Δ = Kate fair − PM YES. Positive Δ ⇒ Polymarket is underpricing the "above" outcome (cheap YES); negative Δ ⇒ overpriced. ±2pp deadband = fairly priced.

Δ = Kate fair − PM YES. Positive Δ ⇒ Polymarket is underpricing the "above" outcome (cheap YES); negative Δ ⇒ overpriced. ±2pp deadband = fairly priced.

Universe: "above" markets (both sides) + band signals passing entry discipline — price 55–75¢ (amended 2026-07-09; was 15–65¢), adjusted win prob ≥ 51%, net edge ≥ 5pp after cost, no outcome overlap with open same-asset positions. Win prob is the Deribit-implied probability haircut toward 50% (k=0.85, anti-overconfidence); EV% is expected ROI on capital at risk at that adjusted prob; Score = net edge × adjusted win prob. Mirrors the kate-runner execution limits.

Record coverage is computed at render time.

Realized results from the Polymarket wallet transaction ledger API, grouped by market date. P&L per market = redemptions + sells − buys; losing positions expire worthless with no redemption row, so their full cost is captured. ROI = net P&L ÷ capital deployed into resolved markets. Unresolved (same-day) positions are excluded until they resolve. Every calendar day in the range gets a row. A day with no wallet activity reads 0 / 0 — that is a day on which no trade was found or taken, and it is deliberately not the same as a blank. A day the API’s activity fetch never reached reads no data in amber; we do not report zero trades for a day we did not observe. A 0 / 0 row records the absence of wallet activity; it does not by itself prove the runner evaluated the market and declined — that requires the runner heartbeat still outstanding from Aneep. Educational only — not financial advice.

Same wallet ledger, same fills, one mechanical filter: keep only entries priced 55–75¢ (Amendment v1.1, adopted 2026-07-09) and drop everything the amended algorithm would now refuse — sub-55¢ coin flips, over-75¢ thin favorites, and stacked same-outcome positions (every stacked trade also fell outside the window, so this one filter catches all three leaks this week). The filter uses only information available at entry time (the fill price), not outcomes. Days that produce no qualifying trades are shown explicitly as 0 / 0 rather than left blank — standing down is part of the strategy, not a failure of it, and an empty row would be indistinguishable from a broken feed. This is a hindsight reconstruction on one week of data: it shows the discipline working, not a promised return. Going forward this portfolio runs live — the amended rules are the algorithm as of Jul 9, so future rows here are real, not theoretical. Educational only — not financial advice.

A 67% win rate still netted −$1.69. The full fill-by-fill post-mortem found three leaks, all execution discipline rather than signal quality:

1 · Coin-flip entries had no realized edge. Fills at 45–55¢ won only 29% against a 48% market-implied rate — the model’s fair value is overconfident near the money. Cost: −$13.69.

2 · Expensive favorites breached the price ceiling. 29 of 52 fills came in above the 65¢ ceiling set on Jun 25 (some at 95¢). The 75–85¢ band won 50% against 79% implied; at those prices one miss erases four wins. Cost: −$15.11.

3 · Correlated stacking. On Jul 1, four positions across BTC and ETH all lost on the same outcome (price grinding up into the adjacent bucket). On Jul 8, three BTC positions were the identical “≥$62k” view expressed three ways — one settlement price killed all three. Gross cost of the two incidents: −$27.11. Meanwhile the disciplined middle band worked: entries at 55–75¢ went 16 for 16, +$26.11.

Adaptations now in the spec (Strategy Spec Amendment v1.1, 2026-07-09): A1 — entry window moves from 15–65¢ to 55–75¢, cutting both the no-edge coin flips and the thin-payoff favorites. A2 — a new outcome-overlap gate: no new position may share a losing price region with any open same-asset position, which blocks every stacked bet above. A3 — production parity: the live runner must run the same audited gate code as the reference implementation and log every rejection, so a rule on paper is a rule in the fills. Review checkpoint after 30 resolved trades under the new window.