A2.10 — Final-Hour Reversion v2

a2 / a2-final-hour-reversion-v2 @ a2-v1 · code b29299a

campaign strategyshadow runner armedLIVE SHADOW

A2Dot10ShadowRunner is deployed, wired into lib/live/webhookProcessor.ts, and ARMED at its verified production true-live boundary (2026-09-16T21:12:31.000Z). Only signals entering at or after that instant count toward the prospective sample -- see /api/live/a2dot10-tally.

Lifecycle

current

Outcome

—

Phase

Phase B

Research gate

FINAL_AUDIT

Engine

strategy lab

Market

MNQ

Timeframe

MNQ 1-minute OHLC

Side

BOTH

Tracking

Family, lineage and what this record needs next

Mechanism family

Session / time-of-day

a2-final-hour-reversion-v2

Mechanism parent

finalHourReversion

Phase detail

Prospective live-shadow evaluation, runner armed. No order execution.

Next required test

Continue accumulating prospective live-shadow observations toward its declared gate.

Integrity & certification

Signals as recorded; NOT_PASSED means not passed or not applicable

Engine certified

unknown

Execution certified

unknown

Integrity audit

not passed

Shadow verifier

unknown

Technical invalidation

None recorded.

Note

Holdout expectancy diverges materially from Validation's own -- treat as a fresh, not fully explained, open question.

Artifacts

Everything committed for this record

Campaign record (A2)

research-output/strategy-lab/campaigns/a2/campaign_state.json

Holdout artifacts

research-output/strategy-lab/campaigns/a2

Stage ladder

Each cell carries the campaign runner's own classification label — hover for the verdict and its reason

Intake

passed

Pre-reg

passed

Lightning

passed

Robust

passed

Exec/stat

passed

Validation

passed

H-eligible

passed

Holdout

passed

Audit

inconclusive

Closed

inconclusive

INTAKE

Registered into campaign A2 at code version b29299a.

PREREGISTRATION

Pre-test passed: distinct, causal, executable, required data available.

LIGHTNING

SURVIVED_LIGHTNING_SCREENAggregate expectancy 0.3016R (PF=1.708) over 451 resolved trades, 3/4 qualifying years (n>=15) agree with the aggregate sign, top-10 winner concentration 14.7% (below the 50% concentration flag). Survives the Lightning Screen -- NOT validated.

ROBUSTNESS

ROBUSTNESS_SURVIVEDNo single-year dependence, no severe winner concentration (no preregistered neighborhood declared).

EXECUTION STAT

EXECUTION_STAT_SURVIVEDBootstrap CI excludes zero and the edge survives up to 0.5pt adverse slippage.

VALIDATION

VALIDATION_SURVIVEDValidation expectancy 0.4091R holds sign and reasonable magnitude relative to Discovery's 0.2661R.

HOLDOUT ELIGIBILITY

HOLDOUT_ELIGIBLESurvived Lightning Screen, the Robustness Gauntlet, the Execution+Statistical Gate, and the Validation/Walk-Forward Gate against 357 Discovery and 88 Validation trades.

HOLDOUT

HOLDOUT_OPENED_ONCEDiscovery: n=357, expectancy=0.2661R, PF=1.609. Validation: n=88, expectancy=0.4091R, PF=2.029. Holdout (this run, sealed bank immediately after A12's consumed slice, stop reason=FORMAL_TARGET_100): n=100, expectancy=0.2600R, PF=1.553.

FINAL AUDIT

a2-final-hour-reversion-v2@a2-v1: economic credibility=true, statistical credibility=true, executable=true, stable=true, distinct=true. 1 unresolved risk(s) flagged.

CLOSED

RESERVE_RESEARCH_VERSION_REQUIREDSurvived Holdout but the Final Research Audit flagged unresolved risk(s): Holdout expectancy diverges materially from Validation's own -- treat as a fresh, not fully explained, open question..

Evidence recorded

Quoted from the campaign record — never recomputed here

Screen n (FIXED_2R)

451

Expectancy

+0.302R

Profit factor

1.71

Win rate

57.4%

Total R

136

Max DD

13R

Longest L streak

Trades / year

128.4

Split	Window	n	Expectancy	PF	Max DD
Discovery	2022-06-03..2025-03-31	357	+0.266R	1.61	13R
Validation	2025-04-01..2025-12-31	88	+0.409R	2.03	3R
Holdout	sealed bank slice	100	+0.260R	1.55	7R

Discovery → Validation

Expectancy degradation -0.143R (positive means Validation came in weaker than Discovery).

Bootstrap CI

95% CI +0.169R .. +0.441R · P(expectancy > 0) = 100%

Winner concentration

Top 5 7.4% · top 10 14.7% · top 20 29.4% of total R.

Leave-one-year-out

No sign flip when any single year is removed.

Holdout vs prior evidence

Discovery: n=357, expectancy=0.2661R, PF=1.609. Validation: n=88, expectancy=0.4091R, PF=2.029. Holdout (this run, sealed bank immediately after A12's consumed slice, stop reason=FORMAL_TARGET_100): n=100, expectancy=0.2600R, PF=1.553.

Read from research-output/strategy-lab/campaigns/a2/campaign_state.json

Holdout position

One-shot, sealed, never retried

Eligible

Yes — the campaign recorded Holdout eligibility.

Authorised

Yes, 2026-09-16 02:39Z

Authorised by

AJ <meatham07@gmail.com> -- explicit chat authorization ('AJ APPROVAL -- A2.10 UNTOUCHED HOLDOUT'), trade-count-based deterministic Holdout consumption policy, 2026-09-16

Specification frozen

Fully frozen before the Holdout was opened.

Sealed data consumed

Yes — this Holdout is spent and can never be re-run.

Slice consumed

2019-05-21 11:26Z → 2020-04-20 19:05Z · 315,910 bars · 100 trades · stop reason FORMAL_TARGET_100

Preserved unused

749,233 sealed bars left for later strategies.

Validation data consumed

Yes.

Read from research-output/strategy-lab/campaigns/a2/A2_10_HOLDOUT_LEDGER.json · research-output/strategy-lab/campaigns/a2/A2_10_HOLDOUT_RESULT_REPORT.json

Lineage

Declared at intake, carrying any prior negative result

Relation to prior work

experimental descendant

Parent

finalHourReversion

Note

Adds a cross-day realized-volatility-regime condition (today's RTH range-so-far vs a trailing 20-day baseline of the same statistic) that FLIPS the parent's unconditional fade to continuation on high-volatility days. This is a genuine new causal mechanism (a regime-conditional bidirectional model), not a different threshold/stop/target/window/long-short filter on the same fade -- the parent's own fade logic is reused verbatim for the non-flipped branch.

Bot testers on this strategy

Automated harnesses producing its evidence

armed

A2.10 Final-Hour Reversion v2 shadow runnerlib/strategy-engine/research/a2Dot10Shadow/runner.ts

Frozen specification

Snapshotted at intake — not re-read from a live module

Mechanism hypothesis

Late-session range-extreme mean-reversion (the parent Final-Hour Reversion thesis) only holds on calm days; on days whose own realized range-so-far is unusually large relative to recent history, the late-session extreme instead tends to continue rather than revert.

Long rule

At the first RTH bar at/after 15:00 ET, score that bar's close against the day's own prior RTH high/low (from 09:30, strictly before this bar). Compute ratio = (today's prior RTH range) / (trailing 20-day average of that same statistic, prior days only). If ratio >= 1.5: trade CONTINUATION (long if close is near the range top, short if near the bottom). Otherwise: trade the ORIGINAL fade (short if near the top, long if near the bottom) -- identical to the parent Final-Hour Reversion engine. No trade if fewer than 20 prior days are recorded yet.

Short rule

Symmetric direction mapping as described above.

Entry trigger

The first RTH bar at/after 15:00 ET whose close is within 0.15 of the day's own prior RTH high or low, with direction determined by the realized-volatility-regime ratio described above.

Earliest executable entry

Next 1-minute bar's open after the triggering bar closes.

Initial stop

The day's own prior RTH extreme relevant to the trade direction (already fully known at trigger time) -- identical to the parent.

Exit

FIXED_2R only (this A2 batch's declared scope), via the shared TradeTracker; 5-day max-hold expiry as a backstop.

Causality

priorRthHigh/priorRthLow only accumulate from bars strictly before the current one (identical to the parent). The trailing-day baseline only ever includes days strictly before today, and today's own value is never included in its own baseline. Entry is deferred to the next 1-minute bar's open after the trigger bar closes. No same-bar entry.

Expected frequency

LOW

Required data

MNQ 1-minute OHLC

Campaign data window

MNQ 2022-06-03..2026-09-15

Parameters

extremeThreshold = 0.85 (number) — Range-position threshold for a bar to count as near an extreme (identical to the parent).realizedVolLookbackDays = 20 (integer) — Trading days of trailing baseline for today's own RTH-range-so-far statistic.highVolRatioThreshold = 1.5 (number) — Ratio at/above which the economic thesis flips from fade to continuation.

Read from research-output/strategy-lab/campaigns/a2/campaign_state.json

History

Every dated event this record appears in, newest first

2026-09-16

Holdout result recorded for A2.10RESERVE_RESEARCH_VERSION_REQUIRED

100 trades consumed · stop reason FORMAL_TARGET_100 · 749,233 sealed bars preserved unused · Survived Holdout but the Final Research Audit flagged unresolved risk(s): Holdout expectancy diverges materially from Validation's own -- treat as a fresh, not fully explained, open question..

research-output/strategy-lab/campaigns/a2/A2_10_HOLDOUT_RESULT_REPORT.json

2026-09-16

One-shot sealed Holdout authorised for a2-final-hour-reversion-v2

AJ <meatham07@gmail.com> -- explicit chat authorization ('AJ APPROVAL -- A2.10 UNTOUCHED HOLDOUT'), trade-count-based deterministic Holdout consumption policy, 2026-09-16

research-output/strategy-lab/campaigns/a2/campaign_state.json

2026-09-16

Sealed Holdout slice assigned to A2.10

Evaluated chronologically forward from 2019-05-21T11:26:00.000Z, frozen before evaluation.

research-output/strategy-lab/campaigns/a2/A2_10_HOLDOUT_LEDGER.json

2026-09-16

a2-final-hour-reversion-v2@a2-v1 registeredACTIVE_RESEARCH · HOLDOUT

[a2] Holdout-eligible but awaiting explicit human authorization -- automatic routing stops here per spec Sec. A#.6, never opens Holdout on its own.

research-output/strategy-lab/experiment_registry.json