Complete guide for human-AI collaboration on weekly Friday bench review:
- Part 1: Full wallet screening chain (3 layers: watch_sharps -> backtest -> live)
with copy_pnl vs forward realized explanation (oliman2/leegunner lesson)
- Part 2: Weekly Friday review process (5 steps with exact commands)
- Step 1: Human pulls 6 data sets from server (exact commands)
- Step 2: AI analyzes 5 dimensions (forward P&L, copy_pnl, conv_win, seasonal, floor)
- Step 3: AI outputs structured recommendation
- Step 4: Human makes final decision
- Step 5: Human executes on server (promote/demote/pin commands)
- Part 3: Log analysis guide (how to discover wallet niches, floor drift,
CASH!=CHAIN, RTDS issues from logs)
- Part 4: Data checklist (8 items human must provide from cloud)
- Part 5: AI agent work rules (input format, analysis rules, output format)
- Appendix: Current status snapshot (8 live + 7 bench with forward P&L)
Complete guide for syncing original author's code while preserving our
customizations. Covers:
- When to sync (monthly or when author publishes major updates)
- What's ours vs author's (table of files + handling)
- Why we discard our patches (author's fixes are better)
- 5-step sync process (sync-upstream.sh -> server update -> bankroll ->
wallet config -> verify)
- Post-sync checklist (12 items)
- sync-upstream.sh OUR_FILES list (must update when adding new custom files)
- FAQ (7 common post-sync issues)
- File relationship diagram (GitHub -> Gitea -> server)
sync_floors.py only updates copybot.paper.json (paper bot). The live
bot's config.live.json floors were never auto-updated (HANDOFF.md:
'nothing auto-writes it'), so manually-added wallets kept their
floor forever.
New sync_live_floors.py mirrors paper floors into config.live.json:
- Reads floors (and floor_pin overrides) from copybot.paper.json
- Updates matching wallets in config.live.json
- Only writes if something changed
Added to daily.sh right after sync_floors.py, so both configs stay
in sync every day.
CASH≠CHAIN -$1.75 root-caused: the 07-16 BikesAreTheBikes copy booked with
empty title/outcome/cond, so settle_resolved could never resolve it; the
venue auto-redeemed $1.694914 on-chain (tx 0x4f303e0e…) while the book
held it open. Hand-settled exactly as settle_resolved would (won @ 1.0,
pnl +$0.68, my_pos/conds cleaned, metadata backfilled from the venue
record) and folded the +$0.050314 fill-rounding remainder as an
adjustments entry (non-trading income, zeroes CASH≠CHAIN, cash credited).
Post: cash == chain $49.324270 (3 RPCs agree), raw drift +0.0003 (0.00).
Machine stopped + heartbeats watched to cessation per the HANDOFF ops
procedure; root cause filed as #18.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
FINDINGS: 'The tape era opens' section (sharp screen 742/742 proxy
validation, Study A identity-null, Study B winner's curse, crater refill
clock) + scorecard rows for the two forward-window studies + research/ in
repo layout. HANDOFF: #16/#17 in the open queue, research silo + nightly
launchd in snapshot/ops. README: research/ directory row.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The flat 10s retry was calibrated to nothing; the tape now says craters
refill at very different speeds (research requote_timing, 775k crater
prints): crypto 94% within 4s, esports 83% within 10s, sports only 70% at
10s but 76% by 25s, geo/politics tails run minutes (price guard is the
protection there, not the retry). Waits move to FAK_RETRY_NICHE_DEFAULT
{crypto 4, esports 10, sports/geo/politics/other 25}, classified by the
validated research niche patterns (esports before sports so 'LoL: A vs B'
doesn't fall through on ' vs '). cfg fak_retry_niche_s overrides per key;
fak_retry_s stays the fallback (tennis: no measurement yet) and 0 still
disables the feature. Paper-parity: same map both bots. test t8 covers the
map, first-match classing, fallback, and config override.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
live/tape_sharps.py — insider.py's improbability z transplanted onto the
3-day tape (13.8M fills, ALL wallets, no survivorship bias): proxy-resolve
tokens from terminal-VWAP convergence (0.97/0.03, 2h quiet, sibling veto),
net held positions per wallet-token, z = (W - Σp)/sqrt(Σp(1-p)) over
proxy-resolved held bets, then payouts.py chain-truth overlay on the
shortlist (refunds count as neither). Copyable/algo split: discrete-entry
conviction bettors (<= 6 fills/bet, <= 90 bets, med >= $50) vs continuous
flow accounts that score huge but can't be mirrored at 3-17s lag.
First run (tape through 2026-07-20 16:59): 2,360 wallets screened, 561
score z >= 2 with pnl > 0, 25 copyable candidates at z 4.0-5.5 — chain
validation 742 payout vectors, ZERO proxy flips. Benchmark sanity: benched
sharps (LSB1 +3.28, EdwardIN +2.38) land positive, benched losers negative.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
'no orders found to match with FAK order' is the #1 miss class (13 of the
last 48h's misses) — the copy lands in the crater the sharp just swept,
before makers requote. Instead of recording the miss at the first
rejection, hand the OPEN to Copybot.fak_requote_retry: sleep fak_retry_s
(default 10, 0 disables) OUTSIDE the bot lock, then re-run the whole gated
buy path (fresh quote, price guard, depth gate) exactly once, mirroring
the webhook call site's fill drain so a retry fill books lag + bet rows.
A second rejection records the miss tagged 'twice (re-quote retry)'.
Paper-parity: the hook is installed in both modes, and PaperExecutor's
FAK model re-decides on the fresh book. tests/test_fak_retry.py covers
hook scheduling, no-double-count sizing (their_size=0 re-entry), ADD and
non-FAK bypasses, and the end-to-end retry thread.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ingest: sftp needs stdin=DEVNULL under launchd (no tty) — without it, empty/partial files decoded to 0 lines and the missing integrity guard MARKED+DELETED them (36 segments lost). Guard: a >200B gz that yields 0 lines is a failed fetch, left on box to retry.
digest: retry+backoff moved INTO _post so the digest send (not just the start ping) survives the post-wake DNS gap; ingest ping gets the same.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
copybot[bot]
gavindiaz