Labs

Published research reports live in this folder, one subfolder per report, each with the script that produced it, the data window, the filters, and the losing cases. The scripts themselves live in scripts/; this page is the index.

Convention for a report folder:

labs/YYYY-MM-DD-slug/
  README.md        the report (English), with a Russian version README.ru.md
  run.py           the exact script or command that produced the numbers
  figures/         charts
  data-window.txt  first and last snapshot timestamp, market count, filters

Published

#	Report	One-line result
1	Longshot fade (RU)	On 183k closed markets, outcomes priced 5–20¢ within a week of the event win more often than priced; the textbook overpricing exists only under 2¢ and a month out, worth 0.4–1.7¢ per share. Selling longshots as a taker loses net.
2	Maker rewards (RU)	On 479k book snapshots from May 2026, the rewards program paid ~0.9%/day of corridor capital (median market 1%); fills at the touch were benign on average (+0.7¢/share size-weighted); a 1,000-share two-sided quote modelled to 2.4%/day median. A $100 account is ineligible on 96% of market-days, and live pools today pay a quarter of May.
3	Cross-venue	26 days, 4.9M quotes, 2,447 pairs. Kalshi and Polymarket quote the same game within half a cent and the same politics market within a cent; after fees the cross-venue arbitrage clears in 0.4% of game samples and ~1¢ on politics for hours at a time, on two venues one account cannot use together. On politics Polymarket moves first and Kalshi closes 70% of the disagreement; an hour before close Kalshi's price was slightly closer to the outcome. 1.1% of settled pairs were wrong pairs.
4	Whale of the week	Weekly. The #1 wallet on the Polymarket leaderboard, read from the public API: how many fills, how much of the money is one position, maker or taker, and the resolved losers the public profile never shows. This week: $1.8M on the board, 74% of it one bet on one match, 6 hidden losers worth −$2.5M in 28 days.

Planned reports

#	Report	Question	Scripts
3	Cross-venue gaps	Size and persistence of Kalshi vs Polymarket vs sportsbook gaps on matched events	cross_market_radar.py, odds_api_audit.py, app/integrations/kalshi_seed_map.py

Research scripts by theme

Descriptions are taken from each script's header; open the script for the full method.

Calibration and resolution - category_calibration.py: calibration overlay per market category - external_calibration.py: calibration against Gamma resolved markets - resolution_risk.py: resolution-risk filter for markets with ambiguous rules - resolutions_watcher.py: alert when the count of resolved weather markets crosses thresholds - weather_resolution_scorer.py: first ground-truth Brier check on weather markets

Fade research (buying dips, selling pumps) - fade_shadow_scanner.py: shadow scanner for the fade-any edge (median +3.61% at 30 minutes, 57% hit rate on the validation set) - fade_replay.py: replay analyzer for fade-any trades - fade_any_funnel.py: why the strategy produced zero trades after deploy - fade_any_correlation.py: correlation tagging of fade-any trades - impact_fade.py: impact-fade detector with walk-forward validation - longtail_moonshot_research.py: long-tail (longshot) research - longtail_event_audit.py: event-level audit of long-tail trades

Maker and rewards - paper_maker_forensic.py, _v2.py, _v3.py: forensic and live-readiness audits of the paper maker - paper_tail_maker_sim.py: tail-maker simulator - rewards_scanner.py: Polymarket liquidity-rewards scanner

Cross-venue and arbitrage - cross_market_radar.py: cross-platform radar (Kalshi seed map vs Polymarket) - neg_risk_arb_scanner.py: negative-risk (sum-to-one) arbitrage scanner - odds_api_audit.py: The Odds API usage audit - hedge_shadow.py: hedge-manager shadow simulator

Smart-money (wallet following and fading) - smart_money_poc.py, smart_money_poc_v2.py, smart_money_v2_weather.py: proofs of concept - sm_validator.py: early validator - sm_canary_lab.py: shadow scoring of wallet signals - sm_mirror_backtest.py, sm_dollar_backtest.py: mirror and fade dollar-PnL backtests - sm_fade_canary_runner.py: paper canary that fades tracked losing wallets

Single-family canaries - tennis_underdog_shadow.py, tennis_underdog_resolve.py, tennis_underdog_paper_runner.py, tennis_underdog_paper_resolver.py: tennis underdog canary, shadow and paper phases - weather_bucket_shadow.py: weather bucket shadow - wti_mr_canary_runner.py: WTI mean-reversion canary - asset_target_dollar_backtest.py, asset_target_haircut_backtest.py: asset-target backtests - event_diagnostic.py: event-strategy diagnostic

Reporting and operations - analysis_dashboard_24h.py: cross-strategy 24–48h analysis - mfe_mae_report.py: MFE/MAE retrospective - morning_report.py: daily Telegram summary - poll_fills.py: production fills logger - bot_health_alerts.py: operational health alerts - ai_veto_stats.sh: LLM veto effectiveness from container logs - ai_debate_bridge.py, debate_heartbeat.py, tg_relay_worker.py: tooling for the two-model review loop used during development