EASY WAY OUT!

specialist-follow · live capital

◈ DATA STALE

bot live

70%

— · updating…

build —

🇳🇱 Poly Copy · maker-only weather copy · executes from AMS3 (this box is geoblocked) · money lives in a DEPOSIT wallet Polymarket's UI does not show ·

loading…

🐋 Whale Watch — the weather traders we CANNOT see · these wallets hide their trade tape · but the Climate leaderboard metric is projected_pnl, which is CATEGORY-SPECIFIC and includes OPEN positions — so their weather P&L ticks live ·

◎ Position monitor · decided positions are COLLAPSED by default (measured 07-29: 46 of 69 open rows were already won or dead) · sorted by URGENCY = uncertainty × size × imminence · 🌡 is the live observed temperature from the free NWS feed, and "flips if" is how far reality must move to change the outcome — neither is available in Kalshi's own portfolio tab

🗓 Daily Report · frozen at 09:05 UTC once a day is complete — a frozen day NEVER moves again · today is live and will change ·

◈ Account value — total equity (cash + open marks)

Candles are OHLC of our own account value bucketed by the selected window (open = first sample in the bucket, high/low = extremes, close = last) — real, not synthetic. Purple bars are actual order flow ($ staked per bucket from the order ledger), so a move with no bars was the BOOK repricing, not us trading. The faint red trace is drawdown from the running peak.

◎ Performance gauges

↑ Cumulative banked P&L — realized, per settled trade

▮ Per-trade P&L distribution

⬡ Wallets — ranked by signal P&L · signal p&l judges the forecast (every market counted in full) · pro-rata attributes OUR dollars and sums to the book · they differ on the 12.3% of markets more than one wallet ordered · best to worst ·

◱ Niche win/loss record

▤ Paper cohort — new wallets on trial · ZERO capital · scored exactly like real money (ask we logged, fees in, $1/contract) · a wallet we never observe can never earn its way in ·

◔ Wallet lifecycle — promote / hold / bench · sample-based, patience-biased · size is the risk control until 30 copies · hard stop is the only fast trigger

▦ Settled-day P&L heatmap — banked $ per wallet per day

⧗ Settling — determined, awaiting payout

◈ Open positions — live, soonest settlement first

* = payout time measured from our own settlement history (daily weather pays 7-8am ET the morning after); no star = Kalshi's stated expiry, which for daily weather is a later admin deadline.

✓ Recently settled (banked)

⟟ Execution telemetry

⚙ Execution quality

⌁ Strategy config

⇄ Following

⊘ Skip analysis — why we didn't trade · deliberate filters vs genuine misses (deduped by market)

⚏ Engine scorecard — copy-bot vs Poly bridge

◎ Calibration — win-rate vs price paid

◷ P&L by hour of day (ET)

⧨ Signal funnel — seen → won

⬡ Live vs backtest edge

⟋ Slippage vs specialist (over fills)

▽ Underwater drawdown

⚔ Kalshi vs Polymarket — head to head · exact matches trade live · inexact matches paper-tested

🧠 Learner — self-improving loop

⇄ Poly → Kalshi bridge Arm B · PAPER

▦ Rewards — Kalshi liquidity provision

♥ System health

◎ Price bands — favourites or longshots? · by the price WE paid · a high win rate at 90c is not skill

◷ Clock — when each niche makes money · ET blocks

⚗ Forward test · pre-registered · scored only on data AFTER registration

⇄ What changed since yesterday

$ Cost ledger · banked P&L is already NET of these — this is what execution costs

⑂ Signal funnel — where each wallet leaks · different causes need different fixes

▷ Fired today

⚵ Leaks — what to cut · a leak needs to be BIG, PERSISTENT, and decidable · first-half vs second-half shows whether it is getting worse or reverting

⚙ Jobs · last-run is read from what each job WRITES, not from cron saying it fired

◉ Data feeds · written = the process is alive · moved = it is actually producing · a feed can be written every cycle and still be frozen

💵 Cash moved — SETTLE-day clock · the calendar below buckets by ET ENTRY day (“how did the trades I placed that day do”) · this buckets by ET SETTLE day (“how much money actually moved”) · a weather daily entered on the 15th settles the morning of the 16th, so the two clocks disagree by design — on 2026-09-16 the entry tile read −$2.14 while −$154.06 left the account

▤ Trading calendar — P&L by day traded · settled + still-open, attributed to the ET day the trade was entered · click a day for detail

🌡 Heat Risk — running max vs strike · KXHIGH* markets settle on the DAY'S MAXIMUM, not the current temp · source api.weather.gov ·

head = strike − the day's running max. Negative means the max has ALREADY passed the strike — the NO side is decided even if the market has not repriced. DANGER = within 2°F with heating still to come (15:00–23:00Z). Current temp is deliberately NOT the headline: a position can be dead while the thermometer reads low.

🌡 WX Confirm — our own weather opinion, as a second check on every copy · ADVISORY ONLY — it watches and scores, it does not block a single trade · the switch stays off until the trades it wants to block are PROVEN to lose money

🧭 What we did, in plain English · three versions failed before one worked · the failures are the useful part

🎯 Can it be trusted yet? · the gate only opens on FORWARD results — signals scored live, then graded when they settle

📡 Live scoring · every weather copy the bot makes, scored within 5 minutes · "thermometer" = anchored on the real station reading, the strong case

🧠 What the bot sees right now · the board is in the bot’s real BUY ORDER · it ranks by model probability, not by edge — measured on 219 settled positions, ranking by edge put the top half at +1.2pp and the bottom at +4.6pp (inverted), while ranking by probability put them at +3.5pp and +0.6pp · `edge_yes` decides who qualifies, probability decides who gets bought first when budget binds

⚾ HR Bot — our own home-run model · SEPARATE BOOK from the copy bot · every number here comes from hr_ledger.py, which reads the VENUE-reported fee · the copy bot’s recon_settlements flags these rows ours=0 and misreports them (it showed −$15.76 for a day this ledger scores at +$72.41), so the two books are never merged

🗓 Per game-day · banked only — open positions are never counted as profit

◎ Model calibration · the number that decides whether the edge is real · model_p is what the bot thought the chance was, actual is what happened · bucketed, never pooled — one blended figure hides a broken band

▤ Recent bets

🌡️ WX3 — status & feed health · the bot nowcasts the Kalshi Weather Index and buys near-certain legs in the last 20 min · feed staleness is the thing that breaks it: on 2026-09-16 it priced 47–62 min old observations as if they were 19 min old and went 5-for-10 against a model claiming 97.8%

🧭 Record by MODEL VERSION · eras are never pooled — a trade placed by a superseded model is not evidence about the one running now · only the CURRENT era counts

🎯 Our index vs Kalshi's published index · Kalshi publishes the settling index itself, free, 1-min cadence · this is our error against truth · measured 2026-09-16 at seven live decisions we ran 1.5–3.3°F LOW, always low, never high — larger than the drift it sits next to

◎ Calibration · model_p is what the bot claimed, actual is what happened · bucketed, never pooled — a blended figure hides a broken band

🧠 What it sees right now · every cycle, all cities, including why it did NOT trade

◎ Open trades · money currently at risk · an open position is NOT a loss — charging its cost with no payout is what drove the daily bot to a phantom −$35.03 against a −$50 kill with zero markets settled · a mark is shown ONLY when both sides of the book quote; a closed book marks to cost, so those read “awaiting settlement”

▤ Recent fills · settled only · open positions are never counted as profit

🌙 Night Owl — night-favourites sleeve · buys 85-97c favourites in overnight hourly temp markets · P&L is SETTLEMENT truth built from FILLS (this sleeve shares tickers with the copy bot, so a per-market settlement row cannot separate them) · open cost is never counted as profit

◑ Per night · a night runs 11pm-7am ET; hours after midnight belong to the previous evening · kill rule fires if any single night loses more than $8

◈ Positions · settled first, then still-pending · entry price is what we PAID per contract

Hourly temp markets pay 81 minutes after they close \u2014 i.e. at :20 past the following hour (measured over 10 settlements, zero variance). Because that lag is longer than an hour, the previous hour is always still pending when a new one opens.

📉 Recent Movers · what actually moved the balance ·

⇄ Markets — does PAPER agree with LIVE? · both legs measured by ONE code path (same price source, fee model, settlement, estimator) · FIXED SIZE per market, never volume-weighted · a verdict needs 10+ settled on BOTH legs ·

▤ Promotion queue · clears every gate: 25+ settled · $/market>0 · t≥2 · positive in BOTH halves · no single event >50% of P&L

▦ Paper cohort — full record · zero capital · scored on the same settlement formula as real money · click a row for detail ·

◷ Paper — decay watch · last-10 vs last-25 vs lifetime $/market · a falling line is an edge dying, not variance