docs: document Copy P&L finding (position win% != copyability)

README/FINDINGS/live/README updated for today's discovery: the sharps
win% is a position snapshot that over-counts scalpers; copy_pnl (flat-$50
replay, authoritative clob resolution) is the real copyability signal.
Most "sharps" lose copied; only Kruto2027 + fortuneking are copy-positive
(now the tracked pair). Refreshed stale counts (50 -> ~31 after the
30d-active filter) and noted position-level conviction.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

FINDINGS.md

@@ -184,8 +184,8 @@ time on winning conviction bets** — this is a copyability heuristic, NOT proof
 inside information (a short lead can be a genuine insider or just someone good at
 fast-resolving markets; we can't tell, and for copy purposes it doesn't matter).
 Of the 218, the gate drops the **"last-minute" wallets** (median lead <24h — you
-can't get the trade in before resolution), leaving **50 validated copyable sharps**
-(`watch_sharps.json`) with multi-day leads. The standout `0x60ec1744…` held 80%
+can't get the trade in before resolution), then a 30-day-active filter, leaving
+**~31 validated copyable sharps** (`watch_sharps.json`) with multi-day leads. The standout `0x60ec1744…` held 80%
 win over **1,017 forward conviction bets**; even the suspiciously-perfect `0x72e1…`
 (99/100% win) enters ~7 days early — a real forecaster, clearly not last-minute.
 These 50 are surfaced live on jaxperro.com/trading.
@@ -196,6 +196,32 @@ That funnel produces a copyable, forward-validated set — the strongest evidenc
 this project that
 followable skill exists.*
 
+## Copy P&L: position win% ≠ copyability (the scalper trap, 2026-06-23)
+
+The biggest caveat on the whole sharps table: **a high conviction win% does not mean
+you can profit copying the wallet.** The win%/record are computed from `curPrice ≥ 0.5`
+on resolved positions — a *position snapshot*. For a high-frequency scalper that
+massively over-counts: he buys ~$0.50, sells seconds later for ~+$1, and the snapshot
+records a "win" even though he never held to resolution. `ArbTraderRookie` shows
+**~100% conviction win (398-2)** yet a flat-$50 copy of his conviction bets, held to
+**authoritative clob resolution (winner by token_id)**, nets **−$790 (held 0-19)** —
+two independent replays agree (live portfolio −$687 ≈ standalone clob −$790).
+
+So `validate_timing.display_stats` now also computes **`copy_pnl`** — what a flat-$50
+copier actually realizes since June 1: replay their conviction entries, mirror their
+exits, settle held bets at clob resolution. This is surfaced as the **Copy P&L**
+column on the dashboard (default sort). The verdict it delivers: **most "sharps" lose
+when copied.** Of the ~31, only a handful are copy-positive — `Kruto2027 +$1,184` and
+`fortuneking +$430` (true hold-to-resolution betters); names that looked great on win%
+(`iohihoo` 88.7% → **−$749**, `ArbTrader` 99.5% → **−$790**) are scalpers that bleed.
+The live tracker now follows **fortuneking + Kruto2027** — the two copy-positive
+wallets — at $50/trade.
+
+*Lesson: judge a copy target by Copy P&L (a trade-replay with real resolution), never
+by position win%. Conviction must be measured at the position level (a wallet's total
+stake in a market), not per individual buy — a scalper splits one position across many
+small buys, so a per-trade threshold copies far more (and worse) bets than intended.*
+
 ## Repo layout
 
 - `insider.py` — the detector: z-score/p-value, timing/freshness/sizing signals,

README.md

@@ -197,12 +197,19 @@ which win 70–80% on genuinely-uncertain (~0.4–0.6)
 markets — trained pre-June and validated June: **62/83 stayed profitable forward
 (p≈0)**. A lead-time gate (`validate_timing.py`) then drops wallets whose wins come
 too close to resolution to mirror — "last-minute" entries (median lead <24h), which
-may be genuine insiders or just fast-market specialists — leaving **50 validated
-copyable sharps** (`watch_sharps.json`), surfaced live on
-[jaxperro.com/trading](https://jaxperro.com/trading). One held **80% win over
-1,017 forward bets**. This is the strongest evidence in the project that
-*followable* skill exists — though execution lag/fees and ongoing forward
-validation still gate turning it into real money.
+may be genuine insiders or just fast-market specialists — plus anyone inactive >30d,
+leaving **~31 validated copyable sharps** (`watch_sharps.json`), surfaced live on
+[jaxperro.com/trading](https://jaxperro.com/trading).
+
+**The catch — position win% ≠ copyability.** A high conviction win% (a position
+snapshot) doesn't mean you profit copying the wallet: a scalper buys ~$0.50 and sells
+for ~+$1, which the snapshot scores as a "win" though he never held to resolution. So
+the feed also precomputes **Copy P&L** — the authoritative flat-$50 copy result
+(replay entries, mirror exits, settle held bets at clob resolution). It's the real
+signal: `ArbTraderRookie` shows ~100% win but **−$790** to copy; of the ~31, only a
+few are copy-positive (`Kruto2027 +$1,184`, `fortuneking +$430` — true holders). The
+live tracker follows those two. Execution lag/fees and ongoing forward validation
+still gate turning it into real money.
 
 ---
 

live/README.md

@@ -88,8 +88,13 @@ wallet's scale (a whale's $200 bet isn't conviction; a minnow's is).
   winning conviction bets separates copyable sharps (multi-day lead) from
   "last-minute" wallets (median lead <24h) we can't mirror in time — could be
   genuine insiders or just fast-resolving-market specialists, can't tell which.
-  → **50 validated copyable sharps** in `watch_sharps.json` (shown live on
-  jaxperro.com/trading).
+  Also drops wallets inactive >30d. → **~31 validated copyable sharps** in
+  `watch_sharps.json` (shown live on jaxperro.com/trading). It additionally
+  precomputes every stat the dashboard renders — incl. **`copy_pnl`**, the
+  authoritative flat-$50 copy P&L (replay entries, mirror exits, settle held bets
+  at clob resolution by token_id). **Copy P&L is the real copyability signal** —
+  position win% over-counts scalpers (see FINDINGS "the scalper trap"); the table
+  defaults to sorting by it.
 
 Identifiers for a follow-worthy wallet: on its **≥$200 bets** — win ≥65%, avg
 entry 0.35–0.70 (edge, not favorites), +copy-ROI, FDR-significant, **median lead