Catalyst impulse, balance and restart
The finding
Falsified on its frozen population. No promotion, no shadow launch, no parameter change and no outcome rescue.
The research question
Does a catalyst-driven impulse followed by balance and a restart carry forward information on a population frozen before any outcome was read?
Decision criteria
The population (84 signals on 63 independent dates) and its manifest hash were committed before outcome access, together with the gates. One bounded outcome trial; no rescue.
Result document
Repository source: research/results/catalyst_impulse_balance_restart_outcome_result_v1_2026_08_13.md
Disposition: FALSIFIED. No promotion, shadow launch, parameter change, or outcome rescue.
Integrity
- Evaluation manifest: 84 signals on 63 independent dates
- Manifest SHA-256:
09f2acd4b22b1438ae5e1a3cef4d3250636c506a8cc681158519b027bdc84fa1 - Outcome contract committed before access in
0bf9f383; complete population committed in166f7514 - Outcome ledger SHA-256:
c3c047af3746d5559b9c6fa233ffd39f13a034e767bf54fa3b7db5a1b7e6e9d0 - Machine summary SHA-256:
79eea12d465fa612e431618f71a5d5bcc4aa1b7c0e76a1b296fa0736622060ed
Primary result
| gate | result | pass |
|---|---|---|
| Independent dates | 63 | yes |
| Mean net daily return | +0.0479% | yes |
| Mean without best date | -0.0293% | no |
| Median net daily return | -0.0545% | no |
| One-sided sign-flip p | 0.428360 | no |
| Largest positive-date contribution | 8.75% | yes |
The conjunction fails three of six gates. The small positive headline mean is not robust to its best date and is not statistically distinguishable from noise. The 40/50 CLEAR identity result proved that this was a coherent visual population; it did not prove expectancy.
Permanent memory
Do not rescue v1 with a different balance duration, drift threshold, stop, cost, time exit, CLEAR-only subset, catalyst subtype, gap/RVOL/liquidity slice, or time-of-day subset. Those are outcome-aware variants of spent evidence. A future candidate must add a new point-in-time mechanism or information variable and use disjoint evidence.
Study specification
Repository source: research/specs/catalyst_impulse_balance_restart_outcome_prereg_v1_2026_08_13.md
Frozen before any post-entry bar or outcome access.
Population and execution
Use all 84 frozen signals from the v1 retrieval, including CLEAR and NEAR cases. Do not filter on the visual label. RGTI 2026-05-21 and OKTA 2026-05-29 remain excluded as identity seeds.
- Entry: open of the five-minute bar immediately after the signal.
- Initial stop: 0.25% below the frozen balance low.
- Failure exit: open of the bar after the first completed five-minute close below the stop.
- Time exit: 15:55 ET close on the signal session if no failure exit occurs first.
- Costs: subtract 20 bps round trip from every trade. This is a historical research convention; paper validation must separately measure realized spread and slippage.
- Risk reject: none. Position sizing and R-multiples are descriptive only; primary returns are equal-notional percentage returns.
Integrity and primary gate
Before outcome extraction, freeze a machine manifest containing symbol, date, balance timestamps, levels, signal timestamp, entry timestamp/open, and its SHA-256. The evaluator must verify that hash before opening a database connection for later bars.
The 50-row review manifest has SHA-256
972cbc3b77bdc7399c906de58b22a6930f2e9fa66789a9f0d8044fa909b0a64e; it is not the evaluation
population. The evaluator remains locked until a separate all-84-signal manifest is frozen.
Evaluation manifest: research/results/catalyst_impulse_balance_restart_all84_signal_manifest_v1_2026_08_13.csv,
84 rows, SHA-256 09f2acd4b22b1438ae5e1a3cef4d3250636c506a8cc681158519b027bdc84fa1.
Primary unit is the independent date: average same-date trade returns first, then evaluate dates. Pass only if all are true:
- at least 60 independent dates remain executable;
- mean net daily return is positive;
- mean net daily return without the best date is positive;
- median net daily return is positive;
- one-sided date-level sign-flip p-value is below 0.05;
- no single date contributes more than 35% of total positive P&L.
Trade-level win rate, stop rate, holding time, CLEAR-versus-NEAR split, catalyst type, time of day, gap, RVOL, liquidity, and alternative exits are descriptive only and cannot rescue failure. A fail ends v1. A pass authorizes a new forward shadow contract only, never paper or live capital.