Tests — paper harnesses

Tape-scored forward window — no bot, no book. Study C follows screened maker-sharps' inventory leans: wallets are screened on strictly-prior tape each day (no self-selection), a lean fires when net maker inventory crosses $150 at ≥60% one-sidedness, and both arms are scored at the last print with on-chain truth — follow small leans ($150–500, absorbed flow) and fade whale bags ($2k+, picked-off makers). Entry-price optimism is stated in the pre-registration: a PASS graduates to a real-execution paper arm, never directly to money. Rows come from the nightly forward ledger.

Paper money, measurement mode. Study D trades T9's finding that same-outcome sibling markets reprice slowly: when an event group's most-printed market moves ≥10¢ inside 120s, the harness paper-FAKs $100 at each lagging sibling's live book (≤2 siblings, cooldown 10m/event) — but only within 4¢ of the sibling's stale print, and down-moves buy the sibling's complement token so a crashing book can't pose as a bargain. Every attempt logs the standing ask's premium over the stale print: median ≥ +8¢ over 3 days pre-registers the whole edge as a mirage (the tape "entered" at prints nobody still quotes) — a kill that needs no paper sample. Settles are provisional (CLOB flags) until the nightly re-grade against on-chain payout vectors (CHAIN badge); EV reports per episode and per dollar staked so partial fills can't flatter it. PASS ≥ +$4/$100 @ n≥400 & ≥5 days · KILL ≤ 0 @ n≥300 — the verdict binds to the chain-graded ledger per #23. Feed: wwf-lagbot.fly.dev/feed · refreshes every 20s.

A scoring arm, not a bot. v1 kept sampling every band on purpose — a band-gated harness would have destroyed the evidence that produced this hypothesis. So #27 scores only the 15–40¢ complements filled after its freeze, at a doubled bar (+$8/fill) because the band was found by slicing v1's own data; the other bands ride along as report-only controls. Every guard below is a hard PASS criterion — a headline that dies without its best day or its top five episodes is recorded CONCENTRATED and does not graduate. Bars: PASS ≥ +$8/fill @ n≥300 & ≥10 days with all guards clear · KILL ≤ 0 @ n≥200 or ex-best-day ≤ 0 @ n≥200.

Paper money, measurement mode. Study C passed its tape-scored window (+$2.17/lean, n=20,461) with every entry booked at the lean-side last print — the same assumption that made Study B look profitable until a real book priced it at −$8.51/fill. This harness answers whether the lean side is actually buyable: when a screened maker-sharp's net inventory in one token crosses $150 at ≥60% one-sidedness (follow arm, $150–500), it paper-FAKs $100 at the live ask — but only within 3¢ of the print, and a net-short lean buys the complement token. The screened set is published nightly from strictly-prior tape, so the box can never select the wallets that happened to win during its own window. Median ask premium ≥ +8¢ over 3 days kills the study without needing a paper sample. PASS ≥ +$2/fill & hit ≥ .56 @ n≥400 & ≥5 days, all guards clear. Feed: wwf-leanbot.fly.dev/feed · refreshes every 20s.