Files

T

jaxperro 0b625c994a README: July test now 8 wallets in two stake classes (volume 4% / whale 12%)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

2026-07-04 19:18:51 -04:00

17 KiB

Raw Blame History

🏆 Winning Wallet Finder

Find Polymarket wallets with a real, statistically-verifiable edge, test whether copying them actually makes money — with real fees, lag, and slippage — and copy-trade them with a 24/7 bot (paper today, live-capable).

This started as "copy the smart money." Along the way we tested — and ruled out — six systematic public-data strategies, and found that the only signal that holds up is statistical improbability: wallets that win far more than the prices they paid imply. This repo is the tooling for finding, validating, and copying those wallets, plus an honest record of everything that didn't work.

Read FINDINGS.md for the research story. TL;DR: detection works; profitable copying is plausible but unproven — every backtest here is in-sample by construction, and the July 2026 forward test (live now) is the arbiter. Treat headline returns as ceilings, not forecasts.

The system today (July 2026)

Three deployed pieces + one static dashboard:

piece	where it runs	what it does
daily pipeline (`live/daily.sh`)	this Mac, launchd 10:00	refresh the bet cache → 5-gate skill scan → fee-aware sharp selection → conviction floors → backtest book → publish JSON feeds to GitHub
copybot worker (`copybot.py` via `host/start.sh`)	Railway, 24/7	polls the 8 followed wallets every 60s, paper-copies their conviction bets with real fees/lag/slippage accounting, settles at CLOB resolution, commits its book back to the repo
Discord digest (`live/discord_daily.py`)	end of the daily pipeline	one message/day: the sharp list with profile links + 30-day conviction stats (per-trade pings retired 2026-07-04; the old Alchemy watcher lives in `archive/webhook_receiver.py`)
dashboard	jaxperro.com/trading (static, in the `jaxperro` repo)	renders the three JSON feeds: live bot book, backtest book, sharp table

The July 2026 live test: a fresh $1,000 paper book (started 2026-07-02, on Railway) following eight wallets in two stake classes (follow.wallet_class): volume (4% of equity/bet) — Kruto2027, shisan888, fortuneking, LSB1, imwalkinghere, iohihoo — and whale (12%/bet) — Stavenson and the 0x4bFb… whale, the two big-clip informed holders the trusted-row re-validation surfaced (added 2026-07-04; see FINDINGS "The holder blind spot"). Every fill records detection lag, price slippage, and the taker fee; missed bets are recorded and settled hypothetically. If this month's measured numbers hold up, real money follows (see LIVE_TEST.md).

 data layer          selection                        execution              display
 ──────────          ─────────                        ─────────              ───────
 live/cache.duckdb ─▶ skill.py (5-gate funnel)        copybot.py (Railway) ─▶ jaxperro.com/trading
 (schema v2:          conviction_scan.py (p80 bets)   · 4%-of-equity stakes   · copybot_live.json
  33k wallets,        validate_timing.py (fee-aware   · taker fees modeled    · portfolio.json
  19M resolved bets,   copy replay → watch_sharps)    · lag/slip per fill     · watch_sharps.json
  token-keyed,        portfolio.py (backtest book)    · missed-bet ledger
  archival)           sync_floors.py (bot parity)     · CLOB settle + redeem

File map

path	role
`live/`	the current system: cache, scanners, sharp selection, backtest, daily pipeline (live/README)
`copybot.py`	the copy-trading bot: push/poll trigger → follow filter → execution engine (paper + live)
`archive/copytrade.py`	the execution engine the bot reuses: sizing, risk gates, price guard, paper/live executors
`host/start.sh`	24/7 worker bootstrap for Railway/Fly/VPS (clones repo, resumes committed state)
`LIVE_TEST.md` · `preflight_live.py` · `redeem.py`	real-money runbook, read-only credential preflight, on-chain redemption
`insider.py`	the original detector: z-score, pre-resolution timing, fresh-wallet flags, funding-cluster rings
`smart_money.py`	shared HTTP helper + survivorship-corrected win-rate dashboard (`:8899`)
`archive/webhook_receiver.py`	retired 2026-07-04: Alchemy webhook → per-trade Discord pings (replaced by `live/discord_daily.py`'s daily digest)
`wide/`	frozen-subgraph bulk scanner (1.76M wallets, historical only — subgraph froze Jan 2026)
`archive/`	the six strategies that didn't work, kept honest (archive/README)
`hunt.py` · `huntwide.py` · `oos.py` · `copyback.py`	earlier research sweeps/backtests (superseded by `live/`)

The core idea: z-score, not win rate

Every Polymarket bet has an entry price that is the market's estimate of its odds (a YES at 30¢ ⇒ market thinks 30%). If you have no edge, over many bets you win about the sum of your entry prices — call it expected wins.

z = (actual wins − expected wins) / standard deviation

z = 0 → you won exactly what your prices implied → no edge.
z = 3 → ~1-in-740 by luck. z = 5 → ~1-in-3.5M.

Why this beats win rate: a wallet that bets longshots and wins 14% when the odds implied 8% has a huge edge (high z) despite a low win rate; a wallet buying 90¢ favorites and winning 90% has z≈0. Win rate is also survivorship-biased on Polymarket (losing shares sit unredeemed and invisible — see FINDINGS). And even true win rate over-counts scalpers — so the final selection metric is fee-adjusted Copy P&L: what a flat-$50 copy of the wallet's conviction bets actually returns after taker fees (replay entries, mirror exits, settle at CLOB resolution). Judge by Copy P&L, never win rate.

Quickstart for a new developer

git clone https://github.com/jaxperro/winning-wallet-finder
cd winning-wallet-finder
pip3 install duckdb                    # the only dep for research/selection
cp config.example.json config.json     # secrets live here (gitignored)

# selection layer (live/) — everything reads the local bet cache
cd live
python3 enumerate.py 30                # build a candidate pool (last 30d markets)
python3 collect.py                     # pull their bets into cache.duckdb (resumable)
python3 skill.py                       # 5-gate skill funnel -> watch_skilled.json
python3 conviction_scan.py             # conviction-bet profile scan
python3 validate_timing.py             # fee-aware copy replay -> watch_sharps.json
python3 portfolio.py                   # the backtest book -> portfolio.json
./daily.sh                             # or: the whole thing, end to end

# copy bot (repo root) — paper by default, no orders ever without --live
python3 copybot.py --config live/copybot.paper.json --state /tmp/s.json --poll 60
python3 copybot.py --test-wallet 0x…   # dry-run one wallet's latest trade

# live trading (real money): read LIVE_TEST.md, then
pip3 install py-clob-client web3
python3 preflight_live.py              # read-only credential/balance check

The cache is the point: every score re-runs in seconds from live/cache.duckdb (~33k wallets / 19M+ resolved bets) instead of hours of API pulls. Schema v2 is token-keyed, provenance-tagged, and archival (refreshes upsert instead of wiping; failed pulls are never cached as "no bets") — details in live/README.md.

Config & secrets (all gitignored)

file	holds
`config.json`	`daily_webhook` (the Discord digest), Alchemy key, the followed-wallet list + per-wallet conviction floors (auto-refreshed daily by `sync_floors.py`)
`config.live.json`	live-trading credentials (`private_key`, `funder_address`) + tiny test caps — see `LIVE_TEST.md`
Railway `copybot` service	`GITHUB_TOKEN` (fine-grained PAT, contents-RW on this repo — the bot commits its state/feed back), `DISCORD_WEBHOOK` no longer used (per-trade pings retired)

How the copy bot models reality

The whole point of the July test is that paper ≈ live. Every mechanism the backtest and bot share:

Taker fees (Polymarket V2, since 2026-03-30): fee = shares × rate × p(1−p), sports rate 0.03 — charged on every marketable entry and exit; redeeming at resolution is free. Fee-adjusted Copy P&L also drives selection.
Lag + slippage: the bot fills at the live CLOB ask at detection (~60s poll), logging per-fill detect_lag_s and slippage_pct; the backtest applies a +0.5%/~90s haircut. Measured so far: ~48s avg / +0.8% avg slip.
Dynamic sizing, two wallet classes: each bet stakes a fraction of current working equity set by the followed wallet's class — follow.wallet_class marks a wallet volume (default, 4%) or whale (12%), with the fractions themselves in follow.class_pct. Stakes compound both ways and are halved while equity is below 80% of its high-water mark. The rule binds per market — adds that mirror a sharp scaling in can only top a position up to the current stake size, never past it. (The profit-sweep threshold below stays on the base 4% so which wallet happens to trade doesn't change when profits get banked.)
Profit ratchet (stake_cap_usd: 250): stakes pin at $250; once the book outgrows that level, surplus cash sweeps to a banked reserve that never bets — locked-in profit, and fills stay inside realistic book depth. A per-event correlation cap exists (risk.max_per_event) but is off.
Entry cap 0.95: entries above 95¢ are skipped (follow.max_entry) — the June sweep showed >95¢ favorites lower final equity even while winning (slip + fee eat the 1–3% payouts; the capital compounds better elsewhere).
Asymmetric price guard: a price below the sharp's fill is never blocked (better odds, by rule); only adverse drift >5% is skipped.
Conviction filter: only copy a wallet's top-20%-by-stake bets (per-wallet p80 floor, kept in sync with the dashboard by sync_floors.py).
Missed-bet ledger: every bet the bot couldn't take (cash deployed, price ran up) is recorded and settled hypothetically — capacity costs are measured, not invisible.
Settlement: winners settle at authoritative CLOB winner flags (see the gotcha below — it caused the project's worst bug); live mode auto-redeems on-chain (redeem.py; neg-risk markets need manual redeem).

Safety: paper is the default; live requires mode:"live" and --live and a typed confirmation phrase, under hard caps. The GH-Actions cron runner is retired (GitHub throttled */5 to ~2h in practice — it copied 1 of ~104 qualifying trades in June; the always-on Railway poller replaced it).

Data sources

Source	Used for
`data-api.polymarket.com`	positions, trades, activity (+`eventSlug`), leaderboard
`gamma-api.polymarket.com`	market metadata (NB: `condition_ids` filter returns nothing for resolved markets)
`clob.polymarket.com`	order books, prices, authoritative resolution (`winner` flags), market slugs
Alchemy (Polygon)	funding-cluster traces + the live trade webhook

Candidate next sources (researched 2026-07, not yet wired in):

Source	Would unlock
Goldsky Turbo Pipelines	per-fill order events with timestamps for every wallet (Polymarket killed subgraphs with the 2026-04-28 v2 migration) — fixes the cache's two blind spots: no entry times, and position-level aggregation hiding scalps. See also warproxxx/poly_data, Bitquery
PolymarketData.co	historical order-book snapshots (Aug 2025+) → depth-aware fill model, the known step before sizing up
Pinnacle closing lines via SharpAPI / sportsapis.dev / BettingIsCool (Pinnacle closed its public API 2025-07)	closing-line-value as an independent "was this bet sharp" ground truth; a Pinnacle suspension on an ITF/esports match is itself a fixing signal
Polysights Insider Finder	cross-check for flagged insider wallets

Gotchas a maintainer must know

CLOB winner flags: false means "not yet", not "lost". Every token of an unresolved market reports winner: false; resolution flips exactly one to true. Any settle/replay logic must gate on any(winner is True) first. Treating false as lost made the bot settle live in-play positions as instant losses (four winning bets booked −$180 on 2026-07-02). Slow pollers never see this — markets genuinely resolve between checks — which is why it survived June.
eventSlug sub-splits one game (…-2026-07-01-more-markets, …-second-half-result): group by the …-YYYY-MM-DD prefix (event_key() in copytrade.py / portfolio.py) for anything per-event.
Win rate over-counts scalpers and is survivorship-biased — never select on it; the fee-adjusted Copy P&L replay is the selection metric.
/closed-positions sorts by realizedPnl by default — always pass sortBy=TIMESTAMP or you sample only the biggest wins.
cache.duckdb is single-writer — a running collect.py blocks even read-only connections; the daily pipeline serializes for this reason (and caps stale refreshes at STALE_CAP=2500/run so it stays daily).
GitHub Actions */5 cron actually fires ~every 1.5–2.5h — never use it for anything latency-sensitive (it copied 1 of ~104 trades in June).
GitHub Pages soft-limits ~10 deploys/hour on the jaxperro repo — batch dashboard pushes (see that repo's README).
Cached res_t/won can be fake for high-volume wallets. When the data-api omits endDate, res_t falls back to the wallet's sell time and won is the price direction at pull — a scalper's sold-at-profit position masquerades as a resolved win (ArbTraderRookie's rows were 100% this). Selection must read trusted rows only via live/trust.py (cross-wallet consensus res_t + pulled-after-resolution + resolved not False). Also: never judge a held edge on a replay window shorter than the wallet's entry→resolution lead — that's how the long-lead holders were being filtered out (see FINDINGS "The holder blind spot").

The research (how we got here)

The 5-gate funnel (live/skill.py): a wallet is "skilled" only if it clears n ≥ 15 resolved bets → z > 0 → Benjamini–Hochberg FDR @5% → split-half out-of-sample persistence → not a market-maker (all resolved-only: early-sold positions in unended markets are marks, not outcomes, and never score).

The clean test (June 2026): high-win-rate "favorite-rider" wallets looked +23.6% in-sample and lost −7.4% once selected without look-ahead — exactly the LBS/Yale "~60% of lucky winners become losers" result. Don't copy win rates.

The repeatable find: score wallets on their conviction bets (top 20% by stake, per-wallet p80) — the edge is wallets that win 70–80% on genuinely uncertain (~0.4–0.6) markets. Trained pre-June, validated June: 62/83 stayed profitable forward (p≈0). A fee-aware flat-$50 copy replay then keeps only wallets that are actually profitable to copy (scalpers with ~100% shown win rates lose money when copied) → currently 12 copy-positive holders in watch_sharps.json, refreshed daily.

The backtest (live/portfolio.py): the followed four, June 1 → now, with fees/lag/dynamic sizing: +531% — but June is the month these wallets were selected on, so that's an in-sample ceiling. July, live, is the test.

What didn't work (see FINDINGS.md + archive/): copy-trading raw, win-rate ranking, LP reward farming, binary & multi-outcome arb, cross-venue PM↔Kalshi arb — all efficient or illusory.

The honest verdict

Detection works. z + timing + funding clusters reliably surface anomalous wallets.
Copying is promising but unproven. Selection is fee-aware and execution-realistic now, but every historical return in this repo is in-sample. The running July book — real lag, real fees, measured slippage, missed bets counted — is the first number that deserves trust.
Scale carefully. Above ~$250/clip in thin sports books, best-ask fills turn optimistic; a depth-aware fill model is the known next step before sizing up.

17 KiB Raw Blame History Unescape Escape