Priced financing: the first reclaim of the pricing level, long

Disposition: FALSIFIED

The test used the 30-signal manifest committed at fbce02dc56dd4663a6c4217c2cf6bd9e93d4a513 and the evaluator committed at 0399d0e0. The signal manifest hash matched the preregistered eb13ff7edee281da414710255debca542cce93f58f3f3a9fbf1624887e957fe2 before any outcome database connection was opened.

Primary result

• Qualifying trades: 30
• Mean net return after 100 bps friction: -1.3202%
• One-sided deterministic sign-flip p-value: 0.899672 (1,000,000 draws)
• Largest event share of total positive P&L: 71.62%

All required pass conditions were conjunctive. Sample size passed, but mean return, statistical significance, and the positive-P&L concentration guard all failed. The hypothesis is therefore falsified, not underpowered.

Descriptive diagnostics

• Median net return: -2.2452%
• Win rate: 10.0%
• Deal-failure exits: 26
• Time exits: 4

These diagnostics do not alter the disposition. No subgroup, threshold, provenance split, transaction-role split, repeated-symbol exclusion, or alternative exit was searched for or used to rescue the result.

Artifacts

• Trade ledger: research/results/priced_financing_first_reclaim_history_trades_v1_2026_08_06.csv
  • SHA-256: 4ab53f3cf36f48d66610bcc479a468299971e4278c99542dcd31806d99bc5eb6
• Machine summary: research/results/priced_financing_first_reclaim_history_summary_v1_2026_08_06.json
  • SHA-256: 3c622d232abdeff8039654ac3e5351da697a107753f2612ca7bafb62cfa2b530

Frozen: 2026-08-05 after descriptive validation-state commit 53f62fe1, before selecting or reading any evaluation event.

Discovery basis and separation

The v1 and v2 packets are discovery data. They showed a recurring mechanical sequence in one situation: after a priced issuer financing becomes public, common stock closes below the disclosed deal price and later reclaims it. Neither cohort may enter this test. Their returns and post-signal outcomes must not be calculated.

This registration selects one path only. It does not compare all state labels and choose the best.

Hypothesis

In a clean small-cap issuer financing with an independently verified first-public pricing clock, the first exact five-minute reclaim of the deal price after at least one post-clock close below it is evidence of absorption. Buying after that reclaim and exiting on a fixed deal-level failure or time horizon has positive net expectancy after conservative friction.

Eligible situation

• verified priced primary or mixed issuer financing;
• common stock, ordinary shares, or ADS that maps directly to the traded security;
• deal price from $1 through $20 inclusive;
• independently verified first-public pricing timestamp, not SEC-acceptance fallback;
• no explicit symbol mismatch, later-split distortion, missing tape, or unresolved ADS ratio;
• common shares or ordinary shares sold at a standalone per-share price; pre-funded warrants offered in lieu of shares at the same economic price are allowed;
• exclude units and any price that includes an attached common, ordinary-share, or long-dated warrant.

The final package-price exclusion is frozen from v2 purity evidence before evaluation retrieval. It may reduce power; it may not be relaxed after outcomes are observed.

Disjoint evaluation sampling

Exclude every canonical accession in the frozen v1 and v2 cohorts at the query level. Rank remaining eligible events by md5(canonical_accession || 'priced-financing-first-reclaim-long-v1') and process in that order.

For sample construction, retrieve bars only from the verified pricing clock through the first signal and its immediately following bar. Do not retrieve or inspect later outcome bars while accumulating the sample. Continue through the fixed ordering until 30 qualifying signals are obtained or the eligible population is exhausted. Commit event identities, signal timestamps, and entry-bar timestamps before extracting any outcome window. Do not replace a qualifying signal for subjective chart quality.

If fewer than 30 qualifying signals exist, stop before computing aggregate outcomes and report UNDERPOWERED. Enlarging or relaxing the cohort after that result requires a new prospective registration and may not reuse this evaluation population.

Signal and execution

Work in regular and extended-hours five-minute bars from the verified clock onward.

1. Observe at least one five-minute close strictly below deal price.
2. The signal is the first subsequent five-minute close strictly above deal price.
3. Enter long at the open of the immediately following available five-minute bar.
4. If there is no following bar in the source window, no trade qualifies.

Exact comparisons are used. There is no percentage band, tick tolerance, volume filter, market-cap filter, ranking, or discretionary veto. The signal bar close cannot also be the entry price.

Exit and net return

After entry, exit at the earlier of:

• the open of the bar immediately following the first five-minute close strictly below deal price;
• the final regular-session five-minute close (15:55 ET bucket) of the next observed market session after the signal session.

If a failure close occurs in the final available bar and no following bar exists, use the next observed regular-session open. Apply 100 basis points round-trip friction to every trade:

net_return = exit_price / entry_price - 1 - 0.01

This is a deliberately conservative common friction assumption, not a claim that every name is fillable at that cost. Capacity and realized slippage require separate live validation.

Primary test and falsification

The primary estimand is equal-weighted mean net return per qualifying event. Test the one-sided null that mean net return is not positive using an exact sign-flip randomization test when computationally feasible, otherwise a deterministic Monte Carlo sign-flip test with seed priced-financing-first-reclaim-long-v1 and at least 1,000,000 draws.

The hypothesis passes only if all are true:

• at least 30 qualifying trades;
• mean net return is greater than zero;
• one-sided sign-flip p < 0.05;
• no single event contributes more than 50% of total positive P&L.

Otherwise it is falsified, except that n < 30 is UNDERPOWERED and must return before aggregate outcomes are computed.

Median net return, win rate, distribution quantiles, timestamp-to-signal, holding time, and results by transaction role are descriptive diagnostics only. They cannot rescue a failed primary test.

Dependence and integrity

Canonical accession is the event key. Multiple accessions for one financing remain one event. If two events share a symbol and clock, retain the earliest fixed-order event only. Report repeated symbols and same-day clusters; do not silently count linked financings as independent evidence.

The implementation must enforce the minimum-sample return before loading outcome prices. Any data bug or required rule amendment discovered before outcome extraction must be documented and committed before rerunning. After outcome extraction begins, this specification is immutable for v1.

Amendment 1 — fixed signal horizon and immutable discovery exclusion

Added: 2026-08-05 before selecting or retrieving any evaluation identity or bar.

The original signal language said to scan from the verified pricing clock "onward" but did not set a maximum horizon. That would allow an economically unrelated reclaim weeks or months later to qualify. Bind the pricing clock to an anchor session using the existing financing convention: a clock before 16:00 ET on an observed session binds to that session; a clock at or after 16:00 ET or on a non-session binds to the next observed session. Signal search ends at 20:00 ET on the second observed session after the anchor session. Missing or sparse bars do not extend this boundary. A below close and later above close may span a normal overnight or source gap inside the fixed window; both must be actual observed bars, and entry remains the next actual observed five-minute bar.

The discovery exclusion must not rely only on canonical_accession, because that value is currently min(accession) in a dynamic grouped view and can change when historical accessions are backfilled. Before historical ingestion, materialize an immutable exclusion ledger containing:

• every accession in source_accessions for every frozen v1 and v2 event;
• each frozen economic-event tuple (symbol, filed, deal_price_usd, transaction_role);
• the originating cohort and cohort ID.

An evaluation candidate is excluded if any source accession appears in that ledger or its economic-event tuple matches a frozen tuple. A regression test must add a lexically earlier accession to a synthetic already-viewed event and prove that the event remains excluded.

This amendment repairs time binding and disjointness only. It does not alter direction, entry, exit, friction, sample minimum, or pass/fail criteria, and no evaluation outcome existed when it was added.