Methodology

How evidence is measured

Historical replay, Forward Grade A, and production portfolio results are separate layers. They are never mixed into one scoreboard.

historical-v2-rc2 · RC2_COMPLETE_30TD · HISTORICAL_V2_COVERAGE_LIMITED · e1b790684ce0… · 2026-08-06T12:22:40Z

Historical replay

Decisions are reconstructed in strict as-of order from recovered SEC and market data. Historical replay is not original live publication and is never labelled Grade A.

Forward Grade A

Forward capture timestamps publication, feature state, intended entry, and outcomes as they happen. Capture start 2026-08-05T17:25:12Z. Status: COLLECTING_FORWARD_EVIDENCE. Momentum Shadow start 2026-08-05T16:34:19Z.

As-of clock & T+1

Feature cutoffs use completed bars only. Intended entry is the next US equity session after the filing session. T+1 reconstructability flags are recorded; calendar-session enforcement vs intended T+1 is not proven fail-closed for all campaigns.

Campaign endpoint & path

Completed 30-trading-session campaigns: 4,892, positive completed outcomes 54.97%, median +1.26%. Path thresholds describe observed historical price movement. They are not guaranteed realized exits or broker fills.

Evidence coverage waterfall

4,892 completed 30-trading-session outcomes from a frozen parent universe of 5,517 campaigns.

Parent frozen universe: 5,517 (campaign inventory — not completed results)
Evidence-eligible: 4,943
Complete 30 trading sessions: 4,892
Excluded: 574
Pending 7 / 30 / 90 trading sessions: 19 / 51 / 68
Complete 30td parent coverage: 88.67%

Excluded and pending campaigns remain accounted for separately. Losses included.

Matched baseline & campaign unit

Campaign research is compared against matched baselines in the frozen research registry. Signal-row inventory counts are not independent trades.

Evidence grades & release versioning

Current customer historical performance binds only to historical-v2-rc2 (HISTORICAL_V2_COVERAGE_LIMITED). Historical V1 is FROZEN_INVALID_FOR_PERFORMANCE — retained as a corrected legacy artifact with preserved hashes and methodology history. It must not power current aggregate cards, Archive KPIs, SEO performance descriptions, or customer CTA proof. Related IDs: HISTORICAL_EVIDENCE_V2, MOMENTUM_M3_RESEARCH_V1, FORWARD_GRADE_A_V1.

Independent replay & forward/replay parity

Institutional replay checks include determinism, independent recalculation, and forward/replay schema parity. Replay remains reconstruction-grade.

Production portfolio distinction

Campaign endpoint statistics are not a capacity-constrained portfolio, not live IBKR P&L, and not a production monthly-return claim. Exact production T+1 was not replicated. Performance claim freeze remains enabled where SELL path metrics are under recalculation.

Momentum M3 (research)

Frozen rule: BUY close > MA20 > MA50; SELL close < MA20 < MA50. Historical holdout n=146; holdout median/expectancy suppressed pending SELL recalculation. See /research/momentum-m3. Holdout figures are not expected subscriber return. Momentum Shadow rules unchanged.

AI research context

Insider Edge may attach a structured AI research narrative to eligible signals. The purpose is research narrative only: summarize evidence, note supportive factors, risks, and missing evidence. Deterministic filters and scoring remain authoritative.

Effective AI scoring weight is exactly 0.
AI does not affect VIP eligibility, Public eligibility, ranking, or position authorization.
Outputs are structured with versioned prompts and model configuration.
Source facts and derived metrics are labelled separately from AI interpretation.
Model-reported confidence is uncalibrated and is never shown as a probability of return.
Assessments are timestamped; failures and insufficient-evidence states are shown explicitly.
Historical V2 return metrics and Momentum Shadow outcomes are not live AI inputs.
Legacy free-text narratives do not meet the current provenance standard and are not shown as current assessments.

Selection effect: None — analysis only.

We do not advertise AI selecting signals, AI-selected signals, an AI trading engine, or AI return prediction.