AlphaForge · public research ledger
The ideas we stopped are part of the memory.
Chart Library is the evidence and retrieval layer. AlphaForge is the research operation using it. The Edge Refinery generates hypotheses, audits whether the necessary point-in-time data exists, freezes each test before outcomes, and records terminal failures so the loop cannot rediscover them under new names.
Public snapshot dated 2026-08-16. We publish process, terminal results, and delayed receipts—not active hypotheses, executable rules, positions, or prospective edge. Nothing on this page is a trading recommendation.
The governed loop
Permission narrows as evidence gets more expensive.
01
Generate
Mechanism + public source
02
Qualify
Novelty + source feasibility
03
Freeze
Cohort, estimand, costs, kill gates
04
Test
One bounded outcome trial
05
Validate
Reproduction + forward behavior
06
Deploy
Shadow, paper, then small capital
Loop receipt · completed tasks through 2026-08-16
Task counts are operational throughput, not independent discoveries and not evidence of edge.
Terminal ledger
Recent hypotheses that did not earn another stage.
Issuer buyback-blackout reopening
2026-08-16 · outcomes never opened
Frozen question
Whether public SEC exhibits could identify formulaic issuer repurchase blackouts that reopen after earnings, without relying on private policy data or reading returns.
Disposition
The source census covered every reviewed buyback issuer but found no qualifying dates under the frozen definition. The hypothesis was stopped before an outcome task existed.
- —155 of 155 reviewed-authorization issuers covered
- —6 candidate independent dates
- —0 qualified dates versus 40 required
- —No return, price-path, order, or position fields accessed
New buyback authorization × published days-to-cover
2026-08-15 · outcomes opened once under frozen rules
Frozen question
A frozen high-days-to-cover interaction around source-reviewed new buyback authorizations, matched on date and input-state fields before prices were opened.
Disposition
The attractive headline effect was rejected. It was underpowered, confidence bounds crossed zero, and a large pre-event placebo suggested the move was already underway before the filing.
- —19 high-DTC events across 19 dates
- —+2.53 percentage-point apparent interaction
- —Permutation p = 0.091; date- and issuer-cluster lower bounds below zero
- —+1.65 percentage-point pre-event placebo; lineage retired without opening 2026
Filing OCF transition × prior-session short-flow crossing
2026-08-16 · outcomes opened once under frozen rules
Frozen question
A preregistered interaction between a filing-based operating-cash-flow transition and a lagged short-flow crossing.
Disposition
The frozen discovery evaluation failed permutation, clustered-inference, and pretrend gates. The semantic lineage cannot be renamed and retried as a fresh idea.
- —Seed-label and crossing-label permutation gates failed
- —Date- and issuer-cluster bootstrap gates failed
- —Pretrend magnitude gate failed
- —A deterministic semantic duplicate was detected and terminated
Legacy fundamental-factor deciles
2026-08-15 · outcomes opened once under frozen rules
Frozen question
The older family of monthly liquid-US fundamental ranks, deciles, rebalance frequencies, and horizons.
Disposition
Historical evidence and implementation-cost gates did not support the family. Rephrasing the same search over another rank or horizon is not treated as novelty.
- —Historical gate failed
- —Cost gate failed
- —Family-level retirement prevents cosmetic retries
- —No promotion to shadow or capital
Published
Frozen questions, source sufficiency, sample gates, outcome-access status, terminal verdicts, negative results, and enough methodology to understand why a candidate stopped.
Embargoed
Active candidates, surviving rules, exact execution logic, current positions, and anything whose disclosure could contaminate forward validation or transfer prospective edge.
Separate evidence records