Run recordsreconciled · state/events.jsonl · not curated
started (UTC)
routine
trigger
code
rulebook
model
outcome
act / ref / inc
Fund Brainknowledge base · idea → order → fill → exit → lesson
Every figure on this page is drawn from the fund's record: research entries, the rulebook, intents and run records, broker fills, EOD snapshots and weekly reviews.
The Fund Brain is the knowledge base that links that record per trade and per thesis so the routines read one brief instead of log tails.
Open the Fund Brain →
Provenance. All figures verbatim from origin/main of the TRADING-ROUTINE repo, latest log — always read the remote; the cloud crons push straight to main and local clones go stale within days. Equity, cash %, day P&L, positions, stops: memory/TRADE-LOG.md. Weekly stats: memory/WEEKLY-REVIEW.md. Decisions: memory/RESEARCH-LOG.md. Rules: CLAUDE.md. Everything above is regenerated from those files by scripts/build_dashboard_data.py, which hard-fails rather than publish partial or stale data; the panels marked editorial are hand-written and dated.
Known data limits — stated, not hidden. (1) Benchmark: the logged S&P series re-fetched its starting level in 3 weeks instead of chaining from the prior close, so it was discontinuous. The chart plots the re-chained series (spxc), marked *; weeks marked † logged no index levels and are carried as-is. The correction makes the S&P stronger (+5.12% vs +3.26% as logged) and the bot's relative record worse. Full detail in the Benchmark Data Errata section of WEEKLY-REVIEW.md. (2) Blotter: Apr–May exits (AMD, PLTR, MU, AVGO) predate the weekly-review era and were never logged; those rows say so rather than guess — which is why logged realized P&L is negative while the account is up. (3) Positions and stops were reconciled 1:1 against the live Alpaca API on 2026-08-11: share counts, entries and stop levels matched the log exactly. Paper Alpaca account — no real capital at risk.