📚 Wiki del proyecto cripto (Dan y Diego, personal)
Empieza por el Atlas (dónde está todo), luego el Glosario (qué significa cada cosa), luego las leyes (lo que ya se pagó) y el Handoff (el estado de hoy). Escribe una palabra y la página esconde todo lo que no la contiene.
🗺️ Atlas
Fuente ATLAS.md · 89 KB · modificado 2026-09-17 20:57 UTC
systemctl is-active <unidad>) · 4. Crons de la caja (crontab -l), con su propósito y su log · 5. Pestañas del hub y quién construye cada una · 6. Logs: los vivos y los que engañan · 7. Archivos de estado (lo que el sistema recuerda) · 8. Canon y documentos (qué leer para qué) · 9. Chats de Telegram, por NOMBRE (los ids viven en la bóveda) · 10. Wallets y contratos (todos públicos) · 11. Secretos, SOLO POR NOMBRE (/root/.config/crypto-archive/credentials.env, modo 600) · 12. Las reglas del espejo (Mac contra caja) · 13. Los otros brazos de signal-desk (una línea cada uno) · 14. Data box (178.104.79.3), qué hay en /srv · 15. Primer día: los 10 comandos para ver que todo vive · 16. Cron PROPUESTO para la pestaña Wiki (no instalado; lo instala Dan o su go) · 17. Trampas que hacen tropezar a una sesión nueva (léelas antes de "arreglar" nada) · 18. Verificación (una pasada, 2026-09-17 20:52 UTC)ATLAS. El único mapa del proyecto cripto de Dan y Diego
Escrito 2026-09-17 por el agente Fable (wiki). Proyecto PERSONAL. Nunca infraestructura de Slate. Cada fila trae la ruta, el comando o la URL exacta y una marca de verificación: ✅ = lo vi hoy con ls, systemctl, curl -I o grep. ❌ = no existe o falló, con lo que vi. Los detalles finos están en GLOSARIO.md (términos) y en HANDOFF-2026-09-17.md (estado de hoy). La columna Visto la rellenó un script de una sola pasada el día de la escritura (sección 18).
Orden de lectura para una sesión nueva: este archivo (10 min) -> GLOSARIO.md (5 min) -> el HANDOFF-*.md más reciente -> canonical/logic/THE-RULES.md leyes 34 a 43.
1. Servidores y alias (dónde corre cada cosa)
| Alias / nombre | Dirección | Qué es | Visto |
|---|---|---|---|
ssh signal-desk | root@5.223.82.12 (Hetzner Singapur), llave ~/.ssh/signal_desk | LA CAJA. Corre TODO lo que toca dinero: copiadores, buscadores, hub, crons. Repo /root/signal-desk, python /root/signal-desk/.venv/bin/python. Sin git. | ✅ 4 de 4 |
ssh deaf-box | root@178.104.79.3 (Hetzner Núremberg), misma llave | LA DATA BOX. Muestra y guarda: el hub (Caddy), Baserow db 125, MinIO, espejo del archivo, transcripciones. Hostname slate-desk-data (nombre confuso: NO es de Slate). No puede llegar a la caja (llave de empuje de una sola vía). | ✅ 1 de 1 |
ssh slate | 5.161.87.81 | PROHIBIDO. Servidor del negocio Slate. Un agente lo tocó por error el 09-15. Nunca. | ✅ 1 de 1 |
Mac mini (apple) | ~/bot-archive | El asiento de trabajo. ESPEJO de la caja para código, y la casa del canon y los chats. Git local sin remote. Corre el barrido Robinhood (excepción a la Regla 0). | ✅ 1 de 1 |
Laptop (mac.lan) | su propio ~/bot-archive | Divergido. Ya no corre crons cripto (movidos a la caja el 09-02). | ✅ 1 de 1 |
| Hub (interfaz) | https://portfolio.178.104.79.3.nip.io | Una sola página, pestañas. Login básico (Caddy) con PORTFOLIO_USER / PORTFOLIO_PASSWORD; login de página HUB_SECRET_DAN. Sin credenciales responde 401 = vivo. | ✅ 4 de 4 |
| Baserow | https://baserow.178-104-79-3.sslip.io | db 125 "DEAF Desk" (T.422 conversaciones, T.774 feedback, T.777 trades, T.782/783 calls, T.784 README, tablas 785 a 790 espejo del almacén de wallets). Trampa: puerto 8000, cabecera Host: localhost por túnel. Mapa: canonical/logic/BASEROW-MAP.md. | ✅ 2 de 2 |
Regla 0 (Dan, 2026-09-01): nada vive en un Mac salvo aprobación expresa. La única excepción hoy es el barrido Robinhood (build/wallets/analyzer/), pendiente de mover a la caja.
2. Código, espejo y GitHub (la realidad)
| Qué | Dónde | Estado real 2026-09-17 | Visto |
|---|---|---|---|
| La verdad viva del CÓDIGO | /root/signal-desk en la caja | Sin .git. Los archivos llegan por scp o por deploy.sh de cada carril. | ✅ 2 de 2 |
| El espejo del Mac | ~/bot-archive | Git con 409 archivos sin commit y ningún remote. build/lebot/*.py en el Mac es del 09-06: le faltan rules_tab.py, solana_copy.py, take_profit.py, reconcile.py, routing_audit.py, panel_rules.py, positions_tab.py, venues.py y más. El Mac NO es la verdad del código. | ✅ 9 de 9 |
| El canon (documentos) | ~/bot-archive/canonical/, map/, HANDOFF-*.md | Vive en el Mac. La caja tiene solo el subconjunto de canonical/logic/ del 08-30 (sin THE-RULES.md, sin LEBOT-V2-LOGIC.md, sin INFRA-MAP.md). | ✅ 4 de 4 |
| Export limpio para GitHub | ~/signal-desk-repo | Commit d708717 del 09-02, 0 cambios locales, remote https://github.com/pykleryk/signal-desk.git. | ✅ 2 de 2 |
GitHub pykleryk | https://api.github.com/users/pykleryk | El usuario existe (200 anónimo). El repo signal-desk responde 404 anónimo (no existe o es privado; sin token personal no se distingue). No consta ningún push. | ✅ 1 de 1 |
gh del Mac | gh auth status | Logueado como slatesystemsllc-dot (cuenta de SLATE). No usar para nada de este proyecto. | ✅ 2 de 2 |
| Tokens de GitHub en la bóveda | GITHUB_NEW_TOKEN, GITHUB_THEZBOT1_TOKEN, GITHUB_THEZBOT1_USER | Por nombre. El 09-02 el token empujaba pero no podía CREAR repos (403). | ✅ 3 de 3 |
| Espejo del archivo en la data box | /srv/bot-archive | Copia del archivo (chats WhatsApp: 22,315 archivos, verificado 09-03). | ✅ 1 de 1 |
| Wiki vieja en la data box | /srv/wiki | Git del 08-28 al 08-30 (README, bots, devs, infra, trading, wallets). Parada. No es la wiki de hoy: la de hoy es la pestaña 📚 Wiki del hub. | ✅ 1 de 1 |
Plan propuesto (decisión de Dan, no ejecutado): UN repo privado signal-desk en su cuenta personal (pykleryk o thezbot1), contenido = código + canon + mapas, sin secretos, sin chats. Dirección fija: caja -> repo -> espejo del Mac. La caja pasa a ser un checkout (git pull = despliegue). Plan de corte: ~/signal-desk-repo/CUTOVER-PLAN.md. Comando de push (token de un solo uso, nunca guardado): canonical/logic/INFRA-MAP.md sección "Project 1 status".
3. Servicios systemd en la caja (systemctl is-active <unidad>)
| Unidad | Qué corre | Qué hace | Log | Visto |
|---|---|---|---|---|
signal-lebot.service | build/lebot/lebot.py --paper | Copiador Robinhood Chain (4663). Feed del secuenciador por websocket + sondeo de logs cada 3 s. --paper NO apaga los envíos: el vivo lo deciden targets.json + limits.py. Tope memoria 1 GiB. | /var/log/signal-lebot.log | ✅ 5 de 5 |
signal-lebot-arc.service | build/lebot/lebot.py --chain arc --paper --no-card | Copiador Arc (5042, Circle). Vivo desde 09-16 16:49. Sondeo 6 s. Escáner ciego: decide el ensayo de venta. | /var/log/signal-lebot-arc.log | ✅ 3 de 3 |
signal-lebot-bsc.service | build/lebot/lebot.py --chain bsc --paper --no-card | Copiador BSC (56). | /var/log/signal-lebot-bsc.log | ✅ 3 de 3 |
signal-lebot-base.service | build/lebot/lebot.py --chain base --paper --no-card | Copiador Base (8453). | /var/log/signal-lebot-base.log | ✅ 3 de 3 |
signal-lebot-solana.service | build/lebot/solana_watch.py watch | SOLO MIRA Solana (10 rondas de 6 s y termina; systemd lo relanza: 5,048 reinicios, no es un fallo pero ensucia la métrica). No compra: el que compra es el cron solana_copy.py. | /var/log/signal-lebot-solana.log | ✅ 4 de 4 |
wallet-tape.service | build/wallets/tape/capture.py --serve | La cinta FOMO: lee rhtrenches cada 2 min, acumula toda wallet vista, juzga con bar.py a 30 días. Solo lectura. | /var/log/wallet-tape.log y /var/log/wallet-tape-skips.log | ✅ 5 de 5 |
wallet-solana.service | build/wallets/solana/finder.py --serve | Buscador Solana: runners cada 30 min, cazadores cada hora. Solo lectura. Tarda 30 s en asentarse tras reinicio. | /var/log/wallet-solana.log | ✅ 3 de 3 |
signal-decisions.service | build/alert-card/decisions.py --every 1 | Escucha las pulsaciones de botones de Telegram cada 1 s y las anota. No ejecuta nada. | /root/signal-desk/decisions.log | ✅ 3 de 3 |
signal-hand.service | tools/act_on_press.py --every 2 | LA MANO: convierte una pulsación anotada en compra o venta real por build/exec/swap.py. | /root/signal-desk/act.log | ✅ 4 de 4 |
signal-tgtap.service | tools/tg_tap.py --every 5 --base-usd 10 | Toma de Telegram (canales de callers, solo lectura) y pasada de tarjetas. Sesión Telethon /root/signal-desk/.tg_session.session. | /root/signal-desk/tgtap.log | ✅ 4 de 4 |
signal-watchers.service | tools/start-wallet-watchers.sh | 11 vigilantes de wallets Solana (uno por wallet, solo lectura) -> map/wallet-hunt/watch/feed.jsonl. | /root/signal-desk/map/wallet-hunt/watch/supervisor.log | ✅ 4 de 4 |
discord.service | /usr/bin/discord --remote-debugging-port=9222 | Cliente Discord real (lector de señales) sobre pantalla virtual. Solo lectura del DOM. Tope 2500 MB. | journal | ✅ 1 de 1 |
xvfb.service | Xvfb :99 | Pantalla virtual para Discord. | journal | ✅ 1 de 1 |
Reinicios: de uno en uno, 10 s entre ellos. Un servicio con NRestarts alto se lee con systemctl show <u> -p NRestarts.
4. Crons de la caja (crontab -l), con su propósito y su log
4a. El copiador LEBOT v2 (dinero)
| Cadencia | Script | Qué hace | Log | Visto |
|---|---|---|---|---|
*/5 | build/lebot/take_profit.py --go --chain robinhood | Recorte 2x: vende 30 % cuando la moneda hace 2x desde NUESTRA última compra. Candado .locks/take-profit.lock. | /var/log/take-profit-cron.log | ✅ 3 de 3 |
*/5 | build/lebot/take_profit.py --go --chain arc | Igual en Arc. Candado propio .locks/take-profit-arc.lock. | /var/log/take-profit-arc-cron.log | ✅ 3 de 3 |
*/10 | build/lebot/reconcile.py --go | Cuadre del cuaderno: escribe SELL_OFFBOOK para salidas que LEBOT no hizo (ventas manuales de Dan). | /var/log/lebot-reconcile.log | ✅ 2 de 2 |
*/2 | build/lebot/solana_copy.py --go | Copiador Solana (ML y OD). Candado .solana-copy.lock. 0 a 120 s de retraso. Vende solo a mano. | /root/signal-desk/solana-copy.log | ✅ 3 de 3 |
* * * * * | build/lebot/solana_be_exit.py --mint 5dvX... --target-usd 500 --go | Salida a break-even de EMBER. Ya disparó: marcador build/lebot/be-exit-5dvXTZ5q.done. Sigue corriendo y solo anota la distancia. | /root/signal-desk/solana-be-exit-cron.log | ✅ 3 de 3 |
7 * * * * | build/lebot/routing_audit.py --since 2026-09-10 | Auditoría de rutas: lee el router del RECIBO (tx.to), no del log. | /var/log/routing-audit.log | ✅ 2 de 2 |
*/10 | build/lebot/dep_health.py --quiet | Salud de dependencias externas (1inch, RPCs, GoPlus) antes de que cuesten un trade. | /root/signal-desk/dep-health.log | ✅ 2 de 2 |
4b. Pestañas del hub y el empuje
| Cadencia | Script | Qué hace | Log | Visto |
|---|---|---|---|---|
*/5 | tools/portfolio_html.py --push | Construye la página del hub (map/state/portfolio.html) y la empuja a la data box por ~/.ssh/portfolio_push. | /root/signal-desk/portfolio.log | ✅ 4 de 4 |
*/5 | build/lebot/positions_tab.py --write | Pestaña 🤖 LEBOT: posiciones por wallet copiada, ETH + USD. | /root/signal-desk/positions-tab.log | ✅ 2 de 2 |
*/10 | build/lebot/rules_tab.py --write | Pestaña 🎛️ Reglas: cada límite con su valor, si está bloqueando algo, interruptores. | /root/signal-desk/rules-tab.log | ✅ 2 de 2 |
*/10 | build/wallets/tape/store_tab.py --write | Pestaña 🔎 Carteras: la tabla del almacén que Dan filtra. | /root/signal-desk/wallets-store-tab.log | ✅ 2 de 2 |
*/20 | build/wallets/reader/build_ranked_tab.py | Pestaña 🏆 Ranked: wallets en ETH bajo reglas aprobadas. | ninguno (>/dev/null) | ✅ 1 de 1 |
* * * * * | build/hub/press_pull.py --pull --apply | Trae las pulsaciones firmadas desde la página del hub y las convierte en la misma fila que escribe el tablero de Telegram. | /root/signal-desk/hub-pull.log | ✅ 2 de 2 |
PROPUESTO */30 | build/lebot/wiki_tab.py --write | Pestaña 📚 Wiki (este atlas, el glosario, las leyes, el handoff). Línea exacta en la sección 16. No instalada: su log /root/signal-desk/wiki-tab.log no existe hasta que corra. | (ninguno todavía) | ❌ /root/signal-desk/wiki-tab.log: no existe en la caja (esperado: el cron de la wiki no está instalado) |
4c. El buscador de wallets (solo lectura)
| Cadencia | Script | Qué hace | Log | Visto |
|---|---|---|---|---|
*/2 | tools/store_md5_watch.sh | Vigila el md5 de build/wallets/tape/store.py y anota cada cambio con su hora. | /root/signal-desk/wallets-store-watch.log | ✅ 3 de 3 |
*/10 | build/wallets/tape/record_watch.sh | Los 9 archivos del carril de la cinta contra map/wallets/store/MD5-OF-RECORD.txt. Silencioso si cuadra. | /root/signal-desk/wallets-tape-drift.log | ✅ 3 de 3 |
17 * * * * | build/wallets/monitor/venue_watch.py | Avisa si cambian los contratos de swap, venues o monedas de cotización de una cadena. Resumen map/wallets/venues/SUMMARY.md. | /root/signal-desk/venue-watch.log | ✅ 3 de 3 |
*/5 | build/wallets/alerts/rotate_watch.py | Avisa si una wallet copiada rota su dinero a otra wallet o lo retira. | /root/signal-desk/rotate-watch.log | ✅ 2 de 2 |
4d. Los otros brazos de signal-desk (callers, cerebro, juez, vendedor, X, salud)
| Cadencia | Script | Qué hace | Log | Visto |
|---|---|---|---|---|
*/2 | build/shadow/live_loop.py --once | El bucle vivo: toma -> pipeline -> tarjeta en el teléfono. Necesita claude -p logueado en la caja. | /root/signal-desk/cron.log | ✅ 2 de 2 |
0 3 * * * | build/shadow/daily.py --send | La línea diaria a las 21:00 México en el chat de Dan. | /root/signal-desk/daily.log | ✅ 2 de 2 |
0 13 * * * | tools/morning_brief.py --send | Brief de la mañana (07:00 México): qué tenemos, qué queda en las bolsas, qué decidir. | /root/signal-desk/morning.log | ✅ 2 de 2 |
*/5 | tools/system_pulse.py --send | El pulso: una fila por pieza, viva o muerta, con línea roja si calla más de lo que su cadencia permite. | /root/signal-desk/system-pulse.log | ✅ 2 de 2 |
*/10 | tools/feed_canary.py --send | Canario de feeds: avisa cuando los canales de callers dejan de leerse (TAP_DOWN, TAP_FAILING, LOOP_BLIND...). | /root/signal-desk/feed-canary.log | ✅ 2 de 2 |
0 * * * * | /root/claude-auth-canary.sh | Comprueba que claude -p está logueado en la caja; escribe /root/signal-desk/READER-STATUS.txt. | READER-STATUS.txt | ✅ 3 de 3 |
*/2 | tools/wallet_to_feed.py | El puente: compra on-chain de un caller -> mensaje de feed -> call_to_card. Solo traduce. | /root/signal-desk/map/wallet-hunt/watch/bridge.log | ✅ 2 de 2 |
0 * * * * | tools/caller_last_seen.py --fragment map/caller-last-seen.json | Cuándo vimos por última vez a cada caller, por marca de tiempo del post. | /root/signal-desk/map/caller-last-seen.log | ✅ 2 de 2 |
23 * * * * | build/brain/calls.py y build/brain/post_call_ath.py | El cerebro: libro de calls originales y el ATH posterior a cada call. Salida en map/brain/. | /root/signal-desk/brain.log | ✅ 4 de 4 |
45 3 * * * | build/brain/fomo_measure.py --swaps map/fomo/swaps.ndjson --send | Cada compra FOMO como call: qué hizo la moneda después, contra costes. | /root/signal-desk/fomo-measure.log | ✅ 2 de 2 |
17,47 * * * * | build/judge/opportunities.py --send --max-cards 3 --no-scan-buzz | EL JUEZ (en sombra): tarjetas SCAN y OPPORTUNITY al chat INSIGHTS, sin botones. | /root/signal-desk/judge.log | ✅ 2 de 2 |
*/5 | build/seller/watch.py --wallet <WALLET_ADDRESS> | El vigilante de precios: el máximo que hizo cada posición abierta desde nuestra primera compra. | /root/signal-desk/seller.log | ✅ 1 de 1 |
3,18,33,48 * * * * | build/seller/suggest.py --send --max-cards 2 | El sugeridor de venta: seis reglas sobre las posiciones abiertas, tarjeta cuando una dispara. | /root/signal-desk/seller.log | ✅ 2 de 2 |
*/15 | build/x/x_insights.py --once --callers --send | X: posts nuevos de los callers -> tarjetas INSIGHT / PROPOSAL. Presupuesto 200 posts al día. | /root/signal-desk/x.log | ✅ 2 de 2 |
7 */2 * * * | build/x/x_insights.py --once --buzz | X: ruido por posición abierta. | /root/signal-desk/x.log | ✅ 2 de 2 |
10 0 * * * | build/x/x_report.py --send | Reporte diario de uso y gasto en X, al chat MACHINE. | /root/signal-desk/x.log | ✅ 2 de 2 |
*/15 | tools/wallet_ledger.py --scan --write --json | Todo lo que hay en NUESTRA wallet, cada token que llegó, con precio (pestaña Wallet). Candado .wallet-ledger.lock. | /root/signal-desk/wallet-ledger.log | ✅ 3 de 3 |
*/30 | tools/nose_scan.py --max 6 | NOSE: descubrimiento en DexScreener del ecosistema Robinhood. Solo lectura. | /root/signal-desk/nose.log | ✅ 2 de 2 |
40 2 * * * | tools/winners.py --send | Los ganadores del día y sus cuatro razones simples. | /root/signal-desk/winners.log | ✅ 2 de 2 |
25 */4 * * * | tools/levels_card.py | Las tres definiciones de ATH de Dan y los soportes/resistencias de la tarjeta. | /root/signal-desk/levels.log | ✅ 2 de 2 |
40 12 * * * | tools/yield_track.py --once | Seguimiento en papel del rendimiento delta neutral en dólares. | /root/signal-desk/yield.log | ✅ 2 de 2 |
*/10 | tools/regen_state_pages.py | Regenera las páginas de ESTADO (canonical/logic/WHAT-RUNS-NOW.md y dos más). | /root/signal-desk/regen.log | ✅ 3 de 3 |
*/30 | tools/lane_board.py | LANE-BOARD.md: qué hace cada carril. Sin LLM, cuesta cero. | ninguno | ✅ 2 de 2 |
5 3 * * * | tools/caller_outcomes_v2.py | Resultados por caller, medidos desde NUESTRA toma, no desde su post. | /root/signal-desk/outcomes.log | ✅ 2 de 2 |
Candados: los de dinero viven en /root/signal-desk/.locks/ (take-profit, take-profit-arc, reconcile, routing-audit, solana-be-exit, lebot-rules-tab, lebot-positions-tab, wallets-store-tab). Cuatro viven en la raíz del repo (.solana-copy.lock, .wallet-ledger.lock, .venue-watch.lock, .rotate-watch.lock). Los /tmp/*.lock son restos del 09-10 al 09-15: no los lee nadie.
5. Pestañas del hub y quién construye cada una
Una pestaña se monta escribiendo UN archivo en map/hub-tabs/<slug>.html cuya primera línea es <!-- tab: <etiqueta> --> (tope 1.5 MB, tools/portfolio_html.py función extra_tabs). La página se reconstruye cada 5 min.
| Pestaña | Archivo | Constructor | Cadencia | Visto |
|---|---|---|---|---|
| 🪙 On-chain, 📈 Binance, 📣 Calls, 🧾 History, 🧠 Brain, 📐 Logic, 📡 Sources, 💼 Wallet, 💓 System | integradas | tools/portfolio_html.py | */5 | ✅ 1 de 1 |
| 🤖 LEBOT | map/hub-tabs/lebot.html | build/lebot/positions_tab.py --write (hub_tab.py es solo un envoltorio desde 09-15) | */5 | ✅ 3 de 3 |
| 🎛️ Reglas | map/hub-tabs/lebot-rules.html | build/lebot/rules_tab.py --write | */10 | ✅ 2 de 2 |
| 🔎 Carteras | map/hub-tabs/wallets-store.html | build/wallets/tape/store_tab.py --write | */10 | ✅ 2 de 2 |
| 🏆 Ranked | map/hub-tabs/ranked.html | build/wallets/reader/build_ranked_tab.py | */20 | ✅ 2 de 2 |
| 👛 Wallets | map/hub-tabs/wallets.html | tablero v2 del analizador (Mac), estático desde 09-05 | ninguna | ✅ 1 de 1 |
| 📜 Rules | map/hub-tabs/rules.html | estático del 09-06. VIEJA. Dan decide si se retira. | ninguna | ✅ 1 de 1 |
| 📊 Desk | map/hub-tabs/desk.html | estático del 09-06 | ninguna | ✅ 1 de 1 |
| 🧠 Logic (extra) | map/hub-tabs/logic.html | estático del 09-07; su slug choca con la integrada y se monta como x-logic | ninguna | ✅ 1 de 1 |
| 🔑 API audit | map/hub-tabs/api-audit.html | estático del 09-06 | ninguna | ✅ 1 de 1 |
| 📚 Wiki | map/hub-tabs/wiki.html | build/lebot/wiki_tab.py --write (nuevo hoy) | propuesta */30 | ✅ 2 de 2 |
Interruptores desde la página (Reglas): build/lebot/panel_rules.py escribe build/lebot/panel-overrides.json con auditoría en build/lebot/panel-audit.ndjson. Una anulación escrita DESPUÉS de arrancar un daemon no está en vigor dentro de él: la pestaña lo dice en rojo.
6. Logs: los vivos y los que engañan
| Log | Qué es | Visto |
|---|---|---|
/var/log/signal-lebot.log | Copiador Robinhood. Cómo leer una decisión: BUY 0x… gave nothing; got N TOKEN -> gate 6 … -> plan BUY $X [LIVE] -> líneas ok N / NO N -> result BUY ok=True CONFIRMED o result NOTHING ok=False <puerta>: <razón> -> t_decided: N s after his block -> phases: decided +a s, SENT +b s, landed +c s. 44 MB. | ✅ 1 de 1 |
/var/log/signal-lebot-arc.log, /var/log/signal-lebot-bsc.log, /var/log/signal-lebot-base.log, /var/log/signal-lebot-solana.log | Los otros copiadores y el vigía Solana. | ✅ 4 de 4 |
/var/log/take-profit-cron.log | EL VIVO del recorte 2x Robinhood (cada 5 min). | ✅ 1 de 1 |
/var/log/take-profit-arc-cron.log | Recorte 2x Arc. | ✅ 1 de 1 |
/var/log/lebot-reconcile.log | Cuadre del cuaderno. | ✅ 1 de 1 |
/var/log/routing-audit.log | Auditoría de rutas. | ✅ 1 de 1 |
/var/log/wallet-tape.log, /var/log/wallet-tape-skips.log, /var/log/wallet-solana.log | Los buscadores. | ✅ 3 de 3 |
/root/signal-desk/solana-copy.log | Copiador Solana. 46 MB. | ✅ 1 de 1 |
/root/signal-desk/solana-be-exit-cron.log y /root/signal-desk/solana-be-exit.log | Salida BE de EMBER. | ✅ 2 de 2 |
/root/signal-desk/portfolio.log | El empuje del hub. Última línea "pushed to the data box" = vivo. | ✅ 1 de 1 |
/root/signal-desk/act.log, decisions.log, tgtap.log, hub-pull.log | La mano, las decisiones, la toma de Telegram, las pulsaciones del hub. | ✅ 4 de 4 |
/root/signal-desk/cron.log | El bucle vivo (shadow). 8.6 MB. | ✅ 1 de 1 |
/root/signal-desk/system-pulse.log, dep-health.log, feed-canary.log | Salud. | ✅ 3 de 3 |
/root/signal-desk/brain.log, judge.log, seller.log, x.log, nose.log, winners.log, levels.log, morning.log, daily.log, yield.log, outcomes.log, fomo-measure.log, wallet-ledger.log, venue-watch.log, rotate-watch.log | Un log por brazo, mismo nombre que su cron. | ✅ 15 de 15 |
/root/signal-desk/rules-tab.log, positions-tab.log, wallets-store-tab.log, wallets-store-watch.log, wallets-tape-drift.log | Las pestañas y los vigilantes de md5. | ✅ 5 de 5 |
/root/signal-desk/map/wallet-hunt/watch/supervisor.log y bridge.log | Los 11 vigilantes y el puente. | ✅ 2 de 2 |
/root/signal-desk/build/lebot/lebot.log | Log propio del copiador (además del de systemd). | ✅ 1 de 1 |
/root/signal-desk/build/lebot/take-profit.log | Log propio del recorte (además del cron). | ✅ 1 de 1 |
MUERTO /root/signal-desk/take-profit-cron.log | Parado el 09-10. Lleva una NOTA del 09-16 diciéndolo. Dos agentes lo leyeron como cron muerta. El vivo es /var/log/take-profit-cron.log. | ✅ 2 de 2 |
ENGAÑOSOS /root/signal-desk/regen.log, poswatch.log, poswatch-seed.log | 0 bytes o parados desde el 09-01/09-03. No prueban nada. | ✅ 3 de 3 |
ENGAÑOSOS /tmp/*.lock | Candados viejos del 09-10 al 09-15. Los reales están en /root/signal-desk/.locks/. | ✅ 2 de 2 |
/root/signal-desk/STATUS.md | Del 08-31. VIEJO. El estado de hoy es el HANDOFF y system_pulse. | ✅ 1 de 1 |
7. Archivos de estado (lo que el sistema recuerda)
| Archivo | Qué guarda | Dueño (escritor) | Visto | ||
|---|---|---|---|---|---|
build/lebot/ledger.ndjson | EL CUADERNO: una fila por BUY / SELL / APPROVE / SELL_OFFBOOK, con mode LIVE / PAPER / REPLAY. Leer build/lebot/LEDGER-NOTE.md antes de sumar nada. | decide.py / execute.py / reconcile.py | ✅ 5 de 5 | ||
build/lebot/ledger-paper.ndjson, build/lebot/ledger-replay.ndjson | Cuadernos de papel y de replay. | copiador | ✅ 2 de 2 | ||
build/lebot/positions.json | Posiciones por target | chain | token: cantidades nuestras y suyas, precios de entrada, hashes. | copiador | ✅ 1 de 1 |
build/lebot/targets.json | Las wallets copiadas: etiqueta, cadena, dirección, encendida o no, profile (max_ticket_usd, fixed_ticket_usd). | orquestador / panel_rules.py | ✅ 2 de 2 | ||
build/lebot/ARMED.json | Permiso de vivo: cap_order_usd 500, cap_total_usd 2000 (solo se MUESTRA, nada lo lee), targets_authorised (nadie lo lee; decisión pendiente de Dan), las palabras de Dan con fecha. Su AUSENCIA es papel. | orquestador | ✅ 1 de 1 | ||
build/lebot/deny.json | Tokens que nunca se compran (MONG en Robinhood, Dan 09-10). | Dan por el orquestador | ✅ 1 de 1 | ||
build/exec/limits-state.json | Las bolsas (pots): chain con topes de pérdida = 1,000,000 (Dan 09-16: sin tope), live_enabled: true, max_order_usd 500, 19 correcciones con cita. Un null en un tope rompe check(). | limits.py | ✅ 2 de 2 | ||
build/exec/oneinch-spender-cache.json | El spender de 1inch por cadena, en disco (ley 43: un reinicio no vacía la caché del camino de dinero). | swap.py | ✅ 2 de 2 | ||
build/exec/acted.ndjson, build/exec/sent.ndjson | Lo que la mano actuó y envió. | act_on_press.py | ✅ 3 de 3 | ||
build/lebot/seen.json | Transacciones ya vistas (puerta 2). | copiador | ✅ 1 de 1 | ||
build/lebot/take-profit-state.json | Qué compra recortó cada disparo (la clave nombra la compra). | take_profit.py | ✅ 2 de 2 | ||
build/lebot/exit-alerts-state.json, build/lebot/exit-quotes.json | Estado de las alertas de salida. | exit_alerts.py | ✅ 3 de 3 | ||
build/lebot/solana-copy-state.json | Firmas ya hechas por el copiador Solana (done). | solana_copy.py | ✅ 2 de 2 | ||
build/lebot/sight.json, sight-robinhood.json, sight-arc.json, sight-bsc.json, sight-base.json | Sello de vida por proceso (ley: un sello por proceso). | cada copiador | ✅ 5 de 5 | ||
build/lebot/watch-state.json, watch-state-bsc.json | Último bloque visto por cadena. | copiador | ✅ 2 de 2 | ||
build/lebot/panel-overrides.json, build/lebot/panel-audit.ndjson | Interruptores de la pestaña Reglas y su auditoría. | panel_rules.py | ✅ 3 de 3 | ||
build/lebot/his-spend-cache.json, build/lebot/legs-cache.json | Cachés: lo que ÉL gastó por ráfaga, patas de cada recibo. | copiador | ✅ 2 de 2 | ||
build/lebot/census.ndjson (128 MB) y build/lebot/census-enriched.ndjson (820 MB) | Censo de toda wallet que envía a un venue de swap en Robinhood (fuente 6 de la búsqueda). Se lee por la COLA (ley 34). | copiador | ✅ 2 de 2 | ||
build/lebot/feed.ndjson, build/lebot/cards.ndjson | Eventos del feed y tarjetas emitidas. | copiador | ✅ 2 de 2 | ||
build/lebot/be-exit-5dvXTZ5q.done | Marcador: la salida BE de EMBER ya disparó. | solana_be_exit.py | ✅ 2 de 2 | ||
map/wallets/store/wallets.ndjson (53 MB, solo añadir) y map/wallets/store/wallets-snapshot.json (25 MB) | EL ALMACÉN de wallets: 19,292 filas, 359 juzgadas, 3 pasan la barra. Contrato: build/wallets/tape/SCHEMA.md. Nunca a mano. | build/wallets/tape/store.py | ✅ 4 de 4 | ||
map/wallets/store/STATUS.json | Estado del almacén en una línea (n, juzgadas, pasan, último tick, errores). | capture.py | ✅ 2 de 2 | ||
map/wallets/store/MD5-OF-RECORD.txt | Los md5 de registro de los 9 archivos del carril de la cinta. La caja es la fuente única. | deploy.sh | ✅ 2 de 2 | ||
map/wallets/store/trips/ (149 wallets) | Los libros de viajes por wallet. | store.py | ✅ 2 de 2 | ||
map/wallets/store/capture-state.json, capture.pid, .store.lock, .capture.lock | Estado y candados del capturador (.capture.lock vive mientras corre el servicio; .store.lock se abre en cada escritura del almacén). | capture.py | ✅ 5 de 5 | ||
map/wallets/RANKED-30D.json, RANKED-90D.json, RANKED-90D-P2.json, RANKED-OURS.json, HUNTERS-COPYABLE.json, CALLER-VALUE.json, DEP-HEALTH.json, held.json | Salidas del lector y del ranking (pestaña Ranked, salud). | build/wallets/reader/ | ✅ 2 de 2 | ||
map/wallets/venues/SUMMARY.md | Último resumen del vigilante de venues. | venue_watch.py | ✅ 2 de 2 | ||
build/wallets/solana/state/ | Estado del buscador Solana. | finder.py | ✅ 2 de 2 | ||
map/wallet-hunt/watch/feed.jsonl, cards.jsonl, state-<pubkey>.json | Los 11 vigilantes. | wallet_watch.py | ✅ 2 de 2 | ||
discord-capture/*.ndjson y discord-capture/feed/ | Feeds de Discord y Telegram (contrato: consumir feed/, nunca la raíz). | tomas | ✅ 2 de 2 | ||
map/state/portfolio.html | La página del hub recién construida, antes del empuje. | portfolio_html.py | ✅ 1 de 1 | ||
/root/signal-desk/READER-STATUS.txt | "READER UP" o "READER DOWN" (claude logueado en la caja o no). | canario | ✅ 1 de 1 | ||
/root/signal-desk/LANE-BOARD.md | Qué hace cada carril (sin LLM). | lane_board.py | ✅ 1 de 1 | ||
/root/.config/crypto-archive/credentials.env | LA BÓVEDA (modo 600). Solo nombres en este atlas (sección 11). | Dan / rotaciones | ✅ 1 de 1 | ||
/root/.config/crypto-archive/solana-keypair.json y solana-wallet.pub | La llave Solana y su pública. | Dan | ✅ 2 de 2 | ||
/root/.config/crypto-archive/supervised-session.json | Token de sesión supervisada para el camino vivo. | orquestador | ✅ 1 de 1 | ||
Mac: build/wallets/analyzer/out/search/ | Estado del barrido Robinhood (67 días): pool_state.json, births_covered.ndjson, early_buyers.ndjson; logs, PIDs y build/wallets/analyzer/out/search/logs67/hunters_all.json en logs67/. 545 MB. | search.py / roll_forward.py | ✅ 8 de 8 | ||
Data box: /srv/portfolio/index.html | El hub servido. 1.5 MB. | empuje --push | ✅ 1 de 1 | ||
Data box: /srv/portfolio-presses/presses.ndjson | Pulsaciones firmadas desde la página (hub-press-receiver.service, 127.0.0.1:8790). | press_receiver.py | ✅ 3 de 3 |
8. Canon y documentos (qué leer para qué)
| Documento | Para qué | Visto |
|---|---|---|
ATLAS.md | Este mapa. | ✅ 1 de 1 |
GLOSARIO.md | Cada término del proyecto, dónde vive. | ✅ 1 de 1 |
HANDOFF-2026-09-17.md | El estado de hoy: mandato de Dan, accesos, sistemas vivos, fallos que ya no existen, decisiones pendientes. Idéntico en Mac y caja (md5 1ff864e7…). | ✅ 1 de 1 |
canonical/logic/THE-RULES.md | LAS 43 LEYES. Cada una nombra el incidente que la pagó. Leer 34 a 43 (esta semana). | ✅ 1 de 1 |
canonical/logic/LEBOT-V2-LOGIC.md | El copiador: las cinco capas (detectar, decodificar, decidir, ejecutar, cuadrar), modos, velocidad, plan de pruebas, órdenes de Dan. | ✅ 1 de 1 |
canonical/logic/EXECUTORS.md | Un ejecutor por cadena, el veredicto honeypot por venue, la ruta Robinhood probada en simulación. | ✅ 1 de 1 |
canonical/logic/INFRA-MAP.md | Las cinco máquinas y el plan de GitHub (del 09-02). | ✅ 1 de 1 |
canonical/logic/LANES-AND-INTERFACES.md | Carriles, contratos de datos entre ellos, quién firma cada puerta. | ✅ 1 de 1 |
canonical/logic/LATENCY.md | Pulsación a fill, medido. | ✅ 1 de 1 |
canonical/logic/HAZARDS.md y HAZARDS.md (raíz, 157 KB) | Incidentes que no deben repetirse. | ✅ 2 de 2 |
canonical/logic/CHAIN-FACTS.md, canonical/logic/ROBINHOOD-CHAIN.md, canonical/logic/SOLANA-HAND.md | Hechos por cadena; Robinhood Chain; el copiador Solana del campo. | ✅ 3 de 3 |
canonical/logic/BASEROW-MAP.md, canonical/logic/HUB-ACTIONS.md, canonical/logic/JUDGE.md, canonical/logic/SELLER.md, canonical/logic/X-API.md, canonical/logic/WHAT-RUNS-NOW.md | Baserow; pulsaciones del hub; el juez; el vendedor; X; el estado generado (viejo si su sello pasa de 20 min). | ✅ 6 de 6 |
canonical/logic/00-LOGIC-BASE.md | El plan de registro, cada afirmación etiquetada (DECIDED / EVIDENCED / PROPOSED / OPEN). | ✅ 1 de 1 |
canonical/02-WHAT-DAN-HAS-SET.md | Las palabras de Dan, fechadas. La máxima autoridad: gana a cualquier página. | ✅ 1 de 1 |
canonical/00-EVERYTHING.md y canonical/01-NAMES.md | Los ocho años en una página; la autoridad de nombres (Arm 2 significa cosas distintas en documentos viejos y nuevos). | ✅ 2 de 2 |
canonical/10-LEBOT.md, canonical/LEBOT-MASTERY.md, canonical/21-FOMO.md | LeBot 2023 y lo que sobrevive; el índice de LeBot; FOMO y su API. | ✅ 3 de 3 |
map/INDEX.md, map/PENDING-MASTER.md, LOOKUP.md, START-HERE.md, WORK-QUEUE.md | La historia 2018-2024 y los índices; lo abierto; la tabla generada; cómo trabajar; la cola. | ✅ 5 de 5 |
map/wallets/BOARD.md (sección 2b) | Bitácora del barrido y del método de búsqueda de Dan. | ✅ 1 de 1 |
build/wallets/tape/SCHEMA.md | El contrato del almacén: campos, las tres leyes, read_error, retractación. | ✅ 1 de 1 |
build/wallets/analyzer/ASYMMETRY.md, COVERAGE.md, SYSTEM.md, LOGIC.md | Asimetría (2x a 100x); las 49 métricas del tool viejo de Dan; el sistema buscar-analizar-guardar-seguir. | ✅ 1 de 1 |
build/wallets/solana/SOURCES-MEASURED.md | Límites medidos de pump.fun, DexScreener y Jupiter. | ✅ 1 de 1 |
build/lebot/ARC-2026-09-16.md, TAPE-ROLL-2026-09-15.md, SOLANA-RUNNERS-2026-09-15.md, HUNT-354-2026-09-15.md, TRENCHES-2026-09-14.md, FOMO-WINNERS-2026-09-13.md, AUDIT-DASHBOARD-2026-09-12.md, AUDIT-RANKING-2026-09-13.md, RECHECK-TOP8-2026-09-13.md, UNIPCS-WEEK-2026-09-12.md, FROGE-2026-09-13.md | Informes de esta semana (Arc, cinta, runners Solana, caza de los 354, rhtrenches, ganadores FOMO, auditorías). | ✅ 1 de 1 |
map/wallets/SEARCH-67D-2026-09-15.md | El barrido de 67 días de Robinhood Chain. | ✅ 1 de 1 |
build/lebot/AUTHORIZATION.md, LEDGER-NOTE.md, TEST-CASES.md, LIVE-TEST-RUNBOOK.md, STATUS-CODER.md | La autorización verbatim de Dan; cómo leer el cuaderno; los casos de prueba; el runbook vivo; el estado del coder. | ✅ 1 de 1 |
build/lebot/GATES-2026-09-17.md y build/lebot/gates.json | Registro de TODAS las puertas con la palabra de Dan o "SIN PALABRA DE DAN". No existe todavía (agente gates-fable en marcha). Si al leer esto sigue sin existir: es la primera tarea (HANDOFF sección 10). | ✅ 2 de 2 |
canonical/logic/CARD-DESIGN.md, canonical/logic/PARAMETERS.md | La ley de la tarjeta; cada parámetro con valor, unidad, dueño. | ✅ 2 de 2 |
9. Chats de Telegram, por NOMBRE (los ids viven en la bóveda)
| Nombre en la bóveda | Para qué | Visto |
|---|---|---|
TELEGRAM_CHAT_DAN | El DM de Dan: línea diaria, brief de la mañana, avisos que exigen su decisión. | ✅ 1 de 1 |
TELEGRAM_CHAT_DIEGO | El DM de Diego. | ✅ 1 de 1 |
TELEGRAM_CHAT_DAN_NOTE | Notas para Dan. | ✅ 1 de 1 |
TELEGRAM_CHAT_SIGNALS | El grupo de prueba compartido: tarjetas con botones. Es el chat de RESPALDO de los streams sin chat propio. | ✅ 1 de 1 |
TELEGRAM_CHAT_INSIGHTS | El juez (SCAN, OPPORTUNITY) y los INSIGHT de X. Sin botones. | ✅ 1 de 1 |
TELEGRAM_CHAT_CALLERS | PROPOSAL de X por cashtag nuevo de un caller vetado. | ✅ 1 de 1 |
TELEGRAM_CHAT_MACHINE | Sobre nosotros, no sobre el mercado: reporte de gasto en X, pulso. | ✅ 1 de 1 |
TELEGRAM_CHAT_NOSE | Descubrimiento NOSE. | ✅ 1 de 1 |
TELEGRAM_CHAT_BRANDON | El brazo Binance (Brandon). | ✅ 1 de 1 |
TELEGRAM_CHAT_LEBOT, TELEGRAM_CHAT_REPLAY, TELEGRAM_CHAT_POSITIONS | Nombrados en el código (build/alert-card/streams.py) y AUSENTES de la bóveda. Caen al respaldo: SIGNALS, y si no, DAN. Las tarjetas del copiador Robinhood salen por ese respaldo; los copiadores de bsc/base/arc corren con --no-card. | ❌ TELEGRAM_CHAT_LEBOT: grep -c en la bóveda: 0 (solo la cuenta, nunca el valor); TELEGRAM_CHAT_REPLAY: grep -c en la bóveda: 0 (solo la cuenta, nunca el valor); TELEGRAM_CHAT_POSITIONS: grep -c en la bóveda: 0 (solo la cuenta, nunca el valor) |
TELEGRAM_PRESSERS_ALLOWLIST, HUB_TG_ID_DAN, HUB_TG_ID_DIEGO | Quién puede pulsar un botón. | ✅ 3 de 3 |
Bot: TELEGRAM_BOT_TOKEN / TELEGRAM_BOT_USERNAME; lector: TELEGRAM_API_ID / TELEGRAM_API_HASH | El bot que manda tarjetas; la sesión Telethon que lee canales. | ✅ 4 de 4 |
Tarjetas: build/alert-card/ (renderizado card.py, streams streams.py, decisiones decisions.py, menú menu.py, diseño canonical/logic/CARD-DESIGN.md).
10. Wallets y contratos (todos públicos)
| Qué | Dirección | Visto |
|---|---|---|
| NUESTRA wallet EVM (todas las cadenas EVM) | 0xc6D970893e2458ff5A654d91a62cEaCb8EfDa4d3 (en la bóveda como WALLET_ADDRESS; la llave WALLET_PRIVATE_KEY, la firma build/exec/signer.py) | ✅ 4 de 4 |
| NUESTRA wallet Solana | 3iP5iL1pK81yXxVhqfk82MjPSEQ34Qek6EN6Pef3DEHS (llave solana-keypair.json, pública solana-wallet.pub) | ✅ 3 de 3 |
| MoneyLord (ML), target | 0xfb2f8e284ed4149358775409f163cb917a62f51e en robinhood, bsc, base, arc | ✅ 1 de 1 |
| Overdose (OD), target | 0x37a4e604545c5411d54ef1ee0c2be6729366bf02 en robinhood, bsc, base, arc | ✅ 1 de 1 |
| Unipcs (APAGADO, Dan 09-13) | 0x0a6ebed0155edb4b21d92ad02897a626cd90119e | ✅ 1 de 1 |
| game_for_one (papel, robinhood) | 0xd4ba7169e9f89ffa94cbbb41b0df668eee25737d | ✅ 1 de 1 |
| smol_intern (papel, robinhood) | 0x194db29f5bebacf3201abc9dbdda452116a3ff23 | ✅ 1 de 1 |
| rh-trader-20 / rh-trader-12 (APAGADOS, primeros targets) | 0x758c4b641f59474249253284be27654f13604ef2 / 0x91d2918f5fa833d235f33a2d8caeb953c8c0d554 | ✅ 2 de 2 |
| RelayRouterV3 (la puerta: su pata propia la toca) | 0xb92fe925dc43a0ecde6c8b1a2709c170ec4fff4f | ✅ 1 de 1 |
Receptor Relay (tx.to de las entregas) | 0xccc88a9d1b4ed6b0eaba998850414b24f1c315be | ✅ 1 de 1 |
| Depositario Relay en Solana / programa | 7uTT8Xi5RWXzy7h9XL244GRgEycDYDhLjr3ZyNdXi8pZ / 99vQwtBwYtrqqD9YSXbdum3KBdxPAVxYTaQ3cfnJSrN2 | ✅ 2 de 2 |
| Uniswap v4 PoolManager (Robinhood) | 0x8366a39cc670b4001a1121b8f6a443a643e40951 | ✅ 1 de 1 |
UniversalRouter modificado (Robinhood; minHopPriceX36 extra) | 0x8876789976decbfcbbbe364623c63652db8c0904 | ✅ 1 de 1 |
| Permit2 (canónico) | 0x000000000022D473030F116dDEE9F6B43aC78BA3 | ✅ 1 de 1 |
| USDG (dólar de Robinhood Chain, 6 decimales) | 0x5fc5360d… (completo en build/lebot/chains.py) | ✅ 1 de 1 |
| MONG (DENEGADO, scam, Dan 09-10) | 0x0c975fb035f942e0ba88238c50670daea414805c | ✅ 1 de 1 |
| UpSideDownCat "USDC" (posición abierta, Arc) | 0x8e98a62a995a50eca9979bfa016f91bf36a8f9d9 | ✅ 1 de 1 |
| MUSEBOOK (posición abierta, Robinhood) | 0x91a2dae9699f0b82540b5886b0d8759c22820ba3 | ✅ 1 de 1 |
| EMBER (mint Solana, salida BE hecha) | 5dvXTZ5qwgafnHtwu3Ls3QrWx1U4LQsFeCuJgkk4QEC6 | ✅ 1 de 1 |
| Exploradores | Robinhood https://robinhoodchain.blockscout.com, Base https://basescan.org, BSC https://bscscan.com (sin API gratis), Solana https://solscan.io | ✅ 4 de 4 |
| Puente | Relay https://api.relay.link (sirve Robinhood 4663 y Arc 5042) | ✅ 1 de 1 |
Cadenas (build/lebot/chains.py): Robinhood 4663 (bloque 0.10 s, Arbitrum Orbit, feed del secuenciador wss://feed.mainnet.chain.robinhood.com), Arc 5042 (0.51 s, Circle, gas en USDC nativo), Base 8453 (2 s), BSC 56 (0.45 s), Ethereum 1 (12 s), Solana (0.4 s, sin chain_id).
11. Secretos, SOLO POR NOMBRE (/root/.config/crypto-archive/credentials.env, modo 600)
Nunca se imprime un valor. Se lee así: ssh signal-desk "grep -c ^NOMBRE= /root/.config/crypto-archive/credentials.env" (cuenta, no valor). Un valor en un transcript es una fuga: rotación a Dan.
| Grupo | Nombres |
|---|---|
| Wallet y firma | WALLET_ADDRESS, WALLET_PRIVATE_KEY; Solana en solana-keypair.json |
| Ejecución | ONEINCH_API_KEY |
| Telegram | TELEGRAM_BOT_TOKEN, TELEGRAM_BOT_USERNAME, TELEGRAM_API_ID, TELEGRAM_API_HASH, TELEGRAM_CHAT_* (sección 9), TELEGRAM_PRESSERS_ALLOWLIST |
| Hub | PORTFOLIO_URL, PORTFOLIO_USER, PORTFOLIO_PASSWORD (rotada 09-03 tras una fuga), HUB_SECRET_DAN, HUB_SECRET_DIEGO, HUB_OWNER_NAMES, HUB_TG_ID_DAN, HUB_TG_ID_DIEGO |
| Binance (brazo Brandon) | BINANCE_LIVE_API_KEY, BINANCE_LIVE_API_SECRET, BINANCE_TESTNET_API_KEY, BINANCE_TESTNET_API_SECRET |
| X | X_API_BEARER_TOKEN, X_API_CLIENT_ID, X_API_CLIENT_SECRET, X_API_CONSUMER_KEY, X_API_CONSUMER_SECRET, X_BEARER_TOKEN, X_CONSUMER_KEY, X_CONSUMER_SECRET |
| GitHub | GITHUB_NEW_TOKEN, GITHUB_THEZBOT1_TOKEN, GITHUB_THEZBOT1_USER |
| Data box | BASEROW_BOX_EMAIL, BASEROW_BOX_PASSWORD, CLOUDFLARE_ACCOUNT_ID, CLOUDFLARE_API_TOKEN, CLOUDFLARE_D1_DATABASE_ID |
| Otros | MACWHISPER_LICENSE_KEY; blockscout.key (archivo aparte en la misma carpeta) |
Registro canónico de credenciales (nombre, alcance, quién la dio): map/CREDENTIAL-REGISTER.md. Trampa: curl -w %{url_effective} o -v sobre una URL con credenciales imprime la clave.
12. Las reglas del espejo (Mac contra caja)
- La caja es la fuente única del código. El Mac es un espejo: caja -> Mac, nunca al revés salvo por el despliegue del dueño.
- Un dueño por archivo y por carpeta. Nadie copia la carpeta de otro carril en ninguna dirección (ley 36).
- Se despliega con el
deploy.shdel carril (prueba, escribeMD5-OF-RECORD.txt, reinicia, refleja al Mac, aborta si no coinciden). Parches porscp, nunca heredocs por ssh. - Fixtures grandes solo en el Mac; en la caja las pruebas escriben bajo
/root/signal-desk/.scratch/(ley 37). - Candados en
/root/signal-desk/.locks/, nunca/tmp(tmpfs de 1.9 GB que se llena). - Reinicios de daemons de uno en uno, 10 s entre ellos.
- Hoy el espejo de
build/lebot/está ATRASADO (sección 2). Antes de leer código del copiador:ssh signal-desk cat /root/signal-desk/build/lebot/<archivo>.
13. Los otros brazos de signal-desk (una línea cada uno)
| Brazo | Dónde | Qué es | Visto |
|---|---|---|---|
| Copiador LEBOT v2 | build/lebot/ | Detecta la compra confirmada de un target, decide bajo 11 puertas, ejecuta por build/exec/swap.py, cuadra, tarjeta. Cuatro daemons EVM + cron Solana. | ✅ 2 de 2 |
| Ejecutor | build/exec/ | swap.py (compra y venta por tres rutas: Uniswap > Kyber > 1inch), signer.py (firma, EIP-712 para Permit2), limits.py (bolsas y topes), supervision.py, route_exec en build/lebot/route_exec.py. Las dos únicas puertas al dinero: la mano y el copiador, ambas por aquí. | ✅ 6 de 6 |
| Buscador de wallets | build/wallets/ | tape/ (cinta FOMO + almacén), solana/ (finder), analyzer/ (barrido Robinhood, Mac), reader/ (lector de historial, ranking, pestaña Ranked), bar.py (LA BARRA), monitor/ (venues), alerts/ (rotación). | ✅ 2 de 2 |
| Tarjetas | build/alert-card/ | Renderizado, streams, decisiones (decisions.py), menú, panel-state. Todo lo que llega a un teléfono pasa por aquí. | ✅ 2 de 2 |
| La mano | tools/act_on_press.py | Pulsación anotada -> compra o venta real. Lo único que gasta desde un botón. | ✅ 1 de 1 |
| Hub | tools/portfolio_html.py, build/hub/, build/console/ | La página; press_pull.py trae pulsaciones firmadas desde la página (Path B). | ✅ 4 de 4 |
| Brain (cerebro) | build/brain/ | Libro de calls, ATH posterior, medida FOMO. Salida map/brain/. Doc canonical/logic/BRAIN.md. | ✅ 3 de 3 |
| Judge (juez) | build/judge/opportunities.py | Oportunidades y "double downs" en sombra, tarjetas a INSIGHTS. Doc canonical/logic/JUDGE.md. | ✅ 2 de 2 |
| Seller (vendedor) | build/seller/ | Vigilante de precios + sugeridor de venta (seis reglas). Doc canonical/logic/SELLER.md. | ✅ 2 de 2 |
| Shadow (sombra) | build/shadow/ | El bucle vivo toma -> pipeline -> tarjeta, y la línea diaria. Doc canonical/logic/SHADOW-RUN.md. | ✅ 2 de 2 |
| X | build/x/ | Insights de callers y ruido, medidor de gasto (xapi.py --today). Docs canonical/logic/X-API.md, X-INSIGHTS.md. | ✅ 3 de 3 |
| Discord | discord.service + tools/discord_tap.py + discord-capture/ | Toma de solo lectura del DOM del cliente real (900 USD de membresías: cuidado). Doc canonical/logic/DISCORD-GUARDRAILS.md. | ✅ 4 de 4 |
| tgtap | tools/tg_tap.py | Toma de canales Telegram de callers, solo lectura, pasada de tarjetas. | ✅ 1 de 1 |
| Watchers (vigilantes) | tools/wallet_watch.py + tools/start-wallet-watchers.sh | 11 wallets Solana vigiladas, una por proceso, feed a map/wallet-hunt/watch/. | ✅ 3 de 3 |
| Exit layer (capa de salida) | build/exit-layer/ | Decisiones de salida en modo alerta (nunca vende sola). | ✅ 1 de 1 |
| Brandon / Binance | build/brandon/, build/exec/binance_exec.py | Discord -> futuros Binance (Signal Desk). Construido, no vivo. Doc canonical/logic/BRANDON-COPIER.md. | ✅ 3 de 3 |
| Sources (fuentes) | build/sources/ | token_safety.py (escáner GoPlus + veredicto honeypot), gas.py (gas v5), parsers. | ✅ 3 de 3 |
| Journal (diario) | build/journal/trades.ndjson por tools/trade_journal.py | La fila de diario de cada trade con precio de señal y de fill. | ✅ 2 de 2 |
| txtool, fomo, delta, safety, parsers, solana | build/txtool/, build/fomo/, build/delta/, build/safety/, build/parsers/, build/solana/ | El tool de wallets recuperado; la API FOMO; delta neutral; seguridad; parsers por fuente; sonda de latencia Solana. | ✅ 6 de 6 |
| Salud | tools/system_pulse.py, tools/feed_canary.py, build/lebot/dep_health.py, /root/claude-auth-canary.sh | El pulso, el canario de feeds, las dependencias, el lector. | ✅ 4 de 4 |
| Herramientas | tools/find.py (buscar en todo el archivo), tools/e0_quote.py (cotización por cadena), tools/rh_swap.py (ruta Robinhood en simulación), tools/wallet_ledger.py | ✅ 4 de 4 |
14. Data box (178.104.79.3), qué hay en /srv
| Ruta | Qué es | Visto |
|---|---|---|
/srv/portfolio | El hub servido por Caddy (portfolio.178.104.79.3.nip.io y portfolio.178-104-79-3.sslip.io). | ✅ 1 de 1 |
/srv/portfolio-presses | press_receiver.py + presses.ndjson (hub-press-receiver.service). | ✅ 4 de 4 |
/srv/baserow-data | Baserow (docker baserow), db 125 "DEAF Desk". | ✅ 2 de 2 |
/srv/minio | MinIO (docker minio): binarios de más de 5 MB, nunca git. | ✅ 2 de 2 |
/srv/bot-archive | Espejo del archivo (chats WhatsApp completos). | ✅ 1 de 1 |
/srv/claude-sessions, /srv/claude-feedback, /srv/claude-lane | Transcripciones compartidas de Claude (10), feedback de Dan y Diego (T.774), scripts del lado DEAF. | ✅ 3 de 3 |
/srv/whisper, /usr/local/bin/transcribe, /srv/brandon/transcribe-batch.sh | Transcripción gratis (whisper). Un lote SIEMPRE por transcribe-batch.sh, nunca transcribe en bucle. | ✅ 3 de 3 |
/srv/wiki | Wiki git vieja (08-28). Parada. | ✅ 1 de 1 |
/srv/stage, /srv/rh-bars, /srv/flow | Hojas y media en tránsito; barras de Robinhood; flujo. | ✅ 3 de 3 |
15. Primer día: los 10 comandos para ver que todo vive
1. ssh signal-desk 'systemctl is-active signal-lebot signal-lebot-arc signal-lebot-bsc signal-lebot-base signal-lebot-solana wallet-tape wallet-solana signal-decisions signal-hand signal-tgtap signal-watchers discord'
2. ssh signal-desk 'crontab -l | grep -vc "^#"'
3. ssh signal-desk 'tail -3 /var/log/signal-lebot.log; tail -1 /var/log/take-profit-cron.log; tail -1 /var/log/lebot-reconcile.log'
4. ssh signal-desk 'cat /root/signal-desk/READER-STATUS.txt; df -h /tmp | tail -1; ls /root/signal-desk/.locks'
5. ssh signal-desk 'tail -2 /root/signal-desk/portfolio.log'
6. ssh signal-desk 'cat /root/signal-desk/map/wallets/store/STATUS.json'
7. ssh signal-desk 'tail -2 /root/signal-desk/build/lebot/ledger.ndjson | cut -c1-160'
8. ssh signal-desk 'cd /root/signal-desk && .venv/bin/python build/lebot/take_profit.py --status'
9. ssh signal-desk 'cd /root/signal-desk && .venv/bin/python build/lebot/solana_copy.py --status'
10. curl -sS -I -m 15 https://portfolio.178.104.79.3.nip.io | head -1
Qué esperar: (1) doce active; (2) 44 líneas (41 crons + la línea PATH + 2 vacías); (3) líneas de hoy; (4) READER UP, /tmp bajo 50 %, ocho candados; (5) "pushed to the data box"; (6) n 19,292 o más; (7) filas de hoy o de ayer; (8) y (9) un estado sin traza de error; (10) HTTP/2 401 (vivo, pide login).
16. Cron PROPUESTO para la pestaña Wiki (no instalado; lo instala Dan o su go)
*/30 * * * * cd /root/signal-desk && flock -n /root/signal-desk/.locks/wiki-tab.lock /root/signal-desk/.venv/bin/python build/lebot/wiki_tab.py --write >>/root/signal-desk/wiki-tab.log 2>&1
La pestaña lee ATLAS.md, GLOSARIO.md, canonical/logic/THE-RULES.md y el HANDOFF-*.md más nuevo desde la raíz del repo en la caja. Si un documento cambia en el Mac, hay que copiarlo a la caja (scp) para que la pestaña lo vea.
17. Trampas que hacen tropezar a una sesión nueva (léelas antes de "arreglar" nada)
--paperensignal-lebot.serviceNO significa papel. El vivo lo decidentargets.json+limits-state.json(live_enabled: true)./root/signal-desk/take-profit-cron.logestá muerto desde el 09-10. El vivo:/var/log/take-profit-cron.log./tmp/*.lockno los lee nadie. Los reales:/root/signal-desk/.locks/.- El espejo del Mac va atrasado en
build/lebot/. Lee el código en la caja. build/lebot/GATES-2026-09-17.mdygates.jsonno existen todavía (09-17 20:30 UTC). Comprobar antes de asumir.TELEGRAM_CHAT_LEBOT/_REPLAY/_POSITIONSno están en la bóveda: las tarjetas caen a SIGNALS o DAN.ARMED.jsontargets_authorisedno lo lee ningún código. La lista viva de targets estargets.json(enabled).signal-lebot-solana.servicereinicia cada ~1 min por diseño (10 rondas y sale). 5,048 reinicios no son una avería. El que COMPRA en Solana es el cronsolana_copy.py.- El barrido Robinhood del Mac: sus 4 PIDs en
build/wallets/analyzer/out/search/logs67/están MUERTOS hoy (el HANDOFF lo daba "en curso"). El LaunchAgentio.slate.wallets.rollforwardcorre a las 04:40 (~/Library/LaunchAgents/io.slate.wallets.rollforward.plist->tools/wallets_roll_forward.sh). Antes de relanzar: leerlogs67/y la ley 38. crontab -lrepite 4 veces el comentario de "routing audit". Es ruido, no cuatro crons.STATUS.md(raíz, 08-31),LANE-BOARD.mdycanonical/logic/WHAT-RUNS-NOW.md(sello del 09-02 en el Mac) son viejos. El estado es el HANDOFF +system_pulse.- El hostname de la data box es
slate-desk-data. No es Slate.ssh slatesí lo es y está prohibido. - La solana
wallet-solana.servicetarda 30 s en asentarse; no la reinicies dos veces. census-enriched.ndjsonpesa 820 MB: nunca leerlo entero (ley 34). Uncato unjson.loadtumba el daemon de al lado.- Los ids de Telegram son por canal: dedupe por (feed, id), nunca por id.
- Un
ghlogueado en Slate no prueba nada sobrepykleryk. 404 anónimo = privado o inexistente.
18. Verificación (una pasada, 2026-09-17 20:52 UTC)
Método: cada ruta con ls -d en su máquina (ssh signal-desk, ssh deaf-box, el Mac); cada unidad con systemctl is-active; cada script además contra crontab -l; cada secreto con grep -c (cuenta, nunca valor); cada dirección con grep -il en los json/py del copiador; cada URL con curl -I sin credenciales. La columna Visto de arriba sale de esta tabla. Generada por el script de verificación de la sesión wiki del 2026-09-17.
Totales: 439 ✅ · 4 ❌ · 443 elementos.
| Elemento | Tipo | Dónde | Resultado |
|---|---|---|---|
signal-desk | sshalias | mac | ✅ Host signal-desk en ~/.ssh/config: 1 |
~/.ssh/signal_desk | path | mac | ✅ existe en Mac |
/root/signal-desk | path | box | ✅ existe en caja |
/root/signal-desk/.venv/bin/python | path | box | ✅ existe en caja |
deaf-box | sshalias | mac | ✅ Host deaf-box en ~/.ssh/config: 1 |
slate | sshalias | mac | ✅ Host slate en ~/.ssh/config: 1 (prohibido, no se corre) |
~/bot-archive | path | mac | ✅ existe en Mac |
https://portfolio.178.104.79.3.nip.io | url | net | ✅ curl -I sin credenciales: HTTP 401 |
PORTFOLIO_USER | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
PORTFOLIO_PASSWORD | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
HUB_SECRET_DAN | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
https://baserow.178-104-79-3.sslip.io | url | net | ✅ curl -I sin credenciales: HTTP 302 |
canonical/logic/BASEROW-MAP.md | path | both | ✅ existe en Mac (falta en la caja) |
build/wallets/tape/deploy.sh | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/lebot/*.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/lebot/rules_tab.py | path | box | ✅ existe en caja; en crontab: 1 línea(s) |
build/lebot/solana_copy.py | path | box | ✅ existe en caja; en crontab: 1 línea(s) |
build/lebot/take_profit.py | path | box | ✅ existe en caja; en crontab: 2 línea(s) |
build/lebot/reconcile.py | path | box | ✅ existe en caja; en crontab: 1 línea(s) |
build/lebot/panel_rules.py | path | box | ✅ existe en caja; en crontab: 0 (no es cron) |
build/lebot/positions_tab.py | path | box | ✅ existe en caja; en crontab: 1 línea(s) |
build/lebot/venues.py | path | box | ✅ existe en caja; en crontab: 0 (no es cron) |
~/bot-archive/canonical/ | path | mac | ✅ existe en Mac |
map/ | path | both | ✅ existe en caja + Mac |
HANDOFF-*.md | path | both | ✅ existe en caja + Mac |
canonical/logic/ | path | both | ✅ existe en caja + Mac |
~/signal-desk-repo | path | mac | ✅ existe en Mac |
https://github.com/pykleryk/signal-desk.git | url | net | ✅ curl -I sin credenciales: HTTP 301 |
https://api.github.com/users/pykleryk | url | net | ✅ curl -I sin credenciales: HTTP 200 |
gh | cmd | mac | ✅ cuenta ? |
gh auth status | cmd | mac | ✅ cuenta slatesystemsllc-dot |
GITHUB_NEW_TOKEN | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
GITHUB_THEZBOT1_TOKEN | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
GITHUB_THEZBOT1_USER | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
/srv/bot-archive | path | deaf | ✅ existe en data box |
/srv/wiki | path | deaf | ✅ existe en data box |
signal-lebot.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
build/lebot/lebot.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/lebot/targets.json | path | box | ✅ existe en caja |
build/exec/limits.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/var/log/signal-lebot.log | path | box | ✅ existe en caja |
signal-lebot-arc.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
/var/log/signal-lebot-arc.log | path | box | ✅ existe en caja |
signal-lebot-bsc.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
/var/log/signal-lebot-bsc.log | path | box | ✅ existe en caja |
signal-lebot-base.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
/var/log/signal-lebot-base.log | path | box | ✅ existe en caja |
signal-lebot-solana.service | service | box | ✅ systemctl is-active = active, reinicios 5068 |
build/lebot/solana_watch.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/var/log/signal-lebot-solana.log | path | box | ✅ existe en caja |
wallet-tape.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
build/wallets/tape/capture.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/wallets/bar.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/var/log/wallet-tape.log | path | box | ✅ existe en caja |
/var/log/wallet-tape-skips.log | path | box | ✅ existe en caja |
wallet-solana.service | service | box | ✅ systemctl is-active = active, reinicios 1 |
build/wallets/solana/finder.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/var/log/wallet-solana.log | path | box | ✅ existe en caja |
signal-decisions.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
build/alert-card/decisions.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/root/signal-desk/decisions.log | path | box | ✅ existe en caja |
signal-hand.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
tools/act_on_press.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/exec/swap.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/root/signal-desk/act.log | path | box | ✅ existe en caja |
signal-tgtap.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
tools/tg_tap.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/root/signal-desk/.tg_session.session | path | box | ✅ existe en caja |
/root/signal-desk/tgtap.log | path | box | ✅ existe en caja |
signal-watchers.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
tools/start-wallet-watchers.sh | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
map/wallet-hunt/watch/feed.jsonl | path | both | ✅ existe en caja + Mac |
/root/signal-desk/map/wallet-hunt/watch/supervisor.log | path | box | ✅ existe en caja |
discord.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
xvfb.service | service | box | ✅ systemctl is-active = active, reinicios 0 |
build/lebot/take_profit.py | path | both | ✅ existe en caja + Mac; en crontab: 2 línea(s) |
.locks/take-profit.lock | path | box | ✅ existe en caja |
/var/log/take-profit-cron.log | path | box | ✅ existe en caja |
.locks/take-profit-arc.lock | path | box | ✅ existe en caja |
/var/log/take-profit-arc-cron.log | path | box | ✅ existe en caja |
build/lebot/reconcile.py | path | both | ✅ existe en caja (falta en el Mac); en crontab: 1 línea(s) |
/var/log/lebot-reconcile.log | path | box | ✅ existe en caja |
build/lebot/solana_copy.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
.solana-copy.lock | path | box | ✅ existe en caja |
/root/signal-desk/solana-copy.log | path | box | ✅ existe en caja |
build/lebot/solana_be_exit.py | path | both | ✅ existe en caja (falta en el Mac); en crontab: 1 línea(s) |
build/lebot/be-exit-5dvXTZ5q.done | path | both | ✅ existe en caja (falta en el Mac) |
/root/signal-desk/solana-be-exit-cron.log | path | box | ✅ existe en caja |
build/lebot/routing_audit.py | path | both | ✅ existe en caja (falta en el Mac); en crontab: 1 línea(s) |
/var/log/routing-audit.log | path | box | ✅ existe en caja |
build/lebot/dep_health.py | path | both | ✅ existe en caja (falta en el Mac); en crontab: 1 línea(s) |
/root/signal-desk/dep-health.log | path | box | ✅ existe en caja |
tools/portfolio_html.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
map/state/portfolio.html | path | both | ✅ existe en caja + Mac |
~/.ssh/portfolio_push | path | box | ✅ existe en caja |
/root/signal-desk/portfolio.log | path | box | ✅ existe en caja |
build/lebot/positions_tab.py | path | both | ✅ existe en caja (falta en el Mac); en crontab: 1 línea(s) |
/root/signal-desk/positions-tab.log | path | box | ✅ existe en caja |
build/lebot/rules_tab.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/rules-tab.log | path | box | ✅ existe en caja |
build/wallets/tape/store_tab.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/wallets-store-tab.log | path | box | ✅ existe en caja |
build/wallets/reader/build_ranked_tab.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
build/hub/press_pull.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/hub-pull.log | path | box | ✅ existe en caja |
build/lebot/wiki_tab.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/root/signal-desk/wiki-tab.log | path | box | ❌ no existe en la caja (esperado: el cron de la wiki no está instalado) |
tools/store_md5_watch.sh | path | both | ✅ existe en caja (falta en el Mac); en crontab: 1 línea(s) |
build/wallets/tape/store.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/root/signal-desk/wallets-store-watch.log | path | box | ✅ existe en caja |
build/wallets/tape/record_watch.sh | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
map/wallets/store/MD5-OF-RECORD.txt | path | both | ✅ existe en caja (falta en el Mac) |
/root/signal-desk/wallets-tape-drift.log | path | box | ✅ existe en caja |
build/wallets/monitor/venue_watch.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
map/wallets/venues/SUMMARY.md | path | both | ✅ existe en caja + Mac |
/root/signal-desk/venue-watch.log | path | box | ✅ existe en caja |
build/wallets/alerts/rotate_watch.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/rotate-watch.log | path | box | ✅ existe en caja |
build/shadow/live_loop.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/cron.log | path | box | ✅ existe en caja |
build/shadow/daily.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/daily.log | path | box | ✅ existe en caja |
tools/morning_brief.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/morning.log | path | box | ✅ existe en caja |
tools/system_pulse.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/system-pulse.log | path | box | ✅ existe en caja |
tools/feed_canary.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/feed-canary.log | path | box | ✅ existe en caja |
/root/claude-auth-canary.sh | path | box | ✅ existe en caja; en crontab: 1 línea(s) |
/root/signal-desk/READER-STATUS.txt | path | box | ✅ existe en caja |
READER-STATUS.txt | path | box | ✅ existe en caja |
tools/wallet_to_feed.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/map/wallet-hunt/watch/bridge.log | path | box | ✅ existe en caja |
tools/caller_last_seen.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/map/caller-last-seen.log | path | box | ✅ existe en caja |
build/brain/calls.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
build/brain/post_call_ath.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
map/brain/ | path | both | ✅ existe en caja + Mac |
/root/signal-desk/brain.log | path | box | ✅ existe en caja |
build/brain/fomo_measure.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/fomo-measure.log | path | box | ✅ existe en caja |
build/judge/opportunities.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/judge.log | path | box | ✅ existe en caja |
/root/signal-desk/seller.log | path | box | ✅ existe en caja |
build/seller/suggest.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
build/x/x_insights.py | path | both | ✅ existe en caja + Mac; en crontab: 2 línea(s) |
/root/signal-desk/x.log | path | box | ✅ existe en caja |
build/x/x_report.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
tools/wallet_ledger.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
.wallet-ledger.lock | path | box | ✅ existe en caja |
/root/signal-desk/wallet-ledger.log | path | box | ✅ existe en caja |
tools/nose_scan.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/nose.log | path | box | ✅ existe en caja |
tools/winners.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/winners.log | path | box | ✅ existe en caja |
tools/levels_card.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/levels.log | path | box | ✅ existe en caja |
tools/yield_track.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/yield.log | path | box | ✅ existe en caja |
tools/regen_state_pages.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
canonical/logic/WHAT-RUNS-NOW.md | path | both | ✅ existe en caja + Mac |
/root/signal-desk/regen.log | path | box | ✅ existe en caja |
tools/lane_board.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
LANE-BOARD.md | path | both | ✅ existe en caja + Mac |
tools/caller_outcomes_v2.py | path | both | ✅ existe en caja + Mac; en crontab: 1 línea(s) |
/root/signal-desk/outcomes.log | path | box | ✅ existe en caja |
map/hub-tabs/lebot.html | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/hub_tab.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
map/hub-tabs/lebot-rules.html | path | both | ✅ existe en caja (falta en el Mac) |
map/hub-tabs/wallets-store.html | path | both | ✅ existe en caja (falta en el Mac) |
map/hub-tabs/ranked.html | path | both | ✅ existe en caja (falta en el Mac) |
map/hub-tabs/wallets.html | path | both | ✅ existe en caja (falta en el Mac) |
map/hub-tabs/rules.html | path | both | ✅ existe en caja (falta en el Mac) |
map/hub-tabs/desk.html | path | both | ✅ existe en caja (falta en el Mac) |
map/hub-tabs/logic.html | path | both | ✅ existe en caja (falta en el Mac) |
map/hub-tabs/api-audit.html | path | both | ✅ existe en caja + Mac |
map/hub-tabs/wiki.html | path | both | ✅ existe en caja (falta en el Mac) |
/root/signal-desk/solana-be-exit.log | path | box | ✅ existe en caja |
decisions.log | path | box | ✅ existe en caja |
tgtap.log | path | box | ✅ existe en caja |
hub-pull.log | path | box | ✅ existe en caja |
dep-health.log | path | box | ✅ existe en caja |
feed-canary.log | path | box | ✅ existe en caja |
judge.log | path | box | ✅ existe en caja |
seller.log | path | box | ✅ existe en caja |
x.log | path | box | ✅ existe en caja |
nose.log | path | box | ✅ existe en caja |
winners.log | path | box | ✅ existe en caja |
levels.log | path | box | ✅ existe en caja |
morning.log | path | box | ✅ existe en caja |
daily.log | path | box | ✅ existe en caja |
yield.log | path | box | ✅ existe en caja |
outcomes.log | path | box | ✅ existe en caja |
fomo-measure.log | path | box | ✅ existe en caja |
wallet-ledger.log | path | box | ✅ existe en caja |
venue-watch.log | path | box | ✅ existe en caja |
rotate-watch.log | path | box | ✅ existe en caja |
positions-tab.log | path | box | ✅ existe en caja |
wallets-store-tab.log | path | box | ✅ existe en caja |
wallets-store-watch.log | path | box | ✅ existe en caja |
wallets-tape-drift.log | path | box | ✅ existe en caja |
map/wallet-hunt/watch/bridge.log | path | box | ✅ existe en caja |
/root/signal-desk/build/lebot/lebot.log | path | box | ✅ existe en caja |
/root/signal-desk/build/lebot/take-profit.log | path | box | ✅ existe en caja |
/root/signal-desk/take-profit-cron.log | path | box | ✅ existe en caja |
poswatch.log | path | box | ✅ existe en caja |
poswatch-seed.log | path | box | ✅ existe en caja |
/tmp/*.lock | path | box | ✅ existe en caja |
/root/signal-desk/.locks/ | path | box | ✅ existe en caja |
/root/signal-desk/STATUS.md | path | box | ✅ existe en caja |
build/lebot/ledger.ndjson | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/LEDGER-NOTE.md | path | both | ✅ existe en caja + Mac |
build/lebot/decide.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/lebot/execute.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/lebot/ledger-paper.ndjson | path | both | ✅ existe en caja + Mac |
build/lebot/ledger-replay.ndjson | path | both | ✅ existe en caja + Mac |
build/lebot/positions.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/targets.json | path | both | ✅ existe en caja + Mac |
build/lebot/ARMED.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/deny.json | path | both | ✅ existe en caja (falta en el Mac) |
build/exec/limits-state.json | path | both | ✅ existe en caja + Mac |
build/exec/oneinch-spender-cache.json | path | both | ✅ existe en caja (falta en el Mac) |
build/exec/acted.ndjson | path | both | ✅ existe en caja (falta en el Mac) |
build/exec/sent.ndjson | path | both | ✅ existe en caja + Mac |
build/lebot/seen.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/take-profit-state.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/exit-alerts-state.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/exit-quotes.json | path | both | ✅ existe en caja + Mac |
build/lebot/exit_alerts.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/lebot/solana-copy-state.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/sight.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/sight-robinhood.json | path | box | ✅ existe en caja |
build/lebot/sight-arc.json | path | box | ✅ existe en caja |
build/lebot/sight-bsc.json | path | box | ✅ existe en caja |
build/lebot/sight-base.json | path | box | ✅ existe en caja |
build/lebot/watch-state.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/watch-state-bsc.json | path | box | ✅ existe en caja |
build/lebot/panel-overrides.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/panel-audit.ndjson | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/his-spend-cache.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/legs-cache.json | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/census.ndjson | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/census-enriched.ndjson | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/feed.ndjson | path | both | ✅ existe en caja (falta en el Mac) |
build/lebot/cards.ndjson | path | both | ✅ existe en caja + Mac |
build/lebot/solana_be_exit.py | path | box | ✅ existe en caja; en crontab: 1 línea(s) |
map/wallets/store/wallets.ndjson | path | both | ✅ existe en caja (falta en el Mac) |
map/wallets/store/wallets-snapshot.json | path | both | ✅ existe en caja (falta en el Mac) |
build/wallets/tape/SCHEMA.md | path | both | ✅ existe en caja + Mac |
map/wallets/store/STATUS.json | path | both | ✅ existe en caja (falta en el Mac) |
map/wallets/store/trips/ | path | both | ✅ existe en caja (falta en el Mac) |
map/wallets/store/capture-state.json | path | both | ✅ existe en caja (falta en el Mac) |
map/wallets/store/capture.pid | path | box | ✅ existe en caja |
map/wallets/store/.store.lock | path | box | ✅ existe en caja |
map/wallets/store/.capture.lock | path | box | ✅ existe en caja |
map/wallets/RANKED-30D.json | path | both | ✅ existe en caja (falta en el Mac) |
build/wallets/reader/ | path | both | ✅ existe en caja + Mac |
build/wallets/solana/state/ | path | both | ✅ existe en caja (falta en el Mac) |
tools/wallet_watch.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
discord-capture/*.ndjson | path | both | ✅ existe en caja + Mac |
discord-capture/feed/ | path | both | ✅ existe en caja + Mac |
/root/signal-desk/LANE-BOARD.md | path | box | ✅ existe en caja |
/root/.config/crypto-archive/credentials.env | path | box | ✅ existe en caja |
/root/.config/crypto-archive/solana-keypair.json | path | box | ✅ existe en caja |
/root/.config/crypto-archive/solana-wallet.pub | path | box | ✅ existe en caja |
/root/.config/crypto-archive/supervised-session.json | path | box | ✅ existe en caja |
build/wallets/analyzer/out/search/ | path | both | ✅ existe en Mac (falta en la caja) |
build/wallets/analyzer/out/search/pool_state.json | path | mac | ✅ existe en Mac |
build/wallets/analyzer/out/search/births_covered.ndjson | path | mac | ✅ existe en Mac |
build/wallets/analyzer/out/search/early_buyers.ndjson | path | mac | ✅ existe en Mac |
build/wallets/analyzer/out/search/logs67/hunters_all.json | path | both | ✅ existe en Mac (falta en la caja) |
build/wallets/analyzer/out/search/logs67/ | path | mac | ✅ existe en Mac |
build/wallets/analyzer/search.py | path | mac | ✅ existe en Mac |
build/wallets/analyzer/roll_forward.py | path | mac | ✅ existe en Mac |
/srv/portfolio/index.html | path | deaf | ✅ existe en data box |
/srv/portfolio-presses/presses.ndjson | path | deaf | ✅ existe en data box |
hub-press-receiver.service | service | deaf | ✅ systemctl is-active = active |
/srv/portfolio-presses/press_receiver.py | path | deaf | ✅ existe en data box |
ATLAS.md | path | both | ✅ existe en caja + Mac |
GLOSARIO.md | path | both | ✅ existe en caja + Mac |
HANDOFF-2026-09-17.md | path | both | ✅ existe en caja + Mac |
canonical/logic/THE-RULES.md | path | both | ✅ existe en caja + Mac |
canonical/logic/LEBOT-V2-LOGIC.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/EXECUTORS.md | path | both | ✅ existe en caja + Mac |
canonical/logic/INFRA-MAP.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/LANES-AND-INTERFACES.md | path | both | ✅ existe en caja + Mac |
canonical/logic/LATENCY.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/HAZARDS.md | path | both | ✅ existe en Mac (falta en la caja) |
HAZARDS.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/CHAIN-FACTS.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/ROBINHOOD-CHAIN.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/SOLANA-HAND.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/HUB-ACTIONS.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/JUDGE.md | path | both | ✅ existe en caja + Mac |
canonical/logic/SELLER.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/X-API.md | path | both | ✅ existe en Mac (falta en la caja) |
canonical/logic/00-LOGIC-BASE.md | path | both | ✅ existe en caja + Mac |
canonical/02-WHAT-DAN-HAS-SET.md | path | both | ✅ existe en caja + Mac |
canonical/00-EVERYTHING.md | path | both | ✅ existe en caja + Mac |
canonical/01-NAMES.md | path | both | ✅ existe en caja + Mac |
canonical/10-LEBOT.md | path | both | ✅ existe en caja + Mac |
canonical/LEBOT-MASTERY.md | path | both | ✅ existe en caja + Mac |
canonical/21-FOMO.md | path | both | ✅ existe en Mac (falta en la caja) |
map/INDEX.md | path | both | ✅ existe en caja + Mac |
map/PENDING-MASTER.md | path | both | ✅ existe en Mac (falta en la caja) |
LOOKUP.md | path | both | ✅ existe en caja + Mac |
START-HERE.md | path | both | ✅ existe en caja + Mac |
WORK-QUEUE.md | path | both | ✅ existe en Mac (falta en la caja) |
map/wallets/BOARD.md | path | both | ✅ existe en Mac (falta en la caja) |
build/wallets/analyzer/ASYMMETRY.md | path | both | ✅ existe en Mac (falta en la caja) |
build/wallets/solana/SOURCES-MEASURED.md | path | both | ✅ existe en caja + Mac |
build/lebot/ARC-2026-09-16.md | path | both | ✅ existe en caja + Mac |
map/wallets/SEARCH-67D-2026-09-15.md | path | both | ✅ existe en Mac (falta en la caja) |
build/lebot/AUTHORIZATION.md | path | both | ✅ existe en caja + Mac |
build/lebot/GATES-2026-09-17.md | path | both | ✅ existe en caja + Mac |
build/lebot/gates.json | path | both | ✅ existe en caja + Mac |
canonical/logic/CARD-DESIGN.md | path | both | ✅ existe en caja + Mac |
canonical/logic/PARAMETERS.md | path | both | ✅ existe en caja + Mac |
TELEGRAM_CHAT_DAN | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_DIEGO | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_DAN_NOTE | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_SIGNALS | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_INSIGHTS | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_CALLERS | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_MACHINE | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_NOSE | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_BRANDON | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_LEBOT | secret | box | ❌ grep -c en la bóveda: 0 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_REPLAY | secret | box | ❌ grep -c en la bóveda: 0 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_POSITIONS | secret | box | ❌ grep -c en la bóveda: 0 (solo la cuenta, nunca el valor) |
build/alert-card/streams.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
TELEGRAM_PRESSERS_ALLOWLIST | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
HUB_TG_ID_DAN | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
HUB_TG_ID_DIEGO | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_BOT_TOKEN | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_BOT_USERNAME | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_API_ID | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_API_HASH | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
0xc6D970893e2458ff5A654d91a62cEaCb8EfDa4d3 | addr | box | ✅ en la bóveda como WALLET_ADDRESS; en build/lebot/venues.py |
WALLET_ADDRESS | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
WALLET_PRIVATE_KEY | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
build/exec/signer.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
3iP5iL1pK81yXxVhqfk82MjPSEQ34Qek6EN6Pef3DEHS | soladdr | box | ✅ en solana-wallet.pub |
0xfb2f8e284ed4149358775409f163cb917a62f51e | addr | box | ✅ en build/lebot/targets.json build/lebot/positions.json build/lebot/ledger.ndjson |
0x37a4e604545c5411d54ef1ee0c2be6729366bf02 | addr | box | ✅ en build/lebot/targets.json build/lebot/positions.json build/lebot/ledger.ndjson |
0x0a6ebed0155edb4b21d92ad02897a626cd90119e | addr | box | ✅ en build/lebot/targets.json |
0xd4ba7169e9f89ffa94cbbb41b0df668eee25737d | addr | box | ✅ en build/lebot/targets.json |
0x194db29f5bebacf3201abc9dbdda452116a3ff23 | addr | box | ✅ en build/lebot/targets.json |
0x758c4b641f59474249253284be27654f13604ef2 | addr | box | ✅ en build/lebot/targets.json build/lebot/positions.json build/lebot/ledger.ndjson |
0x91d2918f5fa833d235f33a2d8caeb953c8c0d554 | addr | box | ✅ en build/lebot/targets.json build/lebot/positions.json build/lebot/ledger.ndjson |
0xb92fe925dc43a0ecde6c8b1a2709c170ec4fff4f | addr | box | ✅ en build/lebot/targets.json build/lebot/decode.py |
0xccc88a9d1b4ed6b0eaba998850414b24f1c315be | addr | box | ✅ en build/lebot/targets.json |
7uTT8Xi5RWXzy7h9XL244GRgEycDYDhLjr3ZyNdXi8pZ | soladdr | box | ✅ en build/lebot/_unipcs/relay.json build/lebot/_unipcs/sol_txs.json |
99vQwtBwYtrqqD9YSXbdum3KBdxPAVxYTaQ3cfnJSrN2 | soladdr | box | ✅ en build/lebot/_unipcs/relay.json build/lebot/_unipcs/sol_txs.json |
0x8366a39cc670b4001a1121b8f6a443a643e40951 | addr | box | ✅ en build/lebot/chains.py build/lebot/decode.py build/wallets/reader/chain_map.json |
0x8876789976decbfcbbbe364623c63652db8c0904 | addr | box | ✅ en build/lebot/chains.py build/wallets/reader/chain_map.json |
0x000000000022D473030F116dDEE9F6B43aC78BA3 | addr | box | ✅ en build/lebot/chains.py build/lebot/venues.py |
build/lebot/chains.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
0x0c975fb035f942e0ba88238c50670daea414805c | addr | box | ✅ en build/lebot/deny.json build/lebot/ledger.ndjson |
0x8e98a62a995a50eca9979bfa016f91bf36a8f9d9 | addr | box | ✅ en build/lebot/targets.json build/lebot/positions.json build/lebot/ledger.ndjson |
0x91a2dae9699f0b82540b5886b0d8759c22820ba3 | addr | box | ✅ en build/lebot/positions.json build/lebot/ledger.ndjson |
5dvXTZ5qwgafnHtwu3Ls3QrWx1U4LQsFeCuJgkk4QEC6 | soladdr | box | ✅ en build/lebot/_dash/facts.json build/lebot/legs-cache.json build/lebot/solana-copy-state.json, crontab |
https://robinhoodchain.blockscout.com | url | net | ✅ curl -I sin credenciales: HTTP 200 |
https://basescan.org | url | net | ✅ curl -I sin credenciales: HTTP 403 |
https://bscscan.com | url | net | ✅ curl -I sin credenciales: HTTP 403 |
https://solscan.io | url | net | ✅ curl -I sin credenciales: HTTP 403 |
https://api.relay.link | url | net | ✅ curl -I sin credenciales: HTTP 404 |
ONEINCH_API_KEY | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
TELEGRAM_CHAT_* | secret | box | ✅ grep -c en la bóveda: 9 (solo la cuenta, nunca el valor) |
PORTFOLIO_URL | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
HUB_SECRET_DIEGO | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
HUB_OWNER_NAMES | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
BINANCE_LIVE_API_KEY | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
BINANCE_LIVE_API_SECRET | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
BINANCE_TESTNET_API_KEY | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
BINANCE_TESTNET_API_SECRET | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
X_API_BEARER_TOKEN | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
X_API_CLIENT_ID | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
X_API_CLIENT_SECRET | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
X_API_CONSUMER_KEY | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
X_API_CONSUMER_SECRET | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
X_BEARER_TOKEN | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
X_CONSUMER_KEY | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
X_CONSUMER_SECRET | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
BASEROW_BOX_EMAIL | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
BASEROW_BOX_PASSWORD | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
CLOUDFLARE_ACCOUNT_ID | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
CLOUDFLARE_API_TOKEN | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
CLOUDFLARE_D1_DATABASE_ID | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
MACWHISPER_LICENSE_KEY | secret | box | ✅ grep -c en la bóveda: 1 (solo la cuenta, nunca el valor) |
/root/.config/crypto-archive/blockscout.key | path | box | ✅ existe en caja |
build/lebot/ | path | both | ✅ existe en caja + Mac |
build/exec/ | path | both | ✅ existe en caja + Mac |
build/exec/supervision.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/lebot/route_exec.py | path | both | ✅ existe en caja (falta en el Mac); en crontab: 0 (no es cron) |
build/wallets/ | path | both | ✅ existe en caja + Mac |
build/alert-card/ | path | both | ✅ existe en caja + Mac |
build/hub/ | path | both | ✅ existe en caja + Mac |
build/console/ | path | both | ✅ existe en caja + Mac |
build/brain/ | path | both | ✅ existe en caja + Mac |
canonical/logic/BRAIN.md | path | both | ✅ existe en Mac (falta en la caja) |
build/seller/ | path | both | ✅ existe en caja + Mac |
build/shadow/ | path | both | ✅ existe en caja + Mac |
canonical/logic/SHADOW-RUN.md | path | both | ✅ existe en caja + Mac |
build/x/ | path | both | ✅ existe en caja + Mac |
build/x/xapi.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
tools/discord_tap.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
discord-capture/ | path | both | ✅ existe en caja + Mac |
canonical/logic/DISCORD-GUARDRAILS.md | path | both | ✅ existe en caja + Mac |
map/wallet-hunt/watch/ | path | both | ✅ existe en caja + Mac |
build/exit-layer/ | path | both | ✅ existe en caja + Mac |
build/brandon/ | path | both | ✅ existe en caja + Mac |
build/exec/binance_exec.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
canonical/logic/BRANDON-COPIER.md | path | both | ✅ existe en caja + Mac |
build/sources/ | path | both | ✅ existe en caja + Mac |
build/sources/token_safety.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/sources/gas.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/journal/trades.ndjson | path | both | ✅ existe en caja + Mac |
tools/trade_journal.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
build/txtool/ | path | both | ✅ existe en Mac (falta en la caja) |
build/fomo/ | path | both | ✅ existe en Mac (falta en la caja) |
build/delta/ | path | both | ✅ existe en Mac (falta en la caja) |
build/safety/ | path | both | ✅ existe en caja + Mac |
build/parsers/ | path | both | ✅ existe en caja + Mac |
build/solana/ | path | both | ✅ existe en Mac (falta en la caja) |
tools/find.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
tools/e0_quote.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
tools/rh_swap.py | path | both | ✅ existe en caja + Mac; en crontab: 0 (no es cron) |
/srv/portfolio | path | deaf | ✅ existe en data box |
/srv/portfolio-presses | path | deaf | ✅ existe en data box |
/srv/baserow-data | path | deaf | ✅ existe en data box |
baserow | docker | deaf | ✅ docker ps: baserow Up 2 weeks (healthy) |
/srv/minio | path | deaf | ✅ existe en data box |
minio | docker | deaf | ✅ docker ps: minio Up 2 weeks |
/srv/claude-sessions | path | deaf | ✅ existe en data box |
/srv/claude-feedback | path | deaf | ✅ existe en data box |
/srv/claude-lane | path | deaf | ✅ existe en data box |
/srv/whisper | path | deaf | ✅ existe en data box |
/usr/local/bin/transcribe | path | deaf | ✅ existe en data box |
/srv/brandon/transcribe-batch.sh | path | deaf | ✅ existe en data box |
/srv/stage | path | deaf | ✅ existe en data box |
/srv/rh-bars | path | deaf | ✅ existe en data box |
/srv/flow | path | deaf | ✅ existe en data box |
Los 10 comandos del primer día, corridos una vez
| # | Primera línea de salida |
|---|---|
| 1 | active |
| 2 | 44 |
| 3 | 2026-09-17T20:52:04Z exit layer: USDC E7:ALERT:1.0 not re-sent, the same verdict was carded 122 min ago; it repeats after 6 h |
| 4 | 2026-09-17T20:00:10Z READER UP |
| 5 | 20:50:09 hub: 0 open, 4 closed, 69 history rows (3 hidden), binance live=ok testnet=ok, brain=present -> /root/signal-desk/map/state/portfolio.html |
| 6 | {"at":1789678241.5974178,"dir":"/root/signal-desk/map/wallets/store","n":19292,"with_evm":19104,"with_solana":573,"judged":359,"by_judge_source":{"tap |
| 7 | {"kind": "BUY", "mode": "LIVE", "target": "0xfb2f8e284ed4149358775409f163cb917a62f51e", "chain": "robinhood", "token": "0x91a2dae9699f0b82540b5886b0d8 |
| 8 | 2026-09-17T20:52:23Z take-profit pass: sell 30% at 2.0x, chain robinhood, mode DRY |
| 9 | watching 3 target(s): |
| 10 | HTTP/2 401 |
📖 Glosario
Fuente GLOSARIO.md · 38 KB · modificado 2026-09-17 20:47 UTC
build/exec/) y las salidas · 4. El buscador de wallets (build/wallets/) · 5. La barra (build/wallets/bar.py): una sola para las tres fuentes · 6. El almacén: los campos que una sesión nueva confunde · 7. El hub, las tarjetas y los botones · 8. Los otros brazos (una línea cada uno; detalle en ATLAS sección 13) · 9. Infraestructura y operación · 10. Cadenas, protocolos y sitios · 11. La forma de trabajar (lo que ya se pagó)GLOSARIO. Cada término del proyecto cripto, en una o dos líneas, y dónde vive
Escrito 2026-09-17 por el agente Fable (wiki). Proyecto PERSONAL de Dan y Diego. Nunca Slate. Orden: por bloques (el proyecto, el copiador, el ejecutor, el buscador, la barra, el almacén, el hub, los otros brazos, la infra, las cadenas y protocolos, la forma de trabajar). Dentro de cada bloque, alfabético. Cada entrada dice DÓNDE está la cosa; el mapa completo es ATLAS.md.
1. El proyecto y sus nombres
- Archivo (bot-archive).
~/bot-archiveen el Mac: ocho años de bots recuperados (2018 a 2024), los chats, el canon y el espejo del código. Historia enmap/INDEX.md; una página encanonical/00-EVERYTHING.md. - Arm 1 / Arm 2 / Arm 3. Numeración de brazos que cambió de significado con el tiempo.
canonical/01-NAMES.mdes la autoridad: en documentos viejos "Arm 2" es Signal Desk; para Dan desde 08-28 "el segundo brazo" es el copiador. - Brief. El encargo escrito con el que se lanza un agente o carril: qué hacer, qué no tocar, dónde mirar. Los de esta semana están en
canonical/logic/LANE-BRIEFS.mdy en el HANDOFF. - Caja (la caja, VPS, trading box).
ssh signal-desk= 5.223.82.12. Donde corre todo lo que toca dinero. Repo/root/signal-desk. - Canon. Los documentos que mandan:
canonical/(la espina),canonical/logic/(la lógica y las leyes),canonical/02-WHAT-DAN-HAS-SET.md(las palabras de Dan, que ganan a cualquier página). - Carril (lane). Un agente con un dueño, unos archivos propios y un brief. Un dueño por archivo y por carpeta (ley 36). Tabla en
canonical/logic/LANES-AND-INTERFACES.md. - Data box (DEAF box).
ssh deaf-box= 178.104.79.3. Muestra y guarda: el hub, Baserow, MinIO, el espejo del archivo. Su hostname diceslate-desk-datapero NO es Slate. - DEAF. El lado personal de Diego y Dan en Baserow (db 125 "DEAF Desk", T.422 conversaciones,
/registerdiego). Nunca en la infraestructura de Slate. - Espejo (mirror).
~/bot-archiveen el Mac respecto de la caja: caja -> Mac, nunca al revés salvo por eldeploy.shdel dueño. Hoybuild/lebot/en el Mac va atrasado (ATLAS sección 2). - Handoff. El documento de traspaso entre sesiones: mandato de Dan, accesos, sistemas vivos, fallos que ya no existen, decisiones pendientes. El más nuevo por nombre:
HANDOFF-<fecha>.mden la raíz. - Hub (portafolio, interfaz).
https://portfolio.178.104.79.3.nip.io. Una página con pestañas, construida portools/portfolio_html.pycada 5 min y empujada a la data box. - LeBot (2023). El copiador original de Asmit (2022 a 2024), dos flotas, +7.565 ETH realizados, el gas se comió el 82 %. Historia:
canonical/10-LEBOT.md; qué sobrevive:canonical/LEBOT-MASTERY.md. - LEBOT v2. El copiador de hoy, reconstruido desde 09-04 para Robinhood Chain y luego BSC, Base, Arc y Solana. Código
build/lebot/, lógicacanonical/logic/LEBOT-V2-LOGIC.md. - Ley. Una regla de
canonical/logic/THE-RULES.md. Cada ley nombra el incidente que la pagó; una regla sin incidente es una preferencia y no entra. Hoy 43. Ley 41: una regla que Dan no pidió se borra, no se afina. - Regla 0 (Dan, 09-01). Nada vive en un Mac salvo aprobación expresa: robots, crons, ejecutores, llaves y estado van a un VPS. Excepción actual: el barrido Robinhood.
- Signal Desk. El nombre del proyecto entero desde 2026 (nació como Discord -> Binance con Brandon). Hoy agrupa el copiador, los buscadores, el hub, el cerebro, el juez, el vendedor, X y Discord.
- Slate. El negocio de Dan (
ssh slate). PROHIBIDO para este proyecto. Nada personal en su infra, nada de Slate aquí. - THE-RULES.
canonical/logic/THE-RULES.md: las 43 leyes. Leer 34 a 43 (esta semana). Están plegadas en la pestaña 📚 Wiki del hub.
2. El copiador (LEBOT v2): detectar y decidir
- ARMED (
build/lebot/ARMED.json). El permiso de vivo: su PRESENCIA permite enviar; su ausencia es papel. Llevacap_order_usd(500),cap_total_usd(2000, solo se muestra, nada lo lee) y las palabras de Dan con fecha.targets_authorisedno lo lee ningún código (decisión pendiente). - Compra real (puerta 5). Una compra suya cuenta si hay un log de Swap en el mismo recibo y él pagó algo. Un token que LLEGA sin swap (airdrop, entrega sin pago) no es compra (la ley del airdrop, puerta 1).
- Cuaderno (ledger,
build/lebot/ledger.ndjson). Una fila por BUY / SELL / APPROVE / SELL_OFFBOOK conmodeLIVE, PAPER o REPLAY.spent()filtra por modo;grepno. Leerbuild/lebot/LEDGER-NOTE.mdantes de sumar. - Decisión (
build/lebot/decide.py). La capa que dice qué hacemos con lo que él hizo: once puertas, el tamaño, la ráfaga. Su salida es unplan BUY $X [LIVE]o unNOTHING <puerta>: <razón>. - Decodificar (
build/lebot/decode.py). Convertir su transacción en una etiqueta: BUY, SELL, ROTATE (token por token), TRANSFER_IN, TRANSFER_OUT, OTHER. Se calcula por deltas netos de los Transfer y el valor nativo, sin ABI del router. - deny.json (
build/lebot/deny.json). Tokens que nunca se compran, diga lo que diga la señal. Hoy MONG en Robinhood (Dan 09-10: "es scam"). - Entrega Relay (delivered buy). Una compra que llega a su wallet por el puente Relay: el receptor
0xccc88a9d…es eltx.toy su pata propia toca RelayRouterV30xb92fe925…. Para MoneyLord y Overdose es LA forma normal de comprar. La regla de la puerta:THE-RULES.mdparte 2, hecho 3. - Escalera (ladder,
SIZE_LADDERendecide.py). Dan 09-09: proporcional por peldaños según lo que ÉL puso en la ráfaga de 30 min: hasta 2,000 -> 100; hasta 5,000 -> 300; más -> 500. Solo se usa si el target no tiene ticket fijo. - Feed del secuenciador.
wss://feed.mainnet.chain.robinhood.com: Robinhood Chain publica cada bloque en el momento en que el secuenciador lo ordena, antes de que cualquier RPC lo sirva. Es el detector primario en Robinhood (build/lebot/feed.py); el sondeo de logs cada 3 s es el respaldo. Dio 403 del 09-11 al 09-17 y el sondeo cubrió. - fill_price (
build/lebot/fill_price.py). Lo que de verdad se pagó por un fill, leído del recibo: una pata es SUYA cuando toca su wallet o el pool que emitió el Swap; una ruta es CADENA (mismo dinero en varios puntos: no se suma) o SPLIT (rutas paralelas: se suma). Ley "Una ruta es una CADENA o un SPLIT". - Gate (puerta). Cada comprobación por la que pasa una señal, numerada y con nombre en el log. Ver "Puertas 1 a 11".
- Modo (LIVE / PAPER / REPLAY). LIVE = envío real; PAPER = decide y anota, no firma; REPLAY = transacciones viejas pasadas por detectar y decidir. El flag
--paperde los servicios NO apaga los envíos: el vivo lo decidentargets.json+limits.py. - MoneyLord (ML). Uno de los dos traders copiados. Wallet EVM
0xfb2f8e28…(misma en Robinhood, BSC, Base y Arc), cuenta 7702 delegada, compra por Relay y vende por bundlers 4337. Su FOMO reconciliado con la cadena al dólar:map/wallets/MONEYLORD-RECONCILIATION.md. Mediana -17 % por viaje; 8 de sus últimas 11 pierden. - Overdose (OD). El otro trader copiado. Wallet
0x37a4e604…. También 7702, también Relay. 28 firmantes distintos en 167 fills: nunca una lista de firmantes como puerta. - Posiciones (
build/lebot/positions.json). Portarget|chain|token: cuánto tenemos, cuánto tenía él al copiar, precios de entrada, hashes de ambos lados. - Puertas 1 a 11 (
decide.py). 1 es suya y confirmada; 2 no vista antes; 3 forma accionable; 4 el token es un token (no moneda, no denegado; la 4b "imita el dólar" se borró el 09-16); 5 compra real; 6 seguridad (escáner GoPlus + ensayo de venta al tamaño real; en Robinhood y Arc el escáner está ciego y decide el ensayo); 7 cooldown (0, Dan 09-06); 8 compras seguidas por target (máximo 10 abiertas, Dan 09-10); 8b él aún tiene al menos el 50 % (MIN_HELD_FRACTION); 9 posiciones abiertas de LEBOT; 9b tope por moneda; 10 tope LEBOT; 11 tamaño. Cada rechazo se escribe con su nombre. - Ráfaga (burst). Dan 09-10: "si un trader hace 6 compras rápidas del mismo token y gasta 10,000 en total, entra el trade por 500". La señal es la RÁFAGA (sus compras del mismo token en 30 min sumadas), no la transacción. Una entrada por ráfaga; una compra suya que sube el peldaño completa la posición, no abre otra. El copiador Solana la tiene desde 09-15 (antes disparó cinco compras de EMBER en 48 s).
- seen.json. Las transacciones ya vistas (puerta 2). Una transacción vista dos veces es una acción.
- Sondeo (poll). Preguntar al RPC cada N segundos "qué hicieron mis wallets en los bloques nuevos" (
eth_getLogs). 3 s en Robinhood, BSC y Base; 6 s en Arc. Es lo que cubre cuando el feed falla. - Target. Una wallet copiada, fila de
build/lebot/targets.json: etiqueta, cadena, dirección,enabled,profile(max_ticket_usd,fixed_ticket_usd). Hoy encendidos: moneylord y overdose en robinhood, bsc, base y arc; game_for_one y smol_intern en robinhood (papel); unipcs y rh-trader-12/20 apagados. - t_decided / phases. Los relojes de cada copia en el log:
t_decided: N s after his blockyphases: decided +a s, SENT +b s, landed +c s. Mediana en Robinhood 2.5 s de decisión, +7 a +18 s confirmado (HANDOFF sección 7). - Ticket fijo (
fixed_ticket_usd) y tope por trader (max_ticket_usd). Dan 09-16/17: 100 fijos para ML y OD en Robinhood, BSC y Base; 75 en Arc. Mandan en TODAS las rutas de tamaño (arreglado 09-16: la ruta proporcional lo ignoraba). Viven en el bloqueprofiledel target. - Tope por orden (
CAP_ORDER_USD= 500). El techo de cualquier orden. No hay tope total ni tope de pérdida (Dan 09-14 y 09-16). - Unipcs. Tercer trader, copiado del 09-12 al 09-13 y apagado por Dan ("mejor no lo copiemos por los momentos"). Informe
build/lebot/UNIPCS-WEEK-2026-09-12.md.
3. El ejecutor (build/exec/) y las salidas
- 1inch. Agregador. Fue la única ruta hasta 09-09; se midió que colapsa a nuestro tamaño real (hasta -36 % en ventas). Hoy es la tercera ruta y solo si es mejor. Clave
ONEINCH_API_KEY; presupuesto diario 12,000 llamadas (75 % vigilantes, el resto trade). Ley 43: toda llamada del camino de ejecución es una llamada de trade. - Bolsas (pots,
build/exec/limits-state.json). Dos bolsas:chainybinance. La de chain tienetotal_loss_usdydaily_loss_cap_usden 1,000,000 (= sin tope, Dan 09-16; unnullrompecheck()),live_enabled: true,max_order_usd500. 19 correcciones con cita. - Breaker (interruptor de pérdida). El tope de pérdida de
limits.pyque latch-eabalive_enabled: falseal primer incumplimiento. Nadie lo eligió; paró el copiador 21 h el 09-15/16 sin avisar. Fuera (ley 40). Regla: cualquier interruptor que pueda parar el copiador PAGINA a Dan y sale en rojo en Reglas. - Ensayo de venta (exit rehearsal,
0a rehearsalenswap.py). Antes de comprar: cotizar la venta del 100 % de la posición al tamaño real y rechazar si no hay cotización o si el viaje redondo pierde más del 15 %. Es una medida ("¿puedo vender esto ahora a mi tamaño?"), no una adivinanza de scam. Doctrina de LeBot 2023 (THE-RULES parte 3). - Escáner (GoPlus,
build/sources/token_safety.py). Leeis_honeypot,cannot_sell_all,sell_tax, tenedores de LP. En Robinhood y Arc no sirve campos (ciego): decide el ensayo. Los hallazgos de LP (lp_risk,lp_puller) son HECHOS en la tarjeta, nunca frenos (ley 42). - Espejo de ventas (mirror sells). Cuando él vende una fracción, vendemos la misma fracción de lo nuestro (
plan SELL 18%= él vendió 18 %). Si su saldo queda en polvo, vendemos el 100 %. Dan 09-05: "quiero copiar una wallet entera". - Gas (gas v5,
build/sources/gas.py). La mitad estructural del diseño de Dan que sobrevive en cadenas con secuenciador: puja relativa a la red con tope, asimetría compra/venta, modo gas máximo. Dan 09-17: "la rapidez no viene por gas, viene por detección y ejecución". - Kyber (KyberSwap). Segunda ruta (uniswap > kyberswap > 1inch). Permiso ERC20 plano, probado.
build/lebot/kyber_venue.py. - Límites (
build/exec/limits.py). El juez de cada orden después de nuestras puertas: bolsas, tope por orden,live_enabled. Paso "0 spend cap" deswap.run. - La mano (hand,
tools/act_on_press.py). Convierte una pulsación anotada (botón de Telegram o del hub) en una compra o venta real. Lo único que gasta desde un botón. Mantiene todas las puertas. - Pasos de
swap.run(compra). 0 tope de gasto, 0a ensayo, 0b wallet, 1 saldo / seguridad, 2 gas y viaje redondo, 2b tope con gas, 3 router (caché en discooneinch-spender-cache.json), 4 cotización de compra en tres rutas, 6 saldo, 7 swap construido, 8 firmado, 9 broadcast, 10 pre-approve al router de venta. Venta: 0 precio nativo, 0b wallet, 1 saldo, 1b cantidad, 2 cotización de venta, 3 router, 4 allowance (4a approve firmado, 4b enviado), construir, firmar, enviar. - Pre-approve (paso 10). Después de comprar, aprobar el monto exacto al router de venta en su propia transacción barata, para que la venta sea un solo envío. Nunca allowance infinito. Mismo router para aprobar y vender (Dan).
- Reconcile (
build/lebot/reconcile.py, cada 10 min). Escribe en el cuaderno las salidas que LEBOT no hizo (ventas manuales de Dan, ventas fallidas cerradas a mano) comoSELL_OFFBOOK, leyendo saldos en cadena. Dan 09-10: "mis ventas manuales deben escribir en el cuaderno". - Recorte 2x (take profit,
build/lebot/take_profit.py, cada 5 min). Vende el 30 % cuando el token hace 2x desde NUESTRA última compra (la base es la compra más reciente, nunca un promedio; el coste es lo que salió de NUESTRA wallet). Estadotake-profit-state.jsoncon la clave de la compra recortada. Log vivo/var/log/take-profit-cron.log. - Router / venue. El contrato al que llamamos para cambiar (
tx.to). Cada cadena nombra su router enbuild/wallets/reader/chain_map.jsoncon la medida que lo eligió a NUESTRO tamaño.build/lebot/routing_audit.py(cada hora) lo lee del recibo: "audita la cadena, no el log". - Ruta (route_exec,
build/lebot/route_exec.py,venues.py). Cotizar tres venues, ordenar, y bajar por el ranking hasta el primero que gana a la base Y puede construir Y simula sin revertir. Leer solo la cabeza hizo decoración del resto (FROGE, 09-12). - SELL_OFFBOOK. La clase de fila que escribe
reconcile: una salida que no hicimos nosotros. Distinta de SELL para que ningún ingreso inventado llegue al P&L. - Signer (
build/exec/signer.py). Firma transacciones y, desde 09-12, mensajes tipados EIP-712 (Permit2). No sabe qué firma a propósito:venues.uni_buildcomprueba spender, token y chainId contra la cotización antes. - Uniswap (v4, Universal Router). Primera ruta. En Robinhood el router es un fork modificado (
0x8876789…, struct conminHopPriceX36extra); el calldata se sacó de una transacción real de la cadena, no de la documentación. Vende por Permit2.build/lebot/uni_pool_quote.py.
4. El buscador de wallets (build/wallets/)
- Almacén (store, la tienda).
map/wallets/store/wallets.ndjson(solo añadir) +wallets-snapshot.json(una fila por clave). Escrito solo porbuild/wallets/tape/store.py. Contratobuild/wallets/tape/SCHEMA.md. 19,292 filas, 359 juzgadas, 3 pasan (09-17). - Analizador (analyzer,
build/wallets/analyzer/, corre en el Mac). El barrido en cadena de Robinhood: pools nacidos, runners, compradores antes del primer 2x, cazadores.search.py(barrido),roll_forward.py(diario, LaunchAgent 04:40), estado enout/search/. DocsASYMMETRY.md,COVERAGE.md,SYSTEM.md. - Asimetría (ASYMMETRY). La vara desde 09-05: cuántas compras de una wallet llegan a 2x, 5x, 10x, 50x, 100x; market cap de entrada; minutos desde el nacimiento; si aguanta runners. Dan: "hay gente pegando 2x 5x 10x 50x 100x 1000x, perseguimos las copias que nos lleven ahí".
- Barrido (sweep). Leer todos los logs de swap de una cadena en una ventana (67 días en Robinhood) para encontrar runners y quién los compró. Bitácora
map/wallets/BOARD.mdsección 2b; informemap/wallets/SEARCH-67D-2026-09-15.md. Hoy sus 4 PIDs del Mac están muertos (ATLAS sección 17). - Cazador (hunter). Una wallet que compró 3 o más runners distintos antes de que doblaran. La clase (bot, sniper, humano) es una columna, nunca un filtro.
- Censo (
build/lebot/census.ndjson,census.py). Toda wallet que envía a un venue de swap en Robinhood, sacada del feed del secuenciador (fuente 6, el método de Dan: "olfateaba los contratos de swap"). Enriquecido cada 5 min leyendo solo la COLA del archivo (ley 34). - Cinta FOMO (tape,
build/wallets/tape/capture.py,wallet-tape.service). rhtrenches.com publica los trades de FOMO sin token:/api/tapey/api/closed(400 filas, se acumulan),/api/trader/<handle>(historial cerrado, 1 llamada cada 9 s obedeciendo Retry-After). El 87 % de los dólares de la cinta son compras falsas marcadas por el sitio: se excluyen. Se mina, no se opera desde ella. - Compradores tempranos (early buyers). Quién compró un runner antes de su primer 2x. Fichero
out/search/early_buyers.ndjsonen el Mac. - Coste no visto (COSTE NO VISTO,
sold_over_bought). Un viaje que vendió más de lo que se vio comprar en la ventana (bolsa de antes). No es un mal viaje: es ilegible. Su exceso no es un viaje; la fila queda SIN veredicto. - Custodial vs propia (
own_wallet: OWN / CUSTODIAL / UNREAD). FOMO publica una dirección de perfil que a menudo NO firma nada (custodial: 0 de 27 fills firmados). La wallet real se recupera por método (hunt-354). Solo la propia es copiable a velocidad de bloque. - Finder Solana (
build/wallets/solana/finder.py,wallet-solana.service). pump.fun (/v2/coins,/v1/...), DexScreener y Jupiter: runners 3x, compradores tempranos, cazadores. Límites medidos enbuild/wallets/solana/SOURCES-MEASURED.md. - FOMO. La app/leaderboard donde MoneyLord y Overdose publican. Página
canonical/21-FOMO.md: endpoints, trampas, por qué FOMO no firma desde tu wallet. Surealized_pnldel tablero está mal; solo valehistory[]. - Gema. Dan 09-15: "perfecciona el sistema de búsqueda de wallets para conseguir unas gemas". Una wallet que pasa la barra y merece ojos; copiar necesita su palabra.
- hunt-354. El método (09-15) que recuperó la wallet real de un handle de FOMO cruzando lo que tiene a precisión completa con la cadena; validado contra tres wallets conocidas. Informe
build/lebot/HUNT-354-2026-09-15.md. - Lector (reader,
build/wallets/reader/wallet_history.py,solana_history.py). Lee el historial de una wallet en cadena: precio desde el lado de la wallet, recibos por lotes de 60. Rankingrank_batch.py; pestañabuild_ranked_tab.py; conversión WETHeth_rate.py. - Método de Dan (2022). Su tool de análisis de wallets (49 métricas,
canonical/80-THE-WALLET-TOOL.md, cobertura enbuild/wallets/analyzer/COVERAGE.md): "preferible un wallet que pegue palazos poniendo poco". La puerta de apuesta (bet gate) y la de antigüedad (hold gate) vienen de ahí. - Nunca rehacer (registro). Dan 09-15: "no rehagas nada". Cada wallet tiene fila desde que se VE; se relee solo con actividad más nueva que su
judged_at(statusseen / judged / new_activity). El salto se anota. - proven_share. Qué parte del libro pudo ponerse precio (0 a 1). Bajo
MIN_PROVEN_SHARE = 0.90no hay veredicto: LIBRO A MEDIAS / PARTLY PRICED. - Retractación (
retract(ident, by, why)). Un veredicto publicado que se retira: un hecho y una escritura, solo por el dueño del veredicto, y debe vaciarjudged_at(si no, la regla de nunca-rehacer esconde el hueco para siempre). - rhtrenches.
rhtrenches.com, la cinta pública de FOMO. SutsES el timestamp del bloque; la fila llega 0.3 s DESPUÉS de la cadena. Informebuild/lebot/TRENCHES-2026-09-14.md. - Runner. Un token que hizo al menos 3x desde su base de la hora de nacimiento (base = VWAP de compras de la primera hora sin el primer swap), con pico de tamaño ≥ $20 y precios EJECUTADOS (un pool drenado imprime
MAX_SQRT_PRICEy no es un trade).n_trusted/count_provisionalmarcan lo que falta por reconstruir. - Sniper / snipe_flag.
deployer,same_slot,first10onone: si entra antes que todos. Confirst_buy_delay_syreplay_delay_pct_15s(precio 15 s después de su fill contra su fill = lo que nos cuesta llegar tarde). - Viaje (trip, round trip). Una compra y su venta emparejadas por token. Libro de viajes por wallet en
map/wallets/store/trips/(trips.py); las ventanas y números de una fila salen derollup.py.
5. La barra (build/wallets/bar.py): una sola para las tres fuentes
"Pasa la barra" = merece ojos, NO "copiar". Copiar requiere la palabra de Dan. Cada cláusula lleva su número y su origen en el código.
- VIVA. Un viaje cerrado o abierto en las últimas 48 h.
- VIAJES. Al menos 10 cerrados (
MIN_TRIPS = 10). - MEDIANA. Mediana del retorno por viaje > 0 (la media engaña).
- PEAJE (toll,
TOLL_PCT) y RETRASO (DELAY_PCT). Eran 1 % por lado y 2 % de entrada tardía; Dan 09-16 los puso en 0. Las cláusulas quedan inertes. (El peaje medido en Robinhood era 1.010 % por lado; ley "Dos por ciento por viaje redondo es el suelo".) - SIN EL MEJOR. El libro sigue positivo sin su mejor viaje. Toda "GANADORA" que descansa en un viaje no pasa.
- BOLSAS (bags). Cada bolsa abierta cuenta como perdida A SU COSTE EN DÓLARES (ley 39). Si ≥ 90 % de los viajes tienen coste, todo el libro es en dólares; si no, método plano y lo dice.
- POLVO (dust,
DUST_USD = 1). Un viaje comprado por menos de 1 USD no cuenta. Mayoría de polvo = falla con razón DUST (ley 35, la wallet 7eYa: 10 de 11). - SIN VEREDICTO (readable = False). La fila vuelve a la cola sin fecha: LIBRO A MEDIAS (< 90 % de viajes con precio), COSTE NO VISTO, BOLSAS SIN COSTE LEÍDO.
read_errorsignifica "no hay veredicto"; una fila dice una sola cosa. - Hold gate / bet gate. Del método de Dan: dinero durante un tiempo respetable (mediana del tablero: 36 días; 0 wallets con más de un año) y apuestas de nuestro tamaño (8 de 49 con un cuarto de compras ≤ $100). Medidos en
map/wallets/DOOR-SCORED.md.
6. El almacén: los campos que una sesión nueva confunde
- Clave (
key).evm:<0x minúsculas>>sol:<base58, mayúsculas respetadas>>handle:<minúsculas>.aliasesguarda toda identidad vista; dos filas que resultan ser una se fusionan (merged_from). - null. NO LEÍDO. Nunca cero. Un cero es una medida.
- proofs. Cada número trae su prueba (
tx,url,at). Sin prueba es una afirmación y la tabla lo imprime como tal. - judged_at / judge_source / history_hash. Cuándo se juzgó, quién (
tape,hunt-354,analyzer,sweep,solana), y el hash de los viajes (mismo hash = nada que rehacer). - coverage_note vs read_error. Cobertura parcial en un veredicto que SÍ vale ("1 de 581 no se pudo leer") va en
coverage_note.read_error= no hay veredicto. Un patch con ambos se rechaza. - Campo desconocido. Se descarta con aviso (
UNKNOWN-FIELDS.json); el validador tiene una lista blanca y un campo que no conoce no aplica en silencio. - Semilla (seed,
seed_registry.py). Las filas sembradas desde los informes viejos (759). Un pase conjudged_atmás viejo nunca retrocede la fila. - MD5-OF-RECORD (
map/wallets/store/MD5-OF-RECORD.txt). Los md5 de los 9 archivos del carril de la cinta, escritos PORdeploy.shdentro del despliegue.record_watch.sh(cada 10 min) compara;tools/store_md5_watch.sh(cada 2 min) vigilastore.py. - deploy.sh (
build/wallets/tape/deploy.sh). Se corre DESDE el Mac; la caja es la fuente única: prueba, escribe el md5 de registro, reinicia el servicio, refleja al Mac, aborta si no coinciden. Un solo mandato porque el md5 como paso aparte se olvidaba. - STATUS.json. El estado del almacén en una línea:
n,judged,bar_passes,alive_48h,last_tick_at, errores http (429).
7. El hub, las tarjetas y los botones
- Pestaña (tab). Un archivo en
map/hub-tabs/<slug>.htmlcon primera línea<!-- tab: <etiqueta> -->, tope 1.5 MB. Hoy: 🤖 LEBOT, 🎛️ Reglas, 🔎 Carteras, 🏆 Ranked, 👛 Wallets, 📜 Rules (vieja), 📊 Desk, 🧠 Logic, 🔑 API audit, 📚 Wiki. Constructores en ATLAS sección 5. - Reglas (🎛️).
build/lebot/rules_tab.py: cada límite que puede rechazar o redimensionar un trade, con su valor vivo, si ESTÁ BLOQUEANDO algo ahora, y su archivo y línea. Interruptores por trader ($100 / $300 / $500 / quitar) y on/off. Dan 09-10: "sigues teniendo reglas de mierda que no tengo visibilidad". - Interruptores (panel,
panel_rules.py). Escribenpanel-overrides.jsoncon una fila de auditoría enpanel-audit.ndjson. Una anulación después de arrancar un daemon no está en vigor dentro de él; la pestaña lo dice ("PENDIENTE DE REINICIO"). - Tarjeta (card). Lo que llega al teléfono por Telegram: qué hizo él (enlace a su tx), qué hicimos (enlace a la nuestra), tamaños, retraso, botones SELL 50 / SELL 100. Ley de la tarjeta:
canonical/logic/CARD-DESIGN.md; códigobuild/alert-card/. PAPER y REPLAY van en la cara de la tarjeta. - Pulsación (press). Un botón tocado.
decisions.pyla anota (1 s), la mano la ejecuta (2 s). Solo los deTELEGRAM_PRESSERS_ALLOWLIST. Path A: el botón del hub abre el bot de Telegram; Path B: pulsación firmada desde la página,build/hub/press_pull.pyla trae cada minuto. - Streams (chats). Cada tarjeta va a un chat por NOMBRE (
TELEGRAM_CHAT_*, ATLAS sección 9). Los que faltan en la bóveda caen a SIGNALS y luego a DAN. - Custom $. Un monto tecleado por Dan o Diego es su decisión: pasa el tope por orden y el de posiciones abiertas; las bolsas mandan.
8. Los otros brazos (una línea cada uno; detalle en ATLAS sección 13)
- Brain (cerebro,
build/brain/). Libro de calls por caller y el ATH posterior a cada call. Doccanonical/logic/BRAIN.md. - Brandon. El caller de Discord cuyas señales van a futuros de Binance (Signal Desk original). Construido, no vivo.
canonical/logic/BRANDON-COPIER.md. - Callers. Las personas cuyas señales leemos: Brandon, MoneyLord, Overdose, Archie, Unfazed. Feeds en
discord-capture/feed/. - Discord (
discord.service,tools/discord_tap.py). Cliente real en pantalla virtual, solo lectura del DOM. 900 USD de membresías en esa cuenta: cuidado.canonical/logic/DISCORD-GUARDRAILS.md. - Double down. Volver a comprar una posición que ya tenemos. Dan 09-02: "el concepto de doubling down es CLAVE". El juez lo mide.
- Exit layer (
build/exit-layer/). Decisiones de salida en modo alerta: nunca vende sola; toda SELL llega como ALERT con fracción y razón. - Judge (juez,
build/judge/opportunities.py). Oportunidades y double downs en SOMBRA, tarjetas al chat INSIGHTS cada 30 min, sin botones.canonical/logic/JUDGE.md. - NOSE (
tools/nose_scan.py). Descubrimiento en cadena primero (DexScreener del ecosistema Robinhood), X después. Solo lectura. - Pulso (
tools/system_pulse.py). Cada 5 min, una fila por pieza: viva, última señal de vida, línea roja si calla más de lo que su cadencia permite. Dan 09-05: "no puede pasar que una pieza se apague X horas y no sabemos". - Canario (
tools/feed_canary.py,/root/claude-auth-canary.sh). El de feeds distingue TAP_DOWN, TAP_FAILING, LOOP_BLIND; el de auth comprueba queclaude -pestá logueado en la caja (READER-STATUS.txt). - Seller (vendedor,
build/seller/).watch.pyguarda el máximo desde nuestra primera compra;suggest.pyaplica seis reglas y manda una tarjeta cuando una dispara.canonical/logic/SELLER.md. - Shadow (sombra,
build/shadow/). El bucle vivo (toma -> pipeline -> tarjeta) y la línea diaria. Necesitaclaude -pen la caja. - tgtap (
tools/tg_tap.py). La toma de Telegram: canales de callers cada 5 s, solo lectura, pasada de tarjetas. Sesión Telethon en la caja. - Watchers (vigilantes,
signal-watchers.service). 11 wallets Solana vigiladas, un proceso por wallet, feed enmap/wallet-hunt/watch/. El puentetools/wallet_to_feed.pyconvierte una compra en un mensaje de feed. - X (
build/x/). Insights de callers cada 15 min, ruido por posición cada 2 h, medidor de gasto (xapi.py --today, 200 posts al día). - Yield (delta neutral,
tools/yield_track.py). Seguimiento en papel de una estrategia conservadora en dólares.canonical/logic/DELTA-NEUTRAL.md.
9. Infraestructura y operación
- Bóveda (credentials.env).
/root/.config/crypto-archive/credentials.env, modo 600. Todo secreto por NOMBRE; nunca un valor en chat, memoria ni log. Registro:map/CREDENTIAL-REGISTER.md. - Candado (lock, flock). Un archivo que impide que dos copias de un cron corran a la vez. Los de dinero en
/root/signal-desk/.locks/; nunca/tmp(tmpfs de 1.9 GB que se llenó el 09-16, ley 37). - Cron.
crontab -len la caja: 41 líneas activas. Cada una con su propósito y log en ATLAS sección 4. - Empuje (push) del hub.
tools/portfolio_html.py --pushconstruyemap/state/portfolio.htmly lo copia a/srv/portfolio/index.htmlde la data box con la llave~/.ssh/portfolio_push(una sola vía). - GitHub. Hoy: export limpio en
~/signal-desk-repo(commitd708717, 09-02) apuntando apykleryk/signal-desk, que responde 404 anónimo; elghdel Mac está en la cuenta de Slate. Decisión pendiente de Dan: un repo privado personal. ATLAS sección 2. - Ley 36 (un dueño, ningún copiado cruzado). Nadie copia la carpeta de otro carril en ninguna dirección; se importa desde la ruta viva.
- MinIO. Almacén de binarios en la data box (
/srv/minio): todo lo inmutable o mayor de 5 MB va ahí, nunca a git. - NRestarts / MemoryMax. Contador de reinicios y tope de memoria de una unidad systemd.
signal-lebottiene 1 GiB; se murió dos veces el 09-15 leyendo 714 MB del censo (ley 34). - Regla del
--paper. Ver "Modo". El flag no apaga el dinero. - Sesión supervisada (
supervised-session.json). Token que el orquestador entrega para el camino vivo; sin él, papel. - systemd (unidad,
.service). Cómo corren los procesos permanentes en la caja.systemctl is-active <u>,systemctl show <u> -p NRestarts,systemctl cat <u>para ver elExecStart. - Telethon. La librería con la que
tg_tap.pylee canales de Telegram como usuario (sesión.tg_session.session). Distinta del bot que manda tarjetas. - Transcribe (whisper). En la data box:
transcribe <audio>para uno; un lote SIEMPRE por/srv/brandon/transcribe-batch.sh(nice 19, dos hilos, espera memoria). Un lote a pelo tumbó Baserow 15 min. - /tmp vs .scratch. Pruebas en la caja escriben bajo
/root/signal-desk/.scratch/con tope; fixtures grandes solo en el Mac.
10. Cadenas, protocolos y sitios
- Arc (5042). La cadena de Circle. Gas en USDC nativo (la moneda de la cadena es USDC, no ETH). Copiador vivo desde 09-16; escáner ciego; ticket 75; quedan ~41 USDC. Informe
build/lebot/ARC-2026-09-16.md. Relay la sirve (0.1 ETH llegó como 237 USDC en segundos). - Base (8453). L2 de Coinbase, bloque 2 s. Copiador
signal-lebot-base. - BSC (56). BNB Chain, bloque 0.45 s. bscscan no tiene API gratis; se lee por RPC (blastapi, NodeReal sirven estado de archivo; los dataseeds no).
- Blockscout. El explorador de Robinhood Chain:
https://robinhoodchain.blockscout.com/tx/<hash>.blockscout.keyen la carpeta de la bóveda. - Bundlers ERC-4337. Los que empaquetan operaciones de cuentas inteligentes. MoneyLord VENDE por bundlers (11 firmantes en 12 transacciones): una venta es un token que sale de él con un Swap en el mismo recibo, lo lleve quien lo lleve. En un bundle hay que casar SU pata con SU swap, no con la de otros usuarios.
- Chainlink. Oráculo de precios en cadena. En este proyecto los precios salen de DexScreener, GeckoTerminal, Jupiter y los propios swaps del pool; Chainlink solo aparece como referencia en los informes de cadena, no en el camino de decisión.
- DexScreener. Sitio y API pública de pares y precios (todas las cadenas). Lo usan NOSE, el precio del hub, el buscador Solana. Lista ocho tokens con nombre de GameStop en Robinhood: por eso murió la puerta 4b.
- EIP-7702. Cuentas delegadas:
eth_getCodedevuelve 23 bytes que empiezan por0xef0100. Es una WALLET (una persona), no un contrato. MoneyLord y Overdose son 7702: "quién firmó" nunca es "quién operó"; se datan por lo que LLEGÓ, nunca por lo que enviaron (txlistvacío). En la puerta Relay, 97.9 % de los receptores son 7702 y los EOA planos son los robots. - Jupiter. El agregador de Solana: cotización (
quote) y swap. Lo usa el copiador Solana (solana_send.py,solana_exec.py) y el finder para precios. Un fee de Token-2022 va en el precio; un freeze lo lee el escáner. - Native USDC / USDG. Arc paga el gas en USDC nativo; Robinhood Chain tiene USDG (
0x5fc5360d…, 6 decimales) como dólar. Un pool cotizado en una acción tokenizada (NVDA) es una UNIDAD, no dólares: E0 rehúsa un precio en dólares hasta leer el de esa acción. - Permit2 (
0x000000000022D473030F116dDEE9F6B43aC78BA3). El contrato de permisos de Uniswap: la allowance va a Permit2, no al router, y la venta lleva una firma EIP-712. Una firma Permit2 es un PERMISO PERMANENTE a un spender, por eso se comprueban spender, token y chainId contra la cotización antes de firmar (ley "Permit2: una firma es una permisión, no una transferencia"). - pump.fun. La rampa de memecoins en Solana. Su API (
/v2/coins,/v1/...) alimenta el finder; su API de usuarios resuelve handle o wallet a userId, X y wallets canónicas (Overdose confirmado 09-05). - Relay (
api.relay.link). El puente que usan los traders de FOMO: pagan en Solana y reciben el token en la cadena EVM por el receptor0xccc88a9d…; el router0xb92fe925…emite el evento de entrega0xafbab204…con(from, to, token). En Solana el depositario es7uTT8Xi5…y el programa99vQwtBw…; el trader es el firmante #1 (el #0 es el pagador de Relay). Su API pública da el USD de ambas patas. - Robinhood Chain (4663). Arbitrum Orbit (Nitro), bloque 0.10 s, secuenciador de Robinhood, feed público del secuenciador, Uniswap v4 con hooks, USDG y acciones tokenizadas en la misma tabla que las memecoins. Verdict de la investigación: "solo tamaño degen".
canonical/logic/ROBINHOOD-CHAIN.md. - Secuenciador. El nodo que ordena las transacciones en una L2 (no hay mempool pública): copiamos transacciones CONFIRMADAS, uno a pocos slots por detrás. El front-running de L1 está muerto (2024, 99.86 % sin presencia en mempool).
- Solana. Cadena del copiador de cron (2 min) y del finder. Wallet
3iP5iL1p….getBlockatransactionDetails: "accounts"una llamada por slot; una hora de Solana cuesta dos horas de lectura en el host gratis. Base58 es sensible a mayúsculas: nunca poner en minúsculas. - Uniswap v4 / hooks. El DEX de Robinhood Chain: un PoolManager único, pools con hooks (que cobran 3 a 4 %), y el settlement en ETH nativo que NO emite Transfer (el evento del router
0x8619026aes lo que la wallet cobró; el fee event0x205442d6es el 1 % del router). - Wallet EVM nuestra.
0xc6D97089…, la misma en todas las cadenas EVM; llave en la bóveda, firmasigner.py.
11. La forma de trabajar (lo que ya se pagó)
- Ficha de verificación. "Verificar en la caja o en la cadena antes de decir hecho. Nunca desde un resumen" (HANDOFF 11). Un log muerto que parece vivo (
take-profit-cron.logde la raíz) ya engañó a dos agentes. - Ley 34 (leer la cola). Un trabajo periódico en un daemon de dinero vivo lee un número acotado de bytes y nunca se solapa consigo mismo.
- Ley 35 (coste real). Un número demasiado bueno es un bug del lector hasta que se pruebe lo contrario.
- Ley 37 (la caja no es banco de pruebas). Fixtures en el Mac; candados en
.locks/. - Ley 38 (reparar un archivo que un proceso tiene en memoria es una carrera). Parar el proceso primero.
- Ley 40 (un breaker que nadie eligió). Cualquier parada pagina a Dan y sale en rojo.
- Ley 41 (regla no pedida). Se borra con sus tests, no se afina. Cada umbral lleva la cita de Dan al lado.
- Ley 42 (un freno se juzga por lo que deja pasar y por lo que rechaza en ESTA cadena). El stop de un solo tenedor de LP rechazó MUSEBOOK (5x) y CPU (3x) y dejó pasar FROGE y PARE.
- Ley 43 (presupuesto de vigilantes). Nunca toca el camino de ejecución; un reinicio nunca vacía una caché del camino de dinero.
- Una lectura fallida nunca es un hecho. Escribir "no pude ver" y la cuenta. Un
exceptdesnudo alrededor de una lectura está prohibido. Control negativo antes de creer un cero. - Prueba lo que corre. La suite en verde no prueba un nombre de campo que la suite nunca da; una repetición en vivo de una transacción real corre antes de cualquier cambio de tamaño.
- Reportes. Cortos, en español, sin rayas largas, "como si tuviera 5 años"; dinero y estado verificados en cadena; Dan decide con 2 o 3 opciones y un voto.
- Reinicios. De uno en uno, 10 s entre daemons;
wallet-solanatarda 30 s en asentarse. - Sin agentes despertadores. Pocos agentes, brief completo, que terminen; sin monitores que los despierten (gastan la cuota de Dan).
⚖️ Las leyes (THE-RULES)
Fuente canonical/logic/THE-RULES.md · 73 KB · 43 secciones · modificado 2026-09-17 20:47 UTC. Cada ley se abre al tocarla.
THE RULES. One page, for everything we are building.
Dan, 2026-09-06: "establish the fucking rules well for all we are working on".
Every rule here was paid for. Each one names the incident that bought it. A rule with no incident behind it is a preference and does not belong on this page.
1. PART 1. THE OBJECTIVE
Dan, 2026-09-06: "there is no what I do. I can do a fucking lot of it ends up me with millions. In theory I'm open to copy anyone anyhow if it's my most likely and efficient way to."
RANK ON EXPECTED DOLLARS PER COPIED TRADE, at a $100 ticket, after our own measured cost, with an exit we can actually see. Three terms, never blended into one score:
- What the wallet makes, on the positions we could have entered.
- What our entry costs us, its OWN measured penalty at our lag, never a generic number. Measured 2.81 to 4.12 percent across four wallets, so one figure mis-ranks them.
- Whether we get an exit signal at all, and how much of the position it covers.
Style is not a criterion. Sell-everything is the CLEAN case: one exit, full coverage, nothing to guess. Sell-some-and-hold is not better because Dan does it; it is better only if the moonbag beats the exit signal it costs us. Never-sell is out for a copyability reason, not a style one.
The row that proves the objective: 0x585cf5b5…900d makes +5.58 percent and still LOSES a copier money, because its own entry penalty is 3.86 and the round trip is 2.3.
And Dan, 2026-09-06: "the market started a month ago". Old wallets with proven records that are ACTIVE NOW are the prize. So is whoever causes the pumps. Tuning our own entries and exits comes AFTER we understand exactly how they bought and sold, never before.
2. PART 2. IS THIS HIS TRADE? The four-fact test.
This is the most expensive question in the project and it cost three lanes a day to get right. Every single-fact version has a counterexample, measured.
Fact 0. eth_getCode FIRST. It tells you which other test to trust.
| result | meaning |
|---|---|
| 0 bytes | plain EOA. The sender test applies. |
23 bytes starting 0xef0100 | EIP-7702 delegated. The sender is ALWAYS somebody else. |
Overdose and MoneyLord are both delegated. A from == wallet test rejects 100 percent of the real trades of the two wallets Dan pays to copy.
But delegation does not FORCE a wallet to stop signing (0xcb4d28c2106f… signs for itself anyway). Fact 0 narrows the question; it never answers it.
Fact 1. Something was PAID for it.
A swap log in the same receipt, and the transaction pays a HANDFUL of addresses rather than hundreds.
Its counterexample: 0x73691cee…3b1 passes this on 83.5 percent of its arrivals and is an airdrop magnet. A modern BSC airdrop routes through a DEX and pays about seven recipients.
And it is CIRCULAR on a population selected for swaps. On the Robinhood 1,911 it passed 9,515 of 9,515, because the analyzer built that population with "a Transfer in a transaction that also swapped this pool". A 100 percent pass rate that is the selection rule echoing back is not validation.
Fact 2. THE WALLET ALSO SENDS.
Buy-only is not trading whatever the receipt looks like. 0x73691cee…3b1: 227,179 in, 136 out, 0.1 percent. A real trader: 58.7 percent.
Fact 3. THE DOOR. Where the value actually passed.
The strongest single fact, and the only one that catches a relay trader hiding in a plain EOA. Measured by the relay lane over Relay's own record plus the chain, 167 of 167 hashes read, zero could-not-see:
| rule | true positives | false positives |
|---|---|---|
his own leg faces RelayRouterV3 0xb92fe925dc43a0ecde6c8b1a2709c170ec4fff4f | 228 / 228 | 0 / 564 |
Relay receiver 0xccc88a9d1b4ed6b0eaba998850414b24f1c315be as tx.to | buys only | 0 |
| Solana payment within 15 s | 78 / 78 | 1 / 346 |
tx.to == Multicall3 | — | 49 of 101 airdrops |
| the signer list | stale by design | — |
The router rule works in BOTH directions (token FROM it on a buy, TO it on a sell) and describes HIS OWN LEG rather than the transaction's front door.
NEVER GATE ON A SIGNER LIST. Overdose has 28 distinct signers over 167 fills, 6 appearing once. MoneyLord's 13-address list would miss 25 of Overdose's 78 buys, 32 percent, SILENTLY, because an unlisted signer just reads TRANSFER_IN.
The residual risk, named rather than carried quietly
An attacker willing to SPEND could buy a token through Relay and have it delivered to a wallet we are known to copy, and it would pass. Overdose publishes his own address. It costs the attacker real money, so it is not free, but it is possible. The Solana lane put it best: anyone who airdrops a wallet we copy can aim our money. Any wallet receiving Relay deliveries it never paid for is a live attack and goes to the orchestrator the same hour.
🚨 THE LAW THAT REPLACES THREE FAILED METHODS
EVERY METHOD THAT ASSUMES A WALLET TOUCHES THE CHAIN ITSELF BREAKS ON A 7702 ACCOUNT. Three separate techniques died on that one shared assumption in a single night, and each looked sound in isolation:
| method | how it broke |
|---|---|
| "did he sign it" | drops a relay trader entirely; Overdose signs nothing and holds $103,730 |
| the pending-state gate | died with the public mempool; it was Dan's single biggest filter, 546 rejections |
| the CEX-funder trace | txlist is EMPTY on a relay wallet, not because it is idle but because it never sends, so it was never funded with gas, so there is no first funder |
The wallets we want are exactly the wallets these methods cannot see.
THE INVERSION THAT WORKS: enumerate the DOOR, not the person. Every relay trader receives through 0xccc88a9d1b4ed6b0eaba998850414b24f1c315be. Grouping that contract's deliveries by recipient finds the population directly, with no handle needed, and it also finds traders nobody has ranked.
RUN 2026-09-06, AND IT WORKED: 120,606 addresses in 72 hours across BSC and Base, and 49 leaderboard names given a real tradeable wallet. Full page and every control: map/wallets/DOOR-HUNT.md. Three things it taught that the plan did not contain:
READ THE HINGE THAT LOGS. The receiver is the transaction's to and carries no recipient in an indexed topic, so reading it costs 11 KB of calldata per delivery. One log later RelayRouterV3 0xb92fe925dc43a0ecde6c8b1a2709c170ec4fff4f emits 0xafbab204e8271965231d37baed9b1abca8725b7409c70314455f68bc89142b91 with (from, to, token) in its data, and that to is the end user. Same door, same transaction, one hinge logs and the other does not. 17,423 of those an hour against 18,000 deliveries an hour, so it is one per delivery and not a sample.
🚨 ON THIS DOOR A PLAIN EOA IS THE ROBOT AND A 7702 ACCOUNT IS THE PERSON. Of 22,747 addresses taking 5+ deliveries on BSC: 22,260 are 7702 (97.9%), 455 are plain EOAs (2.0%), 32 are contracts (0.1%). The two busiest recipients in the whole window are plain EOAs, and they are solvers. A "is it a contract" filter keeps both and throws nothing useful away. 87.3% of the 7702 accounts delegate to one implementation, 0xe6cae83bde06e4c305530e199d7217f42808555b, so the door population is one app's wallet base, which is the base the leaderboard is drawn from.
🚨 A MEDIAN DEPENDS ON THE UNIT YOU COUNT. DOLLARS DO NOT.
FIFO lot matching splits one sell across every buy lot it consumes, so a wallet that accumulates in small pieces gets many lots per sale and its median lot return is dragged down by construction. NoGiOlMoKiNi reads -40.69% by lot and +6.20% by matched buy-sell pair over the same two chains. Both are honest. Only the realized dollars are invariant to how the lots were cut, which is why money is the ranking rule and every ratio is printed with its count beside it.
AND THE POOL'S MOVE IS NOT THE TRADER'S RETURN. walk.py prices a round trip as the pool's price at the sell block over the buy block: a fact about the token. Measured against his own dollars on 188 matched pairs: median gap -1.86%, mean -5.08%, and he did worse than the pool in 141 of 188. So a ratio-based ranking flatters these wallets by about two points a round trip, consistently. Subtract it; do not discard the walk over it.
🚨 THE FUNDER TRACE NEEDS TWO FILTERS BEFORE IT SAYS ANYTHING TRUE
Dan's method, run on the 49 real wallets, 2026-09-06. Full page: map/wallets/THE-49-CHARACTERISED.md.
On half of them there is nothing to trace: the first thing that ever arrived was Relay itself, on 25 of 49. A 7702 wallet pays no gas, so it was never funded the way a 2022 Ethereum wallet was.
And on a DEX everyone's counterparty is everyone. Every address that ever sent these wallets a token gives 1,205 senders, median 59 each, because the token arrives FROM THE POOL on every buy. eth_getCode removes 522 contracts; fan-out removes the rest, since an address that sent to 44 of our 49 is a market participant, not a second wallet. The 96 human-shaped senders shared by exactly two people track TRADING VOLUME, not identity: rodotfun alone is in 27 of them at 100 trades a day.
What survives is small and real: 15 of 49 have a human-shaped first funder, and TWO PAIRS share one. That is a lead, not a conclusion; one person with two wallets and two people funded by the same friend look identical here.
🚨 ON SOLANA, WHO SIGNED IS NOT WHO TRADED
Solana lets one account pay the fee for a transaction another account also signs, and Relay does exactly that. Checked against Relay's own named payer on six fills: the trader was signer #1 every time, and AgmLJBMDCqWynYnQiPCuj9ewsNNsBJXyzoUhD9LJzN51 was signer #0 every time.
Taking signers[0] as the payer collapsed 2,961 real traders into 570 and handed that one address 78% of all payments in the window. Nothing errored. The only reason it was caught is that 78% is not a plausible share for a trader, which is the general defence: a leaderboard where one row holds most of the volume is a reading error until proven otherwise.
The Solana door, both proven on 1,563 of 1,563 Solana-paid fills: depository 7uTT8Xi5RWXzy7h9XL244GRgEycDYDhLjr3ZyNdXi8pZ, program 99vQwtBwYtrqqD9YSXbdum3KBdxPAVxYTaQ3cfnJSrN2. Read it with getBlock at transactionDetails: "accounts", one call a slot, because getSignaturesForAddress covers 54 seconds per 1,000 rows and names no signer. One hour of Solana costs about two hours to read on the free host, so a 90-day Solana sweep is five months of reading and must not be promised.
🚨 COVERAGE IS NOT A CAVEAT, IT IS THE ANSWER OR THE OPPOSITE OF IT
Re-walked the eight shortlist wallets with archive-verified hosts and 60 pages of history instead of 6: priced round trips went from 61 to 373, and the verdict REVERSED on most of the list.
| wallet | priced before | priced now | median round trip before | now |
|---|---|---|---|---|
| freyaa | 5 | 62 | -0.37% | -26.12%, wins 17.7% |
| BlankCanvas4K | 1 | 6 | +156.95% | -51.68% |
| bigwarzeth | 1 | 25 | +35.83% | -9.46% |
| athathath10 | 9 | 88 | -30.53% | -34.38%, wins 19.3% |
| BREWto100M | 6 | 51 | -56.06% | -10.18% |
| tikopumps | 5 | 69 | +10.04% | -4.46% |
Only 2 of 8 have a positive median round trip once properly measured. A low win rate under a high published PnL is the "pego un mega palazo" shape Dan rejected by hand in 2022, and it is most of this list.
🚨 AND THE HOST WAS THE HIDDEN HALF. 78 of 210 trips were unpriced because the rotation contained hosts that cannot serve archive state at ALL: both BSC dataseeds error at every depth including 1,000 blocks back, and so does publicnode on Base. Only blastapi and NodeReal (BSC) and blastapi, mainnet.base.org, drpc and developer-access (Base) answer. Verify a host for the METHOD you will call, at the DEPTH you will call it.
🚨 A WALLET WHOSE ANSWER MOVES UNDER MORE DATA IS NOT SETTLED. cosmic_xbt went 30 priced to 13 while its trips FOUND went 40 to 63, because a deeper pull re-pairs buys to sells; its median moved +369.69% to +128.13%. Both are real measurements of different pairings. Report the instability, never the prettier number.
🚨 A LAG CURVE BELONGS TO A POPULATION, NOT TO A CHAIN
Re-measured 2026-09-06 on the 49 door wallets, both ends walked at 8.159 s: 295 round trips priced, median lag cost +0.00%, trip-weighted mean -0.13%, and OUR WIN RATE EQUALS HIS IN 44 OF 46 WALLETS. Full page: map/wallets/DOOR-SCORED.md.
The earlier curve said 8 s was ruinous, break-even about 2.7 s, all the damage inside the first 4 seconds. That curve was true and it was true about 88-second scalpers, where 8 seconds is a tenth of the whole hold. These wallets hold for hours and days. The same 8 seconds barely touches them. A lag penalty is a ratio of our delay to THEIR hold time, so it has to be re-measured on the population being copied and can never be carried across.
🚨 AND A LAG COST NEAR ZERO IS NOT A LICENCE TO COPY. It only says the clock is not what stops us.
🚨 A TRANSFER-BASED READER IS BLIND TO NATIVE CURRENCY, AND BLIND QUIETLY
Two ways to read the same door, run on the SAME 2,000 BSC blocks: Transfer where from is RelayRouterV3 gave 13,364 logs and 2,697 recipients; the router's own delivery event 0xafbab204... gave 3,656 logs and 2,453 recipients. They find the same traders. The Transfer filter carries 3.7x the volume because it also fires on every router-to-pool hop, and of the 259 recipients it has and the event filter does not, every one of the first 300 checked is a contract, which is exactly what the eth_getCode step exists to delete.
But the event filter finds 15 addresses the Transfer filter CANNOT, 12 of them 7702 traders. Their deliveries are native BNB, token 0x0000...0000, and native currency emits no ERC20 Transfer log at all. There is no error and no gap in the output: those wallets simply are not there. The receiver, for the record, returns 0 rows on the same query, which is the proof that it hands nothing over.
🚨 THE HOLD GATE: THIS BOARD IS YOUNGER THAN DAN'S FILTER ALLOWS
Step 7 of his own hunt, Metodos de busqueda!C3: "te vas a analytics para ver si ha tenido dinero durante un periodo de tiempo respetable". Measured on the 49 matched wallets, dated from each one's FIRST inbound token transfer:
| age on chain | wallets |
|---|---|
| under a week | 5 |
| under 30 days | 23 of 49 |
| 30 to 90 days | 13 |
| 90 to 365 days | 13 |
| over a year | 0 |
Median age 36 days. The oldest is 322 days. This is a board of new money and his gate exists precisely to refuse it.
🚨 DATE A 7702 WALLET BY WHAT ARRIVED, NEVER BY WHAT IT SENT. txlist is empty on every one of them, so an age taken from sends dates the whole population to never.
THE TWO GATES TOGETHER: 49 wallets -> 8 pass the bet gate -> 26 pass the hold gate -> 6 pass both. Named on map/wallets/DOOR-SCORED.md.
🚨 THE BET SIZE IS THE BINDING GATE, NOT THE CLOCK
Dan wrote it in 2022, Sheet1!J10: "Preferible un wallet que pegue palazos poniendo poco a que lo haga poniendo mucho, para tener mas odds de replicar". Measured on the same 49, from Relay's own record which carries the USD on both legs:
| median bet across the 49 | $1,452 |
| range | $50 to $24,997 |
| wallets with a quarter or more of buys at or under our $100 | 8 of 49 |
| wallets with fewer than one buy in ten at our size | 30 of 49 |
So the leaderboard is mostly unusable to us, and not for any reason to do with speed. At $100 against a $2,000 median bet we are not copying a trader, we are taking a tip. The eight that pass are named on the scored page.
THE JOIN BACK TO A NAME IS WHAT THEY HOLD, AT FULL PRECISION. FOMO's API returns 24956945.150284983, not the screen's rounded "24.9M", so the match is an equality and not a band. The risk is TIME, not precision: MoneyLord matched one holding to 1.4e-16 and missed another by 1.2% because he traded after the snapshot. A near miss is never promoted to a match. Positive control: the bridge rediscovered his known wallet out of 4,592 candidates.
🚨 THE PUBLISHED WALLET IS THE DECOY
Measured at n=30, then proved on a person whose real wallet we already knew. Of the top 30 traders on FOMO's board, 30 of 30 have never sent a single transaction from their published EVM address.
MoneyLord's published address carries 53 transfers ending at block 119,731,367. So do DumbCrayonEater's and ogle's. The same 53, the same last block, the same tokens. One airdrop batch, three dormant addresses, two belonging to people who traded within the hour on wallets they do not publish.
His real one traded 3 minutes ago and holds 12,027,773 GME that nobody else in our set holds a unit of.
A leaderboard address is a label. It is not where the money moves.
🚨 HOLD TIME DECIDES COPYABILITY BEFORE PROFIT DOES
Measured both ends, at our real 8.159 second lag, not modelled on one end:
| population | median hold | our lag as a share of it | verdict |
|---|---|---|---|
| the BSC wallets we spent a day scoring | 88 seconds | 9.3 pct | dead to us at any profit |
| the FOMO board | 1.58 hours | 0.14 pct | copyable |
Only 4 of 272 board traders are scalpers. The scalpers also make the most money, median $1,436,373 over 30 days, and not one dollar of it is reachable.
SCREEN ON HOLD BEFORE YOU SCREEN ON ANYTHING ELSE. A wallet can be right and still uncopyable, and only a both-ends measurement finds it: a one-ended entry penalty scored the best wallet at +1.77 percent when the truth was −3.43.
🚨 A DISPLAY WARNING IS NOT A TRADING GATE
impersonates_currency was built so a card would not print "5,881.92 USDG" for a scam token, and its own docstring says the DECISION was already safe because is_currency compares addresses. Somebody then wired that warning as a refusal.
Cost: three real MoneyLord buys of "Greatest Meme Ever" in seven minutes, all three decoded correctly as BUYs and all three refused, because Robinhood Chain lists tokenised equities in the same table as money and a memecoin named after a stock ticker collides with one. Dexscreener lists EIGHT GameStop-named tokens on that chain.
Before a warning becomes a refusal, read what it was built to protect.
3. PART 3. NOT GETTING TRAPPED. LeBot's own doctrine, 2023, still law.
From 133 real mainnet attempts, one shape dominates:
IT IS NOT "WE PICKED BAD TOKENS". IT IS "WE GOT IN AND COULD NOT GET OUT."
So the gate is an EXIT REHEARSAL, NOT A SCAM CLASSIFIER. "Can I sell this, right now, at my size" is a measurement. "Is this a scam" is a guess.
| check | refuses when |
|---|---|
| exit rehearsal | a simulated sell returns nothing (honeypot or dead pool) |
| exit rehearsal | the round trip loses more than 15 percent |
| balance | the wallet holds the ticket and no gas, because no gas cannot SELL |
| all of them | the input could not be read |
Jose Ruggiero, 1 Feb 2023: "si es muy complicado, yo preferiria q no copie la trx en vez de copiarla infinito y fallar".
A REFUSAL COSTS ONE MISSED CALL. A TRAPPED POSITION COSTS THE WHOLE TICKET. "Could not simulate" is NOT "fine".
It is live and it fires. Measured 2026-09-06 at a real $100 ticket: sell back $99.95 of $100, round trip 0.0 percent. And it refused a real Overdose buy the same day because the scanner could not confirm the coin can be fully sold.
What it does NOT do: it cannot see a token that turns malicious LATER. That is the exit layer's job.
4. PART 4. READING. A number you did not measure is not a number.
- A FAILED READ IS NEVER A FACT. Write "I could not see" and the count. 2026-09-06: reported "0 transactions in 42 hours" for both target wallets. It was a wrong function signature swallowed by a bare
except. They had traded 12 times in 27 hours and 6 minutes earlier. - NEVER a bare
exceptaround a read. Count and print what you swallowed. - RUN A NEGATIVE CONTROL before you trust a zero. Query something you know exists. The control that caught the above returned 3,644 rows.
- TWO READERS AGREEING IS NOT PROOF WHEN BOTH SAMPLED THE SAME WRONG STRATUM. 2026-09-06: two lanes independently "confirmed" a wallet was not a trader. Both samples landed in its airdrop overlay, which outnumbers its trading twenty to one. Cross-source agreement is only as good as the sampling.
- A FULL PAGE THAT SAYS NOTHING ABOUT BEING CUT IS A SILENT TRUNCATION. Etherscan V2 caps at 10,000 rows and does not say so.
- TEST THE METHOD YOU ACTUALLY USE, never a proxy.
solana-rpc.publicnode.compassed agetSlothealth check and refused every real request. - A DEFAULT THAT NAMES ANOTHER ENVIRONMENT IS A CROSS-WIRE. A chain with no sequencer feed fell back to Robinhood's. MoneyLord uses the same address on both chains, so his next Robinhood trade would have been bought ON BSC.
- AN ESTIMATE IS NOT A TIMING, AND THE SIZE MATTERS. The 2.3 percent round trip was measured at $1 tickets. At $100 the real numbers are 0.19 to 1.3 percent by chain.
- A MEDIAN COLLAPSES THE DISTRIBUTION YOU CARE ABOUT. A wallet with 18 of 21 positions partially sold was reported as "median 0.80" and vanished. Print the count beside every median.
- AGGREGATE PER POSITION, NEVER PER LEG. Dan's own
agg.pyalways summed every buy and sell leg per token. A rebuild divided by ONE purchase, so a wallet that bought more read as "never sold" on 39 percent of positions. His tool was right. Check it before rebuilding a metric.
5. PART 5. MONEY. Nothing here is negotiable.
- NO SPEND WITHOUT DAN'S WORDS IN THE TURN. A file cannot raise a cap Dan set in words; his words are recorded in code with the date and the quote.
- ARMED IS NOT FIRING. Four of six targets were armed at $100 and could fire on none, because the wallet signs nothing. Prove a target with a replay of a REAL transaction before calling it live.
- MEASURE EACH CHAIN SEPARATELY. A door proven on BSC is not proven on Base.
- FAIL CLOSED, AND SAY WHY. An empty allowlist is deliberate and says so. A chain with no executor refuses BY NAME, never by omission.
- NEVER PRINT A SECRET. Names in the register, values in a file outside the repo at mode 600, created at 0600 rather than chmod'd after.
- A CREDENTIAL GOES IN THE REGISTER THE SAME TURN, and the row names EVERY machine that needs it. The paid Etherscan key sat on the Mac only, so every server call ran anonymous. A paid key on one machine is not a paid key.
- ONE LIVENESS STAMP PER PROCESS. Three watchers shared one file, so a blind chain looked alive because another had stamped for it.
- A TEST THAT ASSERTS A SUPERSEDED INSTRUCTION IS A LIAR. Six tests asserted the retired $1 cap and failed against a correct system. Six known failures hide the seventh. Update the test, keep the history in its docstring.
6. The top traders pay off-chain, so the chain shows WHAT they bought and never WHAT they paid
Measured 2026-09-07 on unipcs, BSC, 90 days. 2,222 token transfers over 2,174 transactions. Of those:
| shape | count |
|---|---|
| a token leg AND a native leg (a readable swap) | 13 |
| a native leg with no token leg | 266 |
| a token leg with NO payment visible anywhere | ~1,895 |
And the 13 "both legs" transactions are not swaps either: in every one, BOTH the wrapper and the token arrive (dir=in), which is a claim or a distribution, not a purchase. txlist confirms the cause: he is the sender in 0 of 5 rows, because a relayer sends for him. These are EIP-7702 smart accounts and the money settles somewhere else, exactly as Overdose and MoneyLord already do.
What it costs us. Profit cannot be measured from transfer data alone for this class of wallet, which is most of the ones worth copying. We can read, exactly: which token, how much, when, and whether it was later sold. We cannot read what they paid. Any profit number for them needs a price lookup per fill, which is a second measurement with its own coverage, not a field on a transfer.
The rule. A ranking built only on transfer reads will report every relay-funded wallet at zero profit and rank them last. That is not a measurement of them; it is a measurement of our reader. Coverage must be stated per wallet, and a wallet whose payment legs are invisible is PRICE UNREAD, never no profit.
How it was caught. Not by the ranking, which happily printed BNB OUT 0.0000 on 1,903 trades three times in a row while I patched the wrong layer twice. It was caught by printing one real transaction's legs. Print the raw row before patching the aggregate.
7. A flash loan is not his fill
2026-09-07. fill_price took the largest transfer of the quote currency in a receipt as the wallet's payment. Correct for a plain swap. Wrong for a complex one: wallet 0xad01c20d, an arbitrage bot with 18,915 round trips in 90 days, reported one position at 1,031,908 ETH over 3 fills and a wallet total of minus 1.3 million ETH. The number was a flash loan passing through the same transaction.
The rule. A leg is HIS when it touches his wallet or the pool that emitted the Swap. The whole-receipt maximum is the last resort only, because a relay-funded fill touches neither and would otherwise be lost.
How it was caught: not by the total, which had already been wrong three different ways today, but by ranking the wallet's own positions and reading the worst one. An impossible number always has one row behind it. Find that row.
8. pgrep -f <name> matches your own command
2026-09-07. Checking a background job with ssh box 'pgrep -f rank_batch && echo running' prints "running" forever, because the ssh command line CONTAINS the string it greps for. A dead batch was reported alive for half an hour on that basis.
The rule. Liveness is a PID file plus kill -0, which cannot match the checker. This is the same shape as the recorded law that a wait loop grepping its own name never exits; it came back wearing a different hat.
9. nohup ... & inside an ssh one-liner dies with the connection
2026-09-07. The batch was launched as ssh box 'nohup cmd &' and kept dying minutes later. It looked like a timeout in the job and it was the SSH session closing and taking the process group with it. Two wallets were recorded as "the read timed out" at 15 minutes when the timeout was set to 90.
The rule. setsid nohup cmd > log 2>&1 < /dev/null & and write the PID. Detach stdin too: an inherited stdin is enough to kill it.
10. Two of our own jobs can starve each other
2026-09-07. Every chain read in a probe was refused, which read as "the pair stopped trading". The cause was another of our own jobs holding the same host's rate limit. The evidence was identical to a dead market.
The rule. Before believing a refusal, check what else WE are running. A rate limit reached by our own second job is not a fact about the chain.
11. A subprocess picks up a fixed file; a long-running parent does not
2026-09-07. A bug was fixed mid-batch. Because each wallet is read in a FRESH subprocess that imports from disk, the fix went live from the next wallet onward with no restart. The six wallets already read kept the bug, so they went on a written redo list rather than being quietly kept.
The rule. When a fix lands mid-run, say exactly which results predate it and redo them. A mixed batch with no record of the split is worse than either half.
12. The venue is measured, never inherited
2026-09-09. LEBOT routed every buy and every sell through 1inch because that is what got wired first. Nothing ever compared it to anything. Dan put the 1inch and Uniswap swap screens side by side for the same 35,000 ASTEROID and found:
| size | 1inch versus Uniswap |
|---|---|
| buy 0.01 ETH | +6.0% |
| buy 0.20 ETH | -18.9% |
| sell 10,000 | -19.8% |
| sell 35,000 | -26.8% |
| sell 82,898 | -35.8% |
1inch is fine on dust and collapses at our real ticket size, in BOTH directions, so a copy bought ~10 to 19 percent fewer tokens and then sold them for ~20 to 27 percent less. It compounds on both ends of every position.
Worse, it poisoned a safety gate. Gate 6 rehearses a $100 buy and an immediate sell and refuses anything whose round trip costs over 15 percent. That round trip was measuring 1inch's own spread, so honest tokens were being judged as traps.
And it fooled me. From the same evidence I concluded "this token is expensive to trade" and told Dan the pool was thin. The pool was not thin; the router was bad. A cost you observe through one venue is a fact about that venue until you price it somewhere else.
The rule. Every chain names its router in chain_map.json with the measurement that chose it, at OUR ticket size, against at least one alternative. A router that was never compared is an assumption wearing a number.
13. Read the calldata off a trade that worked HERE
2026-09-09. Building the Uniswap route means producing the exact bytes we sign. The shape was taken from a real Universal Router trade on Robinhood Chain (0xdcb97367ff779be4d1663d9dcb03aed94115380524aa4a0e932363832e008ab9) rather than from another chain's docs: execute(commands, inputs, deadline) with command 0x10 V4_SWAP, whose input is a second layer of abi.encode(actions, params) carrying 07 0b 0e = SWAP_EXACT_IN, SETTLE, TAKE.
Two layers, and getting the inner one wrong signs a transaction that reverts or swaps the wrong way. Documentation describes the general case; the chain shows the one that works.
14. The cost basis is HIS OWN last buy, and only what left HIS wallet
2026-09-10. The take-profit rule sold 30 percent of a DARK position Dan had opened five minutes earlier, at a price that had not moved. It reported 2.19x. He said, in order: "Pero como calculo el 2x si compre hace 5 min y lo que hizo fue mantenerse o bajar un poco!", then "El 2x debe ser desde que yo compro!", then, when I showed him my numbers, "Esos precios son erroneos, no hubo tanta diferencia entre uno y otro". He was right all three times, before a line of code had been read, and each correction found a different bug.
Bug one: it averaged a lifetime of buys. Four DARK entries went into one spent / qty, including a token-for-token swap it had misread as a sell. The blended basis sat far under what he had just paid, so flat read as 2x.
Bug two: it read a leg of the ROUTER instead of his payment. On his first buy it took 0.021040 WETH as the cost when he had paid 0.105200 ETH. A router splits one purchase across pools; the reader saw the smaller half. A cost read at one fifth makes a position look five times more profitable than it is.
Bug three, found by trying to fix bug two. Reading the router's payouts instead was graded against the two buys whose amount the transaction states outright, and failed both ways: +20.1% on one, because it counted an intermediate currency twice (the WETH, and again the USDG that same WETH had just bought), and -100% on the other, because a Uniswap v4 route settles native ETH without emitting a single Transfer. Both are inferences from someone else's bookkeeping.
Bug four, which would have been silent. The state key was chain:token, so a token trimmed once could never be trimmed again. Dan bought back in minutes later and that new position was permanently unwatched, with no error anywhere. The docstring already promised a re-buy is a new position; the key did not keep the promise. A docstring is not an implementation.
The rule. The basis is his MOST RECENT buy, never an average. The cost is what left HIS address: the transaction's own value for native ETH, plus WETH or a stable he sent. A buy paid in another token is REFUSED, not estimated, because pricing it needs that token's value at that block and that is a guess. The state key names the buy it trimmed.
The control that caught bugs two and three is free and must always be used. Two of the four buys were paid in native ETH, so the transaction states the amount outright. Any cost rule can be graded against a number it never saw. Both wrong versions looked completely reasonable in isolation.
And the scale is not the multiple. He remembered entering at $0.000016 and $0.000014; the chain says $0.001622 and $0.001452, exactly 100x apart, and the contract says 18 decimals with a 1,000,000,000 supply, which puts the market cap at $1.62M and settles it. His RATIO was perfect. When the operator and the reader disagree by a round power of ten, check the ratio before the scale: the ratio is what the rule trades on, and he is usually right about it.
15. A route is a CHAIN or a SPLIT, and they need opposite arithmetic
2026-09-10, same afternoon, found by Dan sending a DeBank screenshot. With his own numbers to grade against, the reader for token-for-token buys was wrong by about 2x, and the two buys it had to price took opposite shapes:
SPLIT GME -> USDG -> DARK two parallel paths; the cost is the SUM
GME -> WETH -> DARK
CHAIN UBIK -> USDG -> WETH -> DARK one path; the USDG and the WETH are
THE SAME MONEY at two points on it
Summing the chain read $155.36 where DeBank showed $77.34: exactly double. Taking the largest leg instead reads the chain right and the split at 63 percent. Neither sum nor max is a rule; they are two guesses that each happen to be right half the time.
The test that separates them: what was sold to get it. Value the router obtained by selling HIS token is his money entering the route. Value it obtained by selling ANOTHER value currency is the same money moving along the route, and was already counted upstream. With that, both shapes land inside 1 percent of DeBank on the same code.
Why it took a screenshot. Every version of this reader looked reasonable and returned a plausible number. Nothing in the receipt says "you paid $77". The operator's own wallet UI was the only independent surface, and asking for it would have cost one message at any point in the three hours before he offered it. When a derived number has no second surface, go and find one.
16. A gate must count the world, not its own notebook
2026-09-10. LEBOT bought nothing for a day. Twelve signals refused, JUGGERNAUT among them, and no error anywhere. Gate 8 caps open positions per target and it counted LEDGER ROWS. MoneyLord's row count said three. The wallet held zero.
Three separate events pulled the notebook away from the world, and not one of them was a bug on its own:
- LEBOT bought a token three times following MoneyLord.
- LEBOT SOLD it, twice, following OVERDOSE, because Overdose sold first and Overdose was the signal. The ledger books each row under the target that caused it, so MoneyLord's counter never saw a sell.
- A third sell was refused by the 1inch HTTP 402 outage, and Dan closed the rest by hand. A manual sale writes no row at all.
The rule. A risk gate reads the chain. still_held returns True, False or None, and only a balance positively read as FALSE drops a row: None means "could not see", and an unreadable balance must never open a gate. Exits LEBOT did not make get written back as SELL_OFFBOOK by reconcile.py every ten minutes, a separate kind from SELL so that no invented proceeds reach the P&L.
And the fix itself failed silently first. swap.SG.wallet_address() does not exist; the accessor is SG.creds()["WALLET_ADDRESS"]. The AttributeError hit my own except, which returned None, which means "could not see", which kept the gate shut. It failed SAFE, which is right, and failed SILENT, which is why it survived a verification run and was caught only because the printed count still said 3 instead of 0. Print the number the fix is supposed to change.
17. A suite that is nine red is a suite nobody reads
2026-09-10. test_decide had been 47 green and 9 red for days. Every one of the nine traced to a single stale number: the fixture write_armed() defaulted to a 1.00 per-order cap, from when LEBOT copied a dollar a trade, and Dan moved to 100 on 2026-09-06. Nothing was broken. But nine permanent reds are indistinguishable from ten, so the real regressions had nowhere to show.
Four different shapes of the same disease, all found in one afternoon:
- a FIXTURE freezing a number the code owns (
cap_order=1.0) - a TEST naming a limit the code owns (
* 3where the cap now reads 10) - a DOCSTRING updated without its body: the Solana test still set
p.chain = "bsc"and died on a fake token, long after BSC was opened - a test asserting a DISABLED feature: the cooldown gate, with COOLDOWN_S at 0
The rule. A test names the CONSTANT, never the number. When a test goes red after a deliberate change, decide which one is stale before touching either: the BSC send test was red because the code was right and the test held the old world, and the answer was to retarget the guard at Solana, not to delete it. 424 green, 0 red, and a red now means something.
18. The signal is the BURST, not the transaction
2026-09-10, Dan's rule, in his words. "Si un trader hace 6 compras rapidas del mismo token y gasta en total 10.000$, entra el trade por 500$. No hace un trade por cada trade. La suma total de sus compras sucesivas/rapidas es lo que se toma en cuenta para el sizing."
Sizing off one transaction reads a conviction bet as several small ones and answers the ladder with the wrong rung. Copying each one opens six positions where he opened one.
And the ladder could not read him at all. his_usd_of prices his side from a currency leg he gave: a stable, or ETH. His buys are token-for-token swaps, so there is no currency leg, and it returned None on EVERY MoneyLord signal that day. The floor applied every time. His real JUGGERNAUT burst, six buys inside 3 minutes 37 seconds, every one logged "could not price what he spent":
24,862 + 24,303 + 12,434 + 12,166 + 12,431 + 12,090 = $98,286
The fix is the router-leg pricing proved that same morning against Dan's DeBank screenshot, applied to HIS side instead of ours. The burst then sums it, one entry is opened at the rung, and a later buy of his that lifts the burst onto a higher rung TOPS UP the position instead of opening a second one.
The rule. Size on what he put in across the whole run, enter once, top up to the rung, never one copy per transaction of his.
19. I wrote the same bug twice in one patch, and the silent half is the dangerous one
2026-09-10. Decoded keeps the token on a LEG, reached by dec.token_bought().token. There is no dec.token. I used it in both halves of the same change:
- in
decide._size_buy, asdec.token- an AttributeError on the first real BUY, after every gate had passed. Loud, immediate, findable. - in
lebot.handle, asgetattr(dec, "token", "")- which returns "", so the burst never ran,his_usdstayed None, every ladder answer would have been the $100 floor, and NOTHING anywhere would have said why.
The crash is the good outcome. The default swallowed the mistake and left a feature that was written, wired, tested green and completely inert: armed and never firing, in my own patch, on the same day I wrote that law down.
The rule. A getattr with a default on a field you believe exists is not defensive, it is a silencer. Read the attribute directly and let it raise, or assert the field is there. And the suite passing means nothing about a field name the suite never supplies: only the live replay of a real transaction found this, which is why one now runs before any sizing change ships.
20. A ranking exists so the second best can be used
2026-09-12. The copier compared three venues, Uniswap won by 21.8 percent, and the trade executed through 1inch. Dan found it in a block explorer, not in our logs: "se ejecuto a traves de 1inch y no kyberswap, me preocupa que no consiguio el mejor quoting".
Two separate faults, and each alone would have been survivable:
- Uniswap's build was never wired. Their
/v1/swapwants the whole quote object echoed back, so the path was left off as unproven. Honest, and it meant the venue that most often won could never be used. - The fallback went to 1inch, not to the next best. Kyberswap was in the same ranking, ahead of 1inch, and able to build. The code read only the head of a list it had just sorted.
Cost on that one copy: about 43,000 FROGE of a 150,000 fill, near 22 percent.
The rule. When the winner cannot build, walk DOWN the ranking and stop at the first venue that beats the baseline and can build. Reading only the head makes every other row decoration.
21. Audit the chain, not the log
2026-09-12. The copier prints the venue it chose. That is what it BELIEVED. routing_audit.py reads tx.to from every live trade, which is what happened, and it found in one pass what no log line had said in six days:
every SELL on the chain went through 1inch
The buy path learned to compare venues on 2026-09-10 and swap.sell was never touched. Measured the same day, best against worst, on tokens actually held: FROGE 23.8 percent, ASTEROID 6.9 percent. A sell is where the money comes back, so the spread costs exactly as much there as going in.
And my first audit was wrong in the safe direction, which nearly hid it. It named the router as "whoever handed us the token", which on a multi-hop route is the LAST POOL. Two ASTEROID buys that the log correctly showed as Kyberswap read as an unknown contract, and I was one step from reporting working trades as failures. tx.to is who we called; everything else is inference.
The rule. Every claim about where money went is checked against the receipt, and the audit runs hourly so nobody has to think to run it.
22. A venue can win the price and still be unable to take your tokens
2026-09-12. Uniswap's router moves tokens through PERMIT2, not a plain ERC20 allowance: isTokenApprovalApplicable: true, permitData.domain.name: "Permit2", verifying contract 0x000000000022D473030F116dDEE9F6B43aC78BA3. That needs an off-chain EIP-712 signature this signer cannot make.
Found by doing it: a real 2 percent FROGE sell approved their router directly, built cleanly, and reverted at eth_estimateGas with 0xd81b2f2e. The approve had already confirmed on chain, so the attempt cost gas and bought nothing.
uniswap 974,571,906,956,062 needs Permit2
kyberswap 974,379,230,003,239 plain allowance, proven
1inch 920,622,077,053,209
The rule. Excluding a venue costs the spread to the next one, which here is 0.02 percent, and including a venue that cannot settle costs the whole trade. A BUY paying native needs no allowance, so the same venue stays in the ranking there. The question is never "who quotes best", it is "who quotes best AND can take what I am paying with".
23. Permit2: a signature is a standing permission, not a transfer
2026-09-12, Dan said "Hazlo". Uniswap's sells were excluded because their router takes tokens through Permit2 rather than a plain ERC20 allowance. Wiring it took three pieces and each one has a trap:
- The signer learned EIP-712.
eth_account.encode_typed_datawas already installed; nothing had ever asked it to sign a message rather than a transaction. - The build sends the signature with the quote, as
{"quote", "permitData", "signature"}. - The ALLOWANCE GOES TO PERMIT2, NOT TO THE ROUTER. This is the part a reasonable reader gets wrong, because every other venue wants the router.
What a wrong guess costs, measured. Approving the router: the approve confirmed on chain, the build returned clean calldata, and eth_estimateGas reverted with 0xd81b2f2e. Gas spent, nothing bought. With the permit signed and the allowance pointed at Permit2: eth_call returns without a revert, estimate 489,909 gas, and a real 2 percent FROGE sell CONFIRMED in block 61106546 at 6.2 percent above 1inch.
The rule, and it is about safety not plumbing. A transaction moves a named amount once. A Permit2 signature grants a SPENDER the right to move tokens later, so a wrong spender, token or chainId is a standing permission rather than a single loss. venues.uni_build checks all three against the quote they arrived with before signing, and signer.sign_typed deliberately knows nothing about what it signs. Never sign a typed message whose identity fields you have not compared to the trade in front of you.
And validate a patch BEFORE writing it. My first attempt at this change left a duplicated loop header, and the script wrote the file and THEN ran ast.parse, so it reported the error after breaking the money path. The order is parse, then write.
24. On this chain a sale pays in native ETH, and native ETH leaves no log
2026-09-13. The 90-day ranking put 0x1a2919… first at +6.01 ETH realized. Dan looked at the wallet and saw $300. The Opus recheck read it from the chain: about −0.018 ETH, −0.142 after gas, nine days old, funded with ~0.277 ETH, half of it gone. Three faults stacked:
- Native-ETH proceeds emit no Transfer. Uniswap v4 settles the money side in the chain's own coin. A Transfer-only reader books every BUY (WETH out) and misses most SELLs (ETH in). And there is no archive node here:
eth_getBalancereaches ~3,700 blocks back, nodebug_trace. Income has to be read from the router's own event, or by difference. - Three timed-out arrival windows made 242 tokens look sold-without-buying and faked +3.88 ETH. A refused span is a hole, never a zero.
- The reader priced one leg of a split route as the whole cost. Fixed the same week for take-profit; the ranking reader had not been told.
And the number that decides everything is not the wallet. The routers here charge about 1.010 percent per side (median of 479 fee events). Wallet 1 earned +0.5535 ETH gross and paid ~0.5715 in tolls. The median return per closed round trip across the top eight runs −2 to −25 percent; all of them live on a few big trips. Toll both ways plus a copier's later entry is larger than the median edge of any of them. Copying on this chain works only at nearly the trader's own price, so the first thing to measure is how many blocks late we enter, not which wallet.
The rule. A ranking is graded by an independent reader before anyone trades on it; agreement to the decimal on the PAID side (28.443 ETH, both readers) is what earns trust, and a gap on the RECEIVED side names the exclusion. Pass 2 now says in its meta what it leaves out. The two readers converged on the same top three.
25. The router's settlement event is what the wallet got; the pool's number is the gross
2026-09-13, from the Fable audit, verified on chain. On a sell the pool pays the router the gross, the router keeps 1.000 percent (fee event, topic 0x205442d6) and forwards the rest as native ETH (settlement event, topic 0x8619026a, amountOut), which emits no Transfer. WDOG 0xb3a3e109…: pool 0.043971, fee 0.000440, wallet 0.043531. Pass 2 booked 0.043971 on every native-paid sell and its meta did not say so; the audit graded that as an honesty fault, correctly. Hook routes charge 3 to 4 percent; relay routes 2 in tokens on the buy and 3 on the sell. Two percent per round trip is the floor, not the toll.
Where BOTH independent readers were wrong, and the rule each one makes:
- Neither subtracted open bags a trader never sold. For a copier that is a −100 percent trip. A wallet's realized net is not a copier's expected net.
- Both simulated our exit at HIS price. Exit delay is unmodeled everywhere.
- Both weighted by his sizes, hiding that his big trades do worse than his small ones (wallet 8: ladder +0.46 percent, flat +1.9 percent). A ladder keyed on the trader's size assumes his big bets are his best; measure that before assuming it.
- On ERC-4337 bundles one reader took other users' legs; the other called the wallet unmeasurable. The wallet's token leg must be matched to its own swap inside the bundle. Nobody did it yet.
The shortlist rule. A live wallet positive at a 2 percent worse entry, judged after removing its best trip, on a flat stake, in paper for 30 days. Every "GANADORA" so far rests on one trip; the bar is being positive without it.
26. A simulation has no gas cap; a transaction does
2026-09-13, selling FROGE on Dan's "Vende". Four refusals and one failed transaction before the sale went through, each with its own cause:
- Kyberswap's calldata reverted at
eth_callwithTransferHelper: TRANSFER_FROM_FAILEDandroute_execREFUSED the whole trade, although Uniswap sat one row down in the same ranking, buildable. The simulation ran once, after the candidate loop. A build that reverts is not a build: simulate per candidate and fall through. - At 100 percent the built
amountInexceeded the balance by dust, so the pull failed in simulation on every venue while a baretransferFromof the full balance passed. Sell 99 percent; the dust is the price of not guessing a router's rounding. - At 95 percent the simulation passed and the REAL transaction reverted with 866,472 gas used against a fixed 900,000 limit;
eth_estimateGasfor the same calldata says 910,314.eth_callhas no such cap. A fixed gas limit is a hidden refusal that costs gas instead of saving it; size it from the estimate, with a floor and a ceiling. - The patch for 3 was anchored on a string that did not exist and reported "anchor missing" twice; the sale succeeded on the OLD limit only because 99 percent routed cheaper than 95 had. Grep the constant before writing the patch, and never let a gate run the money step when the patch above it failed.
The sale: 99 percent via Kyberswap, tx 0x826632ee…, $63.72, 1,566 FROGE of dust left. Net on the position: about −$54 on $200.52, saved from worse by the 2x trim that took $73 out at 2.52x.
27. A rule that changes its unit must find every caller, or the old caller becomes a trap
2026-09-13. The $2,000 cap became PER TRADER on 09-10 (Dan: "no un tope total"). Gate 10 got the target. The post-fill probe in execute.py did not: it still called cap_check with no target, summed the open money of every trader ($2,310) against the per-trader cap ($2,000), and declared a BREACH one second after a confirmed $500 PARE copy. Buys halted from 16:34; nobody was told; the copier looked healthy. With no position open, breach() would have DELETED ARMED and disarmed the copier outright.
The rule. When a rule's unit changes, grep for every call site the same hour and pin a test that runs the old shape against the new one. A false breach must never be able to delete the authorisation file: a spending cap and the permission to sell are different things, and the second must survive the first being wrong. Pinned by test_post_fill_probe_is_per_trader_not_global.
28. The total cap is gone; the per-order cap and the burst rule are the limit
2026-09-14, Dan's word: "El tope por trader de 2000$ quítalo, eso es para cuando lanza muchas transacciones del mismo token y ya para eso tenemos el tope de 500$." What bounds the money now: $500 per order; one entry per burst topped up to the rung, never one per transaction of his; limits.py's loss cap on the chain pot. The 2,000 had cost two full days in two different readings (a lifetime total that halted at $546; a per-trader total that a post-fill probe summed across traders and halted every buy). cap_total_usd stays in ARMED and on the pages as a number to SHOW; nothing reads it as a gate, and the rules page says so on the row. Pinned: test_case8_total_past_20_is_refused_with_gas_counted now asserts the total does NOT refuse and the per-order cap still does.
29. A public tape finds the trader; the chain decides whether he can be copied
2026-09-14. rhtrenches.com is a public, tokenless tape of FOMO trades. Measured: its ts IS the block timestamp (3 of 3, diff 0 s), the row arrives ~0.3 s AFTER the chain, never before, and the same token moves 5 to 10 percent in the ten seconds after a fill. So the tape cannot be traded from. It can be MINED: 147 traders with full closed-trip histories that reconcile to the cent (/api/trader/<h>.history, not the board's realized_pnl, which changes sign on 14 of 147), and a free bridge handle → EVM wallet → Solana wallet.
Then the chain sorts them into two kinds, and only one is copyable:
- CUSTODIAL: the published wallets never sign (0 of 27 fills), Solana shows 0 signatures ever. The top-3 of the leaderboard were all of these.
- OWN-WALLET: signs its Solana transactions, pays Relay Bridge (
99vQwtBw…) from its own wallet, and its EVM buys arrive through the Relay receiver0xccc88a9d…WITH a swap. MoneyLord's shape; the copier already follows it at block speed.game_for_one(24 of 25 relay deliveries with a swap) andsmol_internare this kind.
The rule. Find on the tape, confirm on the chain, copy on the chain. Paper first, flat $100, 30 days, live only if positive without the best trip. Never a live target on a tape number.
30. A rule Dan has to change by hand is a rule the dashboard does not have
2026-09-15. Two $500 rungs in two days lost $482 following MoneyLord's scalps (PARE 09-13, FROGE 09-15), and the audit had already measured his big trades doing worse than his small ones. The remedy was "MoneyLord at a flat 100" and Dan asked whether that was a dashboard click. It was not: the panel edited global rungs and per-target on/off, nothing per trader about SIZE.
Built: max_ticket_usd in a target's profile block, applied in _size_buy after the ladder and the global order cap; registered in profile.defaults() so the validator's allowlist knows it and caps it at CAP_ORDER_USD (900 was refused, 100 accepted); written by panel_rules.set_target_max_ticket with an audit row; reached from the page as panel:ticket:<label> with $100 / $300 / $500 / quitar buttons beside each trader's on/off switch. Nothing set for anyone until Dan chooses.
The rule. A per-trader knob is the only honest way to say "this trader at this size" without touching the others; a global rung that fires on everyone's biggest bets is a rule about the ladder, not about the trader. And the validator's allowlist is part of the schema: a new field the validator has not been told about is a field that silently never applies.
31. I broke law 27 the day I wrote it: the grep that "found every caller" had filters
2026-09-15. Law 27 (a rule that changes its unit must find every caller) was written after the post-fill probe halted buys. The next day the $2,000 total cap was removed "in the one place that enforces it" on the strength of a grep run with grep -v test_ | grep -v .bak | grep -v "^\s*#" and scoped to two dirs. The Solana sender enforced the same cap in its own words and was not in the list. Every Solana signal from 09-14 on was refused: "$3249.50 already spent, $100 more would pass the $2000 total". MoneyLord's ALLINU was refused six times and did 5x.
The rule. When removing a gate, grep the whole tree for the FIELD NAME with no filters first, then read each hit. A filtered grep is a claim about the filters, not about the code. And after removing a gate, run every copier's own dry pass and read its refusals: a refusal string names the gate that still stands.
32. A copier without the burst rule turns one signal into five buys the moment a gate opens
2026-09-15, 20:18 UTC. The Solana sender's lifetime cap was removed and the cron re-enabled. On its first tick the copier fired five Jupiter buys of about 1.03 SOL each inside 48 seconds: all five EMBER, one per Overdose signature, 5.14 SOL for one position. The EVM copier had had "one entry per burst" since 09-10; the Solana copier keyed done on HIS signature and had no notion of a burst, so a gate that had been refusing everything hid the missing rule until it opened. Twenty-two MoneyLord PUMP signatures were queued behind them; the wallet ran dry first.
Fixed the same hour: one entry per (wallet, token) inside 30 minutes, later signatures of the same burst marked done without a buy, and at most two buys per pass so a backlog can never drain the wallet in one tick.
The rule. Before removing a gate, list what that gate was ACCIDENTALLY holding back. A refusal that fires on every signal masks every other missing rule behind it; the day it stops firing, all of them run at once. And the two copiers must carry the same laws: a rule that lives in one file is a rule the other lane does not have.
33. A staleness rule that cannot read the clock is not a rule
2026-09-15. MAX_AGE_S = 900 in the Solana copier read t/ts/blockTime as a number; the feed stamps t_confirmed and t_seen as ISO text. Age was None for every row, the check age > max_age never ran, and every signal in the file was "fresh" for ever. That is how five EMBER buys from 13:0x fired at 20:18, and how 148 signals sat armed behind a cap. Fixed: parse the ISO stamps; an unreadable age is OLD, never fresh. Dry pass after: 148 too old, 0 acted.
The rule. A guard on a value must be tested with the value's real shape from the real feed, not with the shape the author imagined. And None in a comparison silently means "pass"; the default for an unreadable safety input is REFUSE.
34. A pass reads the END of a file, never the whole file, and never two passes at once (2026-09-15)
What happened. The copier died twice in one evening. At 20:04 UTC the kernel killed it: the service has a 1 GiB memory cap and the process was at 996 MB. At 21:12 it exited with code 1 and systemd brought it back. Between the two, and after each restart, the feed reconnected and the 3-second poll covered the gap, so no MoneyLord buy was missed (FROGE 17:39 was copied in 10 seconds). But a copier that is dead for 10 seconds at the wrong moment is a copier that misses.
The cause. Every 5 minutes census.enrich() read ALL of census-enriched.ndjson (714 MB, json-parsed line by line into a set) and ALL of census.ndjson (169 MB, readlines()) to work on the newest 600 rows. And the watchdog started a new pass every 5 minutes whether or not the last one had finished, so slow RPC meant two or three copies of that in memory at once. The docstring even said "a job that tries to catch up in one pass is the thing that takes the box down". It was talking about the write side. The read side did it.
The fix. _tail_lines(path, n, max_bytes) reads at most 32 MB from the end of a file; the dedup set is the newest 20,000 enriched rows (DONE_WINDOW), never the file; a pass is skipped with a log line when the previous one is still alive (_enrich_thread). Tests: test_census.py (16, incl. a 30,000-row fixture that must cost exactly one receipt). Backups: /root/*.bak-oom-20260915.
The rule. A periodic job on a live-money daemon reads a bounded number of bytes, holds a bounded number of rows, and refuses to overlap itself. The memory cap of the service is a number the job's author must know and name. Log noise counts too: 175,000 lines of "HTTP 403" from a flapping feed (the feed host's own fault, the poll covers it) is what hid two deaths in one log.
Addendum, same evening (law 34b): a flag is not a stop. After the census fix the copier still grew 8 threads a minute. The watchdog rebuilds the feed reader after 15 s of silence, and FeedWatcher.stop() only set a flag that the thread read once per feed message. In a 403 storm no message comes, so every rebuild abandoned a live thread with its own event loop, executor thread and sockets. Fix: stop() cancels the session task on its loop from outside (loop.call_soon_threadsafe(task.cancel)), start() refuses a second thread while one is alive, and the session checks the flag before it begins. Tests: test_feed.py (37): a parked thread dies within 3 s of stop(); start twice is one thread; a stopped watcher can start again. Backups: /root/*.bak-leak-20260915. Rule: anything that "stops" a thread must be proven by the thread being dead, never by a flag being set.
35. A trip needs a real cost, and a sale only counts against what was bought (2026-09-16)
What happened. The Solana finder passed wallet 7eYa7HsAD9QNjePwdJ6spmGy5ajWbxXDxYTjor1KaeeE with a median of +12,125 percent on 11 trips and $24,117 realized. The audit found the reader's fault, not a sniper: the wallet held bags from before the window, made 4-cent buys, then sold the whole bag, and the reader attributed the whole sale to the 4-cent buy. 10 of its 11 trips cost under 1 USD. All 11 sold 95,191x to 40,044,174x more tokens than they bought inside the window. Neither passing Solana wallet is a sniper: first buys 15 to 40 minutes after birth, never in the first 10 buyers, and the price 15 s after their fill is 0.09 to 0.14 percent above it, so the bar's 2 percent delay floor is conservative.
The fix, in two places. The reader (the lane's port of ASYMMETRY section 3): proceeds attribute only to the quantity bought inside the window; a sale over 1.05x the bought quantity is flagged sold_over_bought and its excess is not a trip. THE BAR (bar.py): a trip with a known cost under DUST_USD = 1 is counted as dust and dropped before any clause; mostly dust fails with the reason DUST. Self-test in the patch: the 7eYa shape fails on DUST, a normal book passes, an unknown cost (0) is not dust, and the same returns at a real cost pass (the control). Store fields for every future hunter: snipe_flag, first_buy_delay_s, replay_delay_pct_15s, dust_trips, audit_note.
The rule. A number that is too good is a reader bug until proven otherwise. Every finder measures the entry it can actually replay (price after his fill, not his fill), and a trip whose cost the reader cannot vouch for is not evidence.
Addendum to law 35 (same night): no verdict on a fraction of a book. The sweep's judge built trips from positions whose sells were priced and whose buys mostly were not (0xc87b51c7: 19 of 2,729 round trips proven read as a +4,818 percent median). Two published verdicts were retracted with their dates (a retraction must clear judged_at, or the never-redo rule hides the hole for ever). THE BAR now takes proven_share; under MIN_PROVEN_SHARE = 0.90 there is no verdict, reason PARTLY PRICED. Every reader reports round_trips, round_trips_proven and proven_share.
36. Nobody copies another lane's directory, in any direction (2026-09-16)
What happened. The store owner deployed a fix to build/wallets/tape/store.py, verified it, and minutes later found BOTH the box and the Mac back at an older checksum. Not a second writer on the box: the Solana lane had pulled store.py, trips.py and SCHEMA.md box -> Mac at 03:49:53Z "to refresh a read-only copy", which overwrote the owner's Mac working copy with the box's then-current file. The box itself never regressed (the owner's dated backups prove it); the Mac lost the fix and the owner's local test found it missing. The weakness was real anyway: editing on the Mac and pushing to the box makes any external write to the Mac the next deploy. The sweep lane had done one box -> Mac pull of the same file at 03:21Z and left a backup inside the owner's directory. Three lanes, one directory, three copies in flight.
The rule. A lane copies NOTHING under another lane's directory, in either direction, ever: it imports from the live path (sys.path), and a test asserts its own directory holds no copy of the other lane's files. The owner edits and deploys from ONE source (the box), verifies the md5 on the side that will RUN the code at the moment it runs, mirrors box -> Mac last, and publishes the md5 of record to every lane after each deploy and in map/wallets/store/MD5-OF-RECORD.txt on the box. A 2-minute checksum watch on the box (tools/store_md5_watch.sh, log wallets-store-watch.log) times every change. Sister rules: 34b (a flag is not a stop) and the sweep's law 13 (the md5 that matters is on the side that runs).
37. The live box is not a test bench: no big fixtures there, no locks in /tmp (2026-09-16)
What happened. The store owner's scale tests built 50,000-row fixtures of ~120 MB each in /tmp on the copier's box; a test that raised before its cleanup stranded the directory. Eighteen of them filled the 1.9 GB tmpfs to 100 percent between about 03:30 and 04:20 UTC. Six money crons took their flock in /tmp (take-profit, reconcile, routing-audit and three tabs). This time every lock file already existed, so take_profit ran every five minutes through the window (checked in its log) and nothing failed; a missing lock file on a full tmpfs would have stopped a cron in silence.
The rule. Scale fixtures run on the Mac, never on the box; the box runs the small hermetic suite. A test that must write on the box writes under /root/signal-desk/.scratch/ with a hard cap, atexit cleanup and a pre-run sweep. Cron locks live in /root/signal-desk/.locks/ (moved 2026-09-16, crontab backup /root/crontab.bak-locks-20260916). A test runner prints free space at start and end.
38. A repair to a file a live process holds in memory is a race, not a repair (2026-09-16)
What happened. The sweep found it had been re-reading covered blocks (17.5 percent of a run) since the 09-05 sweep, leaving 921,587 duplicate (block, log_index) entries in pool_state.json. Deduping recomputed 29,264 bases and moved 207 runner tiers (same peak, different base: 60 down, 48 up, 30 out of sized, 19 in, 18 to 100x, 5 from it). The sweep holds that file in memory and rewrites it every 100 spans, so a dedupe under the running sweep would have been clobbered at the next checkpoint in silence. The lane stopped the sweep first (which also freed the 2.6 GB the load needed), deduped in 26 s, restarted on the clean state. The swap count n cannot be recovered; 456 runner rows (6.5 percent) carry n_trusted: false and 6 hunters count_provisional: true until the clean rebuild.
The rule. Before repairing a file, name every process that holds it in memory or rewrites it on a timer, and stop them. A tier is peak over base: a rule that gates the peak's dollars does not protect the multiple when the base moves. Any ranking derived from a repaired input is re-derived after the repair, and the delta is reported.
39. A bag is worth its cost, and the book is measured in dollars when it can be (2026-09-16)
What happened. Dan: "es literalmente IMPOSIBLE que no pasen... con lebot conseguiamos candidatos siempre". He was right about one clause. THE BAR counted every open bag as a whole trip lost at a flat stake: 38 bags = 38 trips at -100 percent, whatever they cost. smol_intern made $241,720 realized in 30 days with 38 bags costing $72,969; with every bag at zero he is still +$168,751, and the count rule failed him. game_for_one the same (+$21,903 realized, 4 bags $12,087). insentos (+$9,669 realized, 32 bags $56,080) fails either way, which is the point: the rule was not too strict, it was measuring the wrong unit.
The fix. When 90 percent or more of the trips carry a known cost, the book is dollars: each trip = its cost times OUR adjusted return (toll twice, 2 percent later entry); the best trip and the bags are dollars; WITHOUT BEST and BAGS AT -100 are dollar clauses. When costs are unknown (a reader that quotes in WETH and passes 0), the flat-stake count rule stays and the verdict names its method. The Robinhood sweep now converts WETH costs at the fill's block through eth_rate.py, never at today's price.
The rule. A penalty is measured in the unit the profit is measured in. A clause that can fail a wallet with more realized profit than the cost of everything it still holds is not a rule, it is a unit error.
40. A breaker nobody chose halts nobody in silence (2026-09-16)
What happened. build/exec/limits.py carried a 250 USD total-loss cap on the chain pot, written 2026-09-04 as "the money lane's proposal, change on one word from Dan". Dan never chose it and had removed every total cap. On 2026-09-15 18:36 UTC the FROGE sale took realised losses to 447 USD, the breaker latched (live_enabled: false, "the first breach is FINAL until a human turns it on") and the copier refused every live buy for 21 hours: two 500 USD MUSEBOOK plans on 09-16 at 09:57 and 09:58 died at "gate 0 spend cap" and no card, no message, no dashboard line said so. Found only by reading the decision log after Dan asked why Musebook was not copied.
The fix. Dan's word: no total or daily loss cap on the chain pot (set to 1,000,000 because check() compares a float against the cap and a null crashes it, which would have refused every buy with a TypeError, caught by --check 100 before the daemons ran it), live back on, tickets 100 USD fixed for MoneyLord and Overdose on Robinhood, BSC and Base (75 on Arc), daemons restarted, all recorded in the file's own corrections list.
The rule. A number nobody chose is not a rule (law from 09-09 again). Any switch that can stop the copier PAGES Dan the moment it flips and shows on the Reglas tab in red until a human clears it; a halt that pages nobody is a bug of the same class as the death it prevents. A cap set to "no cap" must be a value the code can compare.
41. A rule Dan did not ask for is removed, not tuned (2026-09-16)
Four in one day: the 2 percent delay and the 1 percent toll in the bar, the 250 USD loss breaker, the count-based bags clause, and gate 4b (a token named like the chain's dollar). Dan: "no pongas reglas que no te he pedido". Each one either hid a winner, halted the copier in silence, or blocked the very token the copied wallet trades. The rule: every threshold and every gate names whose word it rests on in the code comment beside it. One with no word behind it is not a rule; when Dan objects it is deleted, with its tests, not softened. On Arc the scanner is blind: the exit rehearsal at our real size decides, as on Robinhood (Dan, 2026-09-16: "SI").
42. The single-LP-holder stop refused what MoneyLord wins on (2026-09-17)
What happened. token_safety.py refused any token whose pool had one LP holder (a GoPlus rug heuristic, MIN_LP_HOLDERS = 2, plus "one unlocked holder over 50 percent"). On Robinhood Chain every fresh v4 pool has ONE position, the deployer's, so the stop refused MUSEBOOK on 09-16 09:30 (then 5x) and CPU on 09-17 18:05 (3x within the hour, 7.55 s after his block, the round-trip rehearsal had PASSED at 100 -> 98.56), while it let FROGE and PARE through. Dan: "y si copias las que perdemos". Not his rule (law 41).
The fix. Both LP findings are FACTS on the card (lp_risk, lp_puller), never refusals; the exit rehearsal at our real size stays the hard stop. Four suite cases flipped by exact string, 88/88; every EVM daemon restarted. Audit: our 100 USD on MUSEBOOK at 09:30 would read about 220 to 500 today; on CPU about 300.
The rule. A safety heuristic is judged by what it refuses AND what it lets through on THIS chain, measured against the copied wallet's real trades, before it is a stop. A stop that refuses the target's winners and passes his losers is a losing rule.
43. A budget for watchers never touches the execution path (2026-09-17)
What happened. FLX, 18:38:13Z: plan BUY 100 USD [LIVE], then step "3 router" died: "the 1inch daily budget is spent: 9000 of 12000 calls today, and this is a watch call". The 1inch spender address is a constant of the chain read once an hour; it was classified as a WATCH call, the watchers (alerts, audits, quotes) had spent the 75 percent watch share, and the process had been restarted 20 minutes earlier so it had nothing remembered. FLX did +101 percent in the hour. Third invented stop of the day after the LP rule (law 42) and the loss breaker (law 40).
The fix. The spender read is a TRADE call by its path; the remembered address is persisted to build/exec/oneinch-spender-cache.json on every read and loaded at import, seeded for robinhood, bsc, base and arc; swap suite 164/164; daemons restarted.
The rule. Every call on the execution path is a trade call. A budget meant to starve watchers is proven against the live BUY path with a spent budget before it ships. A process restart never empties a cache the money path depends on.
🤝 Handoff
Fuente HANDOFF-2026-09-17.md · 19 KB · modificado 2026-09-17 20:14 UTC
systemctl is-active y crontab -l) · 4. El copiador: reglas que sí son de Dan (y dónde viven) · 5. El buscador de wallets, definido entero · 6. Posiciones y dinero (verificar en cadena) · 7. Velocidad medida (7 días) y el plan para la sesión nueva (auditar y ejecutar; NO lo hizo esta sesión) · 8. Lo que falló y ya NO existe (no lo vuelvas a meter) · 9. Decisiones que Dan no ha dado (no actuar sin su palabra) · 10. Trabajo abierto al cerrar · 11. Cómo trabajar aquí sin repetir los fallos · 12. Prompt para abrir la conversación nuevaHANDOFF LEBOT v2 · 2026-09-17 · para una conversación nueva
Lee esto entero antes de tocar nada. Proyecto PERSONAL de Dan y Diego (copy-trading cripto). Nunca infraestructura de Slate. Escrito por la sesión que Dan cerró por fallar demasiado; los fallos están en la sección 8 para que no se repitan.
0. El mandato de Dan (sus palabras, 2026-09-16 y 17)
- "TODAS ESTAS REGLAS TIENEN QUE ESTAR EN MI INTERFAZ."
- "No quiero que haya un fallo más nunca."
- "No pongas reglas que no te he pedido." Una regla sin cita de Dan al lado NO EXISTE. Si Dan objeta, se BORRA con sus tests, no se afina (ley 41).
- "NUNCA se copia un wallet sin nosotros autorizar." Ni en papel ni en vivo.
- Tickets: $100 fijos para MoneyLord y Overdose en Robinhood, BSC y Base; $75 fijos en Arc; tope por orden $500; sin tope total, sin tope de pérdida.
- Ruta: uniswap > kyberswap > 1inch, 1inch solo si es mejor. Mismo router para aprobar y para vender.
- 2x desde NUESTRA última compra (recorte 30%). Las ventas manuales de Dan van al cuaderno.
- Velocidad: "la rapidez no viene por gas, viene por detección y ejecución".
- Reportes: cortos, en español, sin rayas largas, "como si tuviera 5 años"; dinero y estado verificados en cadena, nunca de memoria.
1. Accesos (dónde está todo)
| Qué | Dónde | Notas |
|---|---|---|
| VPS del proyecto | ssh signal-desk = root@5.223.82.12, llave ~/.ssh/signal_desk (en ~/.ssh/config) | repo /root/signal-desk, python /root/signal-desk/.venv/bin/python. No hay git en la caja. |
| PROHIBIDO | ssh slate | servidor del negocio Slate. Un agente lo tocó por error el 09-15. Nunca. |
| Interfaz (hub) | https://portfolio.178.104.79.3.nip.io | vive en la "data box" 178.104.79.3; se empuja desde el VPS cada 5 min (tools/portfolio_html.py --push, cron). Login básico: usuario y clave en /root/.config/crypto-archive/credentials.env del VPS, nombres PORTFOLIO_USER / PORTFOLIO_PASSWORD (nunca imprimir valores). Login de página: HUB_SECRET_DAN. Pestañas: 🤖 LEBOT (posiciones), 🎛️ Reglas (interruptores), 🔎 Carteras (buscador), 🏆 Ranked, 👛 Wallets. |
| Telegram | tarjetas del copiador y alertas: build/alert-card/; chats por nombre en credentials.env (TELEGRAM_CHAT_*) | el copiador Robinhood manda tarjetas; los de bsc/base/arc corren con --no-card |
| Wallet EVM | 0xc6D970893e2458ff5A654d91a62cEaCb8EfDa4d3 (todas las cadenas EVM) | llave en el VPS, usada por build/exec/signer.py |
| Wallet Solana | 3iP5iL1pK81yXxVhqfk82MjPSEQ34Qek6EN6Pef3DEHS | ~4.95 SOL |
| Puente | Relay (api.relay.link), sirve Robinhood 4663 y Arc 5042 | 0.1 ETH llegó a Arc como 237 USDC en segundos, $0.77 de comisión |
| Secretos | solo por nombre; /root/.config/crypto-archive/credentials.env | nunca en chat, nunca en memoria |
2. Código y documentación (el mapa)
- La verdad viva es el VPS (
/root/signal-desk).~/bot-archiveen el Mac es un ESPEJO: cada lane copia sus archivos allí al terminar. El espejo tiene 409 archivos sin commit y ningún remote de GitHub. - GitHub, estado real (09-17): existe
~/signal-desk-repoen el Mac con remotehttps://github.com/pykleryk/signal-desk.git(export limpio del 09-02, commit d708717, sin chats ni secretos), pero no consta empujado y el repo puede no existir. Elghdel Mac está logueado comoslatesystemsllc-dot(cuenta de Slate). Tarea para la sesión nueva (sección 10): dejar UN repo privado en la cuenta personal de Dan, empujar código + canon (sin secretos, sin cuadernos), y fijar la dirección: VPS -> repo -> espejo. - Entradas de lectura en
~/bot-archive:START-HERE.md,LOOKUP.md(tabla generada),map/INDEX.md(la historia 2018-2024 y los índices),map/PENDING-MASTER.md,canonical/00-EVERYTHING.md,canonical/logic/(LEBOT-V2-LOGIC.md, THE-RULES.md, EXECUTORS.md, LATENCY.md, INFRA-MAP.md, HAZARDS.md),map/wallets/BOARD.md(bitácora del barrido). - Leyes:
~/bot-archive/canonical/logic/THE-RULES.md, 43 leyes. Leer 34 a 43 (las de esta semana). - Informes de esta semana:
build/lebot/ARC-2026-09-16.md,build/lebot/TAPE-ROLL-2026-09-15.md,build/lebot/SOLANA-RUNNERS-2026-09-15.md,map/wallets/SEARCH-67D-2026-09-15.md,build/lebot/GATES-2026-09-17.md(del agente Fable; si no existe, es la primera tarea),build/lebot/AUDIT-DASHBOARD-2026-09-12.md. - Logs (VPS):
/var/log/signal-lebot.log(Robinhood),-arc.log,-bsc.log,-base.log;/var/log/take-profit-cron.logytake-profit-arc-cron.log;/var/log/wallet-tape.log,wallet-tape-skips.log,wallet-solana.log;/root/signal-desk/solana-be-exit.log. Trampa:/root/signal-desk/take-profit-cron.logestá muerto desde 09-10 (dos agentes lo leyeron como cron muerta). - Cómo leer una decisión del copiador en el log:
BUY 0x… BUY: gave nothing; got N TOKEN(detección) ->gate 6 …->plan BUY $X [LIVE]-> líneasok N …/NO N …->result BUY ok=True CONFIRMEDoresult NOTHING ok=False <puerta>: <razón>->t_decided: N s after his blockyphases: decided +a s, SENT +b s, landed +c s.
3. Sistemas vivos (verificar con systemctl is-active y crontab -l)
| Sistema | Unidad / cron | Qué hace |
|---|---|---|
| Copiador Robinhood | signal-lebot.service (build/lebot/lebot.py --paper; el flag --paper NO apaga los envíos, el vivo lo decide targets.json + limits.py) | feed del secuenciador por websocket + websocket de bloques + sondeo de logs 3 s |
| Copiador Arc | signal-lebot-arc.service | vivo desde 09-16 16:49; websocket + sondeo 6 s; escáner ciego -> el ensayo de venta decide |
| Copiador BSC / Base | signal-lebot-bsc.service, signal-lebot-base.service | websocket + sondeo 3 s |
| Copiador Solana | cron */2 build/lebot/solana_copy.py --go (candado .solana-copy.lock) | 0 a 120 s de retraso; vende solo a mano |
| Salida BE Solana | cron * * * * * solana_be_exit.py | ya disparó en EMBER (marcador be-exit-5dvXTZ5q.done); reutilizable con otro mint |
| Recorte 2x | cron */5 take_profit.py --go --chain robinhood y --chain arc | candados en /root/signal-desk/.locks/ |
| Reconcile | cron */10 reconcile.py --go | SELL_OFFBOOK para saldos cero |
| Auditoría de rutas | cron hourly routing_audit.py | lee el router del recibo, no del log |
| Pestañas | positions_tab.py (/5), rules_tab.py (/10), store_tab.py (/10), build_ranked_tab.py (/20); push del hub portfolio_html.py --push (*/5) | |
| Buscador FOMO | wallet-tape.service | cinta rhtrenches cada 2 min, registro "nunca repetir" |
| Buscador Solana | wallet-solana.service | runners cada 30 min, cazadores cada hora |
| Barrido Robinhood | proceso detached en el MAC (build/wallets/analyzer/search.py, PID en out/search/logs67/) + LaunchAgent io.slate.wallets.rollforward 04:40 | backfill de 67 días en curso; escribe en el almacén del VPS por scp + upsert |
| Almacén de wallets | map/wallets/store/wallets.ndjson (dueño build/wallets/tape/store.py), deploy.sh, MD5-OF-RECORD.txt, record_watch.sh (/10), tools/store_md5_watch.sh (/2) | 19,292 filas, 359 juzgadas |
| Ejecutor | build/exec/swap.py + signer.py + limits.py (limits-state.json: live_enabled: true, bolsa chain sin topes de pérdida = 1,000,000; un null rompe check()) | |
| Otros servicios de la caja (no tocar sin leer INFRA-MAP.md) | signal-decisions, signal-hand, signal-watchers, signal-tgtap, discord, crons de brain/x/judge/seller/shadow | otros brazos del proyecto |
4. El copiador: reglas que sí son de Dan (y dónde viven)
- Targets:
build/lebot/targets.json(moneylord-relay, overdose-evm en Robinhood; moneylord-bsc, overdose-bsc; moneylord-base, overdose-base; moneylord-arc, overdose-arc; unipcs-* apagados; paper: game_for_one-evm, smol_intern-evm). Cada fila llevaprofile(max_ticket_usd,fixed_ticket_usd).ARMED.json:targets_authorisedno lo lee ningún código (decisión pendiente). - Tamaño:
decide._size_buy: ticket fijo/tope por trader manda en TODAS las rutas (arreglado 09-16: la ruta proporcional lo ignoraba). Escalera 100/300/500 según SU ráfaga de 30 min solo si no hay ticket fijo. Una entrada por ráfaga. Tope por orden $500 (CAP_ORDER_USD). - Puertas de decisión (
decide.py): 1 suya y confirmada, 2 no vista, 3 forma accionable, 4 negociable (deny.json), 5 compra real, 6 seguridad (escáner GoPlus + ensayo de venta al tamaño real; en Robinhood y Arc el escáner está ciego y el ensayo decide), 7 cooldown (0), 8 compras seguidas (máx 10, de Dan), 8b él aún tiene ≥50%, 10 tope LEBOT, 11 tamaño. - Ejecutor (
swap.run): 0 tope de gasto (limits.py), 0b wallet, 1 saldo, 2 cotización de venta a este tamaño, 3 router (caché en discooneinch-spender-cache.json), 4-8 cotización tres rutas, construir, simular, firmar, 9 broadcast, 10 pre-approve al router de venta. - Salidas: espejo de sus ventas por fracción (
plan SELL 18%= él vendió 18%); recorte 30% a 2x desde nuestra última compra;reconcilemarca saldos cero. - Hoy el copiador no tiene ninguna regla que Dan no pidiera capaz de frenar una compra: quedan dinero, gas, y "no se puede vender al tamaño real". El registro completo con interruptores lo entrega el agente Fable (sección 10).
5. El buscador de wallets, definido entero
Objetivo (Dan): "un sistema que busca constantemente, con los datos ahí para que yo filtre", "gemas", "nunca rehacer lo ya analizado".
Tres fuentes, un almacén, una barra:
- Cinta FOMO (
build/wallets/tape/):rhtrenches.com/api/tapey/api/closed(400 filas, se acumulan),/api/trader/<handle>(historial cerrado; 1 llamada cada 9 s obedeciendo Retry-After; surealized_pnldel tablero está mal, solohistory[]), 87% de dólares de la cinta son compras falsas marcadas por el sitio (se excluyen). Wallets publicadas de FOMO son custodiadas; la real se recupera por método (hunt-354). - Robinhood en cadena (
build/wallets/analyzer/, corre en el Mac): pools nacidos, runners (≥3x desde la base de la hora de nacimiento, pico con tamaño ≥$20,n_trusted), compradores antes del primer 2x, cazadores (3+ runners), juzgados con el lectorbuild/wallets/reader/wallet_history.py(precio desde el lado de la wallet, recibos por lotes de 60). Doc:ASYMMETRY.md§2-3,COVERAGE.md(las 49 métricas del tool viejo de Dan),BOARD.md§2b. - Solana (
build/wallets/solana/): pump.fun + DexScreener + Jupiter, runners 3x, compradores tempranos, cazadores; el lector aplica la ley anti-trampa (una venta solo cuenta contra lo comprado en la ventana; >1.05x = coste no visto). Límites medidos enSOURCES-MEASURED.md.
La barra (build/wallets/bar.py, una sola para las tres; cada cláusula lleva su número y su origen):
- VIVA: viaje cerrado o abierto en 48 h.
- VIAJES: ≥10 cerrados.
- MEDIANA > 0 del retorno por viaje (la media engaña).
- PEAJE y RETRASO: 0 y 0 por orden de Dan (09-16); las cláusulas quedan inertes.
- SIN EL MEJOR: el libro sigue positivo sin su mejor viaje.
- BOLSAS: cada bolsa abierta cuenta como perdida a su coste en dólares; si ≥90% de los viajes tienen coste, todo el libro es en dólares; si no, método plano y lo dice.
- POLVO: viaje comprado por < $1 no cuenta (
DUST_USD). - SIN VEREDICTO (readable=False, la fila vuelve a la cola sin fecha): LIBRO A MEDIAS (< 90% de viajes con precio), COSTE NO VISTO (vendió más de lo comprado), BOLSAS SIN COSTE LEÍDO.
- "Pasa la barra" = merece ojos, NO "copiar". Copiar requiere la palabra de Dan.
El almacén (build/wallets/tape/SCHEMA.md es el contrato): clave evm:<0x> / sol:<pubkey> / handle:; campos de registro (status visto/analizado/actividad nueva lo calcula el almacén, judged_at, judge_source, history_hash, last_activity_ts), campos de auditoría (snipe_flag, first_buy_delay_s, replay_delay_pct_15s, dust_trips, sold_over_bought_trips, proven_share, coverage_note), read_error = SIN veredicto (una fila dice una sola cosa), retractación = un hecho y una escritura (retract(ident, by, why), solo el dueño del veredicto), campo desconocido = descartado con aviso. Nunca rehacer: se relee una wallet solo con actividad más nueva que su análisis.
Disciplina de despliegue (leyes 36 a 38): un dueño por archivo y carpeta; nadie copia la carpeta de otro lane en ninguna dirección; se despliega en el VPS con deploy.sh (prueba, escribe MD5-OF-RECORD.txt, reinicia, refleja al Mac, aborta si no coinciden); record_watch.sh cada 10 min; fixtures grandes solo en el Mac; candados en .locks/, nunca /tmp.
Estado (09-17): 19,292 wallets, 359 juzgadas. Pasan: smol_intern (FOMO, +$243k/30d, bolsas $75k), 0xcb4d28c2… (Robinhood, +$41.7k, pasa solo porque el peaje es 0), game_for_one (FOMO, +$21.9k, bolsas $9.7k). Ninguna copiada. Pendiente: papel para las dos FOMO si Dan lo dice.
6. Posiciones y dinero (verificar en cadena)
- UpSideDownCat "USDC" (Arc,
0x8e98a62a995a50eca9979bfa016f91bf36a8f9d9): 72,972 tokens en dos lotes ($100 de Dan a $0.003618 + copia $100 a $0.002208). Wallet Arc ~41 USDC. - MUSEBOOK (Robinhood,
0x91a2dae9699f0b82540b5886b0d8759c22820ba3): dos copias de $100 (09-17 10:47 y 19:00). - Cerradas: ASTEROID (ML salió 100% el 09-17; el espejo vendió todo, $685 por un lote de $1,643; cierre ~-$740 incluido el recorte anterior), EMBER (+$38), FROGE x2 (-$486), PARE (-$239).
- MoneyLord: mediana -17% por viaje; 8 de sus últimas 11 pierden. Copiarlo con espejo de salidas rinde lo que él rinde.
7. Velocidad medida (7 días) y el plan para la sesión nueva (auditar y ejecutar; NO lo hizo esta sesión)
- Robinhood: decisión mediana 2.5 s tras su bloque (573 decisiones); compra confirmada +7 a +18 s; 3 a 6 s son el ejecutor (3 cotizaciones + simulación + firma), ~2.3 s las 5 lecturas de decisión, 2 a 3 s aterrizar. Feed del secuenciador: 403 del 09-11 al 09-17 20:06 (el sondeo de 3 s cubrió).
- BSC: +9 a +18 s. Arc: +12.6 s. Solana: 0 a 120 s (cron de 2 min).
- Plan propuesto (para auditar primero, luego ejecutar): 1. Solana: suscripción websocket a las wallets de ML/OD (
logsSubscribe/accountSubscribedel RPC) en vez del cron; meta 1 a 3 s. 2. Ejecutor: cotizar solo la ruta que puede tener el token (Robinhood: Uniswap v4 primero; Kyber solo si Uniswap no cotiza); reutilizar escáner y ensayo dentro de la misma ráfaga; meta 1 a 2 s menos. 3. Medir cada copia en la pestaña LEBOT: "segundos después de su bloque" por fase. 4. Lo que no se compra sin gas: los bots en el mismo bloque. Meta realista: 4 a 6 s Robinhood, 2 a 4 s Solana.
8. Lo que falló y ya NO existe (no lo vuelvas a meter)
| Regla inventada | Qué costó | Estado |
|---|---|---|
Tope de pérdida $250/$150 (limits.py) | 21 h sin comprar; MUSEBOOK 09-16 09:57 | fuera (ley 40) |
Un solo tenedor de LP / tenedor >50% (token_safety.py) | MUSEBOOK 5x, MONEY +86%, CPU 6.5x | HECHO en la tarjeta, nunca freno (ley 42) |
Presupuesto 1inch en el paso "router" (swap.py) | FLX +101% | lectura = operación; caché en disco (ley 43) |
| Peaje 1% + retraso 2% en la barra | tumbaba ganadoras | 0 (Dan) |
| Bolsas por unidad en la barra | tumbaba smol_intern | en dólares al coste (ley 39) |
| Gate 4b "token que imita el dólar" | bloqueaba lo que ML opera en Arc | borrada |
| Tope total $2,000 y tope total Solana | ALLINU 5x | fuera |
| Ticket fijo ignorado en la ruta proporcional | copia Arc a $100 en vez de $75 | arreglado con test que corre decide() |
| Daemon muerto por memoria (census leía 714 MB) y fuga de hilos (feed) | dos muertes 09-15 | arreglado (ley 34) |
/tmp lleno por fixtures de prueba; candados de crons en /tmp | riesgo de cron muda | candados en .locks/, fixtures en el Mac (ley 37) |
Dinero no ganado por reglas inventadas (precios 09-17): ~$1,220 a $100 por ticket; ~$5,400 con los tamaños de entonces.
9. Decisiones que Dan no ha dado (no actuar sin su palabra)
- Papel para
smol_intern/game_for_one. - Fondear más Arc (quedan ~41 USDC).
ARMED.jsontargets_authorised: hacerla real o borrarla. Pestaña vieja 📜 Rules: retirar o no.- Go para el plan de velocidad (sección 7).
- Entradas manuales tardías (CPU a 6.5x, FLX): esta sesión votó no.
10. Trabajo abierto al cerrar
- Agente Fable
gates-fable(en marcha o terminado; comprobarbuild/lebot/GATES-2026-09-17.md): registrobuild/lebot/gates.jsoncon TODAS las puertas (id, dónde, qué, params,dan_wordo "SIN PALABRA DE DAN"), interruptores y números en Reglas (sección PUERTAS), alarma instantánea por Telegram al rechazar una compra de un target (salvo dinero, gas, invendible), sección "COMPRAS NO HECHAS" (razón, regla, estado hoy, precio desde su compra) rellenada desde el 09-10, contadores cada 10 min, tests, reinicios. Si no está: es la primera tarea, con el brief de esta sección. - GitHub ordenado (sección 2): un repo privado personal, código + canon, sin secretos, VPS -> repo -> espejo.
- Barrido Robinhood: al terminar el backfill, reconstrucción limpia (25 h) + pasada de compradores (33 h) + marcado de "viva"; juzgar los 7 entrantes del top 90; mover el barrido al VPS (545 MB de estado, un solo host de logs).
- Lista de Fable "SIN PALABRA DE DAN": entregar a Dan para que mate o deje, una palabra por línea.
11. Cómo trabajar aquí sin repetir los fallos
- Verificar en la caja o en la cadena antes de decir "hecho". Nunca desde un resumen.
- Cada umbral nuevo lleva la cita de Dan al lado o no entra. Un freno se prueba contra las compras reales de ML antes de existir.
- Un dueño por archivo; desplegar en la caja, espejo después, md5 publicado; nunca copiar la carpeta de otro lane; parches por scp, nunca heredocs por ssh.
- Pocos agentes, brief completo, que terminen; sin monitores que los despierten (gastan la cuota de Dan).
- Reinicios de daemons de uno en uno con 10 s entre ellos;
wallet-solanatarda 30 s en asentarse tras un reinicio. - Reportes en español, cortos, con tabla; Dan decide con 2 o 3 opciones y un voto.
12. Prompt para abrir la conversación nueva
"Lee ~/bot-archive/HANDOFF-2026-09-17.md entero y ~/bot-archive/canonical/logic/THE-RULES.md (leyes 34 a 43). En ssh signal-desk verifica que los servicios de la sección 3 están activos y si existe build/lebot/GATES-2026-09-17.md. Luego dime en 10 líneas: qué está vivo, qué está roto, qué decisión me toca, y tu plan para la velocidad (sección 7) antes de ejecutarlo."
Generado 2026-09-17 20:57 UTC por build/lebot/wiki_tab.py desde la raíz del repo en la caja. Secciones: atlas 24, glosario 12, leyes 43, handoff 14. Si un documento cambia en el Mac, hay que copiarlo a la caja para que aparezca aquí.