Files
winning-wallet-finder/HANDOFF.md
T

142 lines
7.9 KiB
Markdown
Raw Normal View History

# Session handoff — 2026-07-10 (rev 4: real money one step from live)
Self-contained pickup for a fresh session (human or AI). Read
[README.md](README.md) gotchas 115, [FINDINGS.md](FINDINGS.md) (the
2026-07-08 sections at the end), [ETHERSCAN_MIGRATION.md](ETHERSCAN_MIGRATION.md),
and [LIVE_ROLLOUT.md](LIVE_ROLLOUT.md). The repo is authoritative.
## THE CRITICAL PATH — finish live placement (one wrap + one port)
The real-money bot (`wwf-copybot-live`, armed, $24.73) cannot place yet.
Root-cause chain, every link PROVEN 2026-07-09/10:
1. First two qualifying signals CRASHED the bot inside the unguarded live
order path (Fly events: exit_code=1 at 21:21:55Z and 21:53:35Z, seconds
after each signal); boot baseline then marked the fresh trades seen.
ALL FIXED: executor never raises, per-wallet poll guard, baseline
exempts trades younger than the stale window.
2. **py-clob-client is ARCHIVED** (May 2026) — the CLOB rejects its orders
('invalid order version') globally. Reads/auth still work; placement is
dead for everyone on it. No newer PyPI release exists.
3. Successor: **Polymarket/py-sdk** (`pip install --pre polymarket-client`,
`SecureClient`). Proven working for us tonight: auth ✓, the signer's
**Deposit Wallet deployed: `0x455e252e45Ee46d6C4cc1c8fAdD3899d68f245a1`** ✓,
gasless relayer transactions execute ✓ (real txs: 0x935be727…,
0x1044d3c7…), `setup_trading_approvals()` ✓ (max-uint on all spenders),
order FORMAT accepted by the exchange ✓.
4. **Sole remaining gap: the wrap.** The user's $24.73 sits SAFE as native
USDC at the deposit wallet (his key controls it). The 2026 exchange
trades a wrapped-collateral token (`0xC011a7…`; wraps native USDC; UI
flows wrap on deposit — our raw transfer bypassed it), and the CLOB's
balance view counts ONLY that token → orders reject with 'balance: 0'.
Attempted: periphery `0x93070a…`.wrap(asset,to,amount) via gasless batch
→ 'batch would revert' (likely caller-gated). Neg-risk adapter ruled out.
**Next session, in order (do NOT guess with live funds):**
1. Read **Polymarket/ts-sdk** (same unified family) for the deposit-wallet
wrap/deposit implementation — it contains the exact supported call.
Also check docs.polymarket.com's new wallet/collateral pages.
2. Replicate that one call via `client.execute_transaction` (the gasless
path is proven). Success = `get_balance_allowance` shows ~$24.73.
3. Port `LedgerLiveExecutor` to the new SDK: `SecureClient.create(
private_key=…)` (wallet auto-resolves; NO api_key needed at runtime once
approvals exist), BUY = `place_market_order(token_id, side="BUY",
amount=<USD>, order_type="FAK")`, SELL = `shares=<n>`; parse the result
for filled shares/price; keep py-clob-client temporarily for reads.
Repoint the 1.4 cash anchor at the deposit wallet's collateral view —
the live bot currently alarms `CASH≠CHAIN +24.73` because the anchor
still reads the emptied legacy proxy (expected, harmless).
4. `pip` add `polymarket-client` (--pre) to fly.Dockerfile, deploy, re-run
`host/order_probe_v2.py` (contains the ENTIRE working onboarding
sequence) until a $5 round trip fills, then run LIVE_ROLLOUT Phase 4 on
the first organic fill.
5. Cleanup: probe runs minted several Builder API keys —
`fetch_builder_api_keys` / revoke extras.
Meanwhile the armed live bot is harmless-by-construction: every attempt
returns ok:False and records an honest missed row.
## System state (all healthy, 2026-07-10 00:47Z)
- **Paper bot** (`wwf-copybot`, push mode): Set E, fresh $1,000 book reset
2026-07-08 20:45Z. **Day 1½: realized +$1,107.57, 36 copies, 6 open,
drift CLEAN** (the 11.32 forensics closed at 0.00 — see below).
Biggest single win: mirrored Kruto's 3¢→48¢ Hive scalp for +$722.
- **Live bot** (`wwf-copybot-live`, poll 60s): armed via LIVE_CONFIRM,
$24.73 book, 0 trades (see critical path), Discord pings wired
(DISCORD_WEBHOOK secret), auto_redeem off. Disarm:
`flyctl secrets unset LIVE_CONFIRM` or `flyctl apps stop`.
- **Set E** (both books + backtest): LSB1, imwalkinghere, Kruto2027,
0xbadaf319, gkmgkldfmg, AIcAIc, 1kto1m — pinned floors, identical rules
(min_price 0.01 everywhere since 2026-07-09 18:05Z, user-approved).
- **Dashboards**: jaxperro.com/trading (paper+backtest+sharps),
jaxperro.com/live (real money). Shared renderer
jaxperro/trading/copybot-section.js; `?botFeed=`/`?rmFeed=` overrides.
- **Watchdog**: Fly /health check (paper app) + GH Actions watchdog.yml →
Discord. NOTE: the live app has NO watchdog (poll mode, no HTTP) —
liveness = flyctl logs + feed freshness; wire one before raising caps.
- **Daily pipeline**: Mac launchd 08:00 + pmset RTC wake 07:58 +
caffeinate; appends live-vs-model calibration.csv row daily.
## Completed 2026-07-09 (the honesty & detection sprint)
- **Fill-split merging**: same-token BUY clips ≤120s apart merge into the
bet the sharp actually made before the floor gate (gkmg's $612 MOUZ entry
read as 3×$204 sub-floor clips; he sold +60% in 5 min). Both books.
- **Missed-ledger truth**: missed bets settle at the sharp's EXIT when they
sold pre-resolution (Kruto Hive miss = honest +$75 counterfactual on $5);
settled won/lost rows get one exit-print re-check (fast-resolve race);
every row relabeled with its diagnosed cause; the two silent follow-path
exits (no-ask book, order rejection) now record misses.
- **Clobber fix + drift forensics CLOSED**: re-entry on a settled token
archives (token#ts keys) instead of overwriting; three vanished records
reconstructed from fills-ledger/state-history; paper drift 0.00.
- **Order-book snapshots** (`book` field) logged per copy → empirical fill
model later. **payouts.resolution_time(cond)** wired (Etherscan V2,
exact on-chain res_t; metadata was 29h wrong on the test market).
- **min_price parity** 0.05→0.01 on the live book (user-approved; the band
— not detection — blocked the 16x Hive copy).
## Queued work (after the critical path)
1. ETHERSCAN_MIGRATION.md phases 0→5 (chain-sweep res_t backfill → shadow
audit → consumer flips → trust.py simplification LAST).
2. Empirical fill model from the accumulating book snapshots; depth gate
before real-money sizing.
3. Funding-cluster tracer port to Etherscan logs (Alchemy free tier caps
getLogs at 10 blocks).
4. Live-app watchdog (no /health in poll mode).
5. Decide the leftover **$32 on polymarket.us** (user's first deposit,
wrong venue — withdraw or abandon).
## Watch list (does Set E hold up?)
- **0xbadaf319** — two-sided arb/merge clips, not directional conviction;
TOP demotion candidate; judge on forward copies.
- **AIcAIc** — 42% held-win; sell-timing edge (lag-fragile). Second.
- **1kto1m** (z=2.4), **gkmgkldfmg** (z=2.05) — near the gate floor.
- Rejected-with-evidence (don't re-add): oliman2 (true lifetime ~$19k, not
PM's $112k), leegunner (negative to copy), 0xb0E43B/ArbTraderRookie
(refund harvesters), lma0o0o0o (negative in shared book), Winnertraders
(6.6-day locks).
## User to-dos (security)
- Revoke the polymarket.us API key `f03d9eb5…` (secret appeared in a chat
screenshot) — if not already done.
- Rotate the Discord webhook that was committed to this repo's history
pre-2026-07-09 (spam-risk only).
## Operational quick-reference
- Follow-set change: edit live/copybot.paper.json → `./live/deploy_bot.sh`;
keep backtest.json mirrored.
- State surgery: `flyctl machine stop` FIRST → edit → push →
`flyctl apps restart` → VERIFY the first heartbeat (gotcha 15; bot memory
wins conflicts; boot clones can read a stale GitHub replica briefly).
- Never 2 Fly machines per app (`flyctl scale count 1`).
- Cache single-writer; three cursors per wallet (gotcha 12).
- Bot commits its own state/feed — `git pull --rebase --autostash` first.
- Image changes (fly.Dockerfile, host/) → `flyctl deploy --remote-only`;
code/config → push + restart. Live app config = fly.live.toml.