Sparse Alpha Compute Governor · Live Pilot
Alpha Compute Governor
A pre-simulation decision layer: evaluate alpha candidates for duplicate risk, self-correlation risk, novelty, structural quality and compute justification before spending full simulation resources.
Demo concept powered by Sparse Alpha Compute Governor · sealed primitive-backed advisory layer.
Projected pilot reduction
—
Estimated share of poor or duplicate candidates that could be held before full simulation during historical replay.
Candidates checked
—
Live pilot metrics
Full sims avoided
—
Advisory hold / reject
Avg sparse eval
—
From live receipts
False hard rejects
0
Pilot is advisory only
Candidate alpha
Paste or load an alpha candidate. Evaluations call the live SACG API (Lane A tokenised).
Governor decision
Decision, scores, reason codes and receipt are generated before full simulation.
TokeniseOperators, fields, lookbacks, groups
SignatureSealed sparse primitive
CompareNovelty and similarity index
GateRisk and cost envelope
ReceiptAuditable decision record
Decision
—
Paste API key, click Connect, then Evaluate.
Novelty—
Correlation risk—
Turnover risk—
Cost justification—
Extracted candidate profile
Family—
Operators—
Lookback—
Signature—
Modeadvisory
Compute impact proxy
Sparse eval—
Proxy energy—
Full sim avoided—
Receipt hash—
Policysacg_v0.1
Signed decision receipt
Safe pilot receipt. It does not expose sparse positions, salt, or primitive internals.
Waiting for evaluation…
Pilot replay summary
Example historical replay results for a lighthouse pilot report.
Replay verdict
Self-corr fails caught1,824 / 2,140
Catch rate85.2%
False holds on pass4.8%
False hard rejects0%
Estimated sim reduction38.0%
| Decision | Count | Meaning |
|---|---|---|
| FULL_SIM | 6,108 | Novel enough to run |
| MICRO_TEST | 1,630 | Try cheaper test first |
| HOLD | 4,502 | Revise before full sim |
| REJECT | 240 | Exact duplicate only |
| Candidate | Family | Decision | Novelty | Corr Risk | Reason | Receipt |
|---|