BLUM 2.3.0 — Verified Alpha Desk

BLUM 2.3 adds a prospective Alpha Desk for equities/ETFs and Forex, evaluated separately. New source-engine decisions register frozen inputs before outcomes; imported history cannot create a forward track record.

• Read-only market panels and a call journal show source hashes, providers, timestamps, cost-qualified closes, missing benchmarks and exclusions.
• Filled time exits and invalidations remain realized outcomes. Unfilled orders stay excluded, older closes stay visible and live-forward rows are not counted twice.
• Realistic fill accounting reconciles explicit cash fees without double-counting spread/slippage. Currency totals remain separate. Paired alpha requires causal benchmark endpoints aligned to the actual holding dates and consistent notional returns.
• A closure-date bootstrap reports uncertainty and minimum coverage. It has no model-promotion authority or multiple-testing correction; confidence remains heuristic.
• Physical base and streamed WAL publish in one recovery object with embedded sizes and SHA-256 checks; extraction streams that object without staging another complete archive. Failed database restores stop startup, including retries with partial tables. Deployments and Cloudflare readiness verify actual protected backend/script content as well as revision and version.

Validation: 905 backend + 17 frontend + 12 gateway tests (934 total), production web build, gateway types/typecheck and unchanged packaged historical acceptance audit. The initial independent review found four functional defects, all reproduced and fixed before merge. A subsequent recovery review found the archive member-size ceiling; it was also reproduced and fixed. No actionable review findings remain.

This is paper research. Bloomberg superiority and market-leading alpha are not established results. The registry begins with newly recorded calls; valid forward performance must accumulate over time. Native Forex paired alpha stays unavailable without the required notional and benchmark contract. Embedded PostgreSQL still uses periodic backups and does not guarantee zero-loss restart recovery. Protected code hashes do not certify model weights, datasets or provider histories.

Live Alpha Desk · Evaluation contract · Reviewed PR

Public deployment verification now retries transient rollout failures at most three times; persistent failures remain blocking. Reviewed workflow fix.

Cloudflare readiness now waits inside the verifier deadline before product snapshots, so transient rollout propagation is retried while persistent revision/hash/readiness failures remain blocking. Recovery and readiness fixes.

The Alpha page retries one eligible transient load failure and offers a manual retry after a persistent failure. Compiled-browser fault cases cover a 503, an actual request abort and manual recovery. Reviewed page recovery.

The physical restore decoder now accepts PostgreSQL concatenated gzip members and validates every gzip trailer. Five archive regressions were reproduced before the correction; actual persisted WAL decoding and focused independent review passed. Decoder correction.

Production verification: exact revision fc371a2, all 20 public HTTP checks passed, and the complete 6.8 GB physical image restored with PostgreSQL ready and no logical fallback. The checkpoint retained all 452 registered calls, 40 checked journal manifests and 12 visible generations. Temporary Forex paper maintenance has ended and its original RUNNING state is restored.

Periodic backup limitations remain observed: aggregate legacy simulation counts decreased after recovery (including 10 paper-game trade rows and 88 learning events). One earlier generation UID is absent from live SQL after the earlier logical fallback; its JSON evidence and previous private image are preserved. These checks do not establish complete row retention or zero-loss recovery. The captured desk has no eligible active calls or qualified paired outcomes yet. See the attached release report and verification bundle for scopes, decisions and evidence.

Exact verified source fc371a2 · Main CI · Successful deployment attempt 2 · Detailed release report · Verification bundle

BLUM 2.2.0 — verified research and Cloudflare observatory

BLUM 2.2 establishes GitHub as the canonical platform for source, review, CI, documentation and releases. Verified main revisions publish one-way to the existing Hugging Face Engine runtime and the Cloudflare observatory.

The complete product includes Brain, Training Ground, Paper Trading, Alpha and the new Evolution page. Public APIs are read-only at both the gateway and origin; administrative research requires a private runtime key. Backend and gateway report the deployed source revision, and release verification checks that they match.

ASCENT runs bounded calibration experiments with restricted arithmetic candidates, chronological selection, independent holdout gates, persistent lineage and explicit resource limits. It retains research artifacts without automatically replacing a production champion, merging code, starting paid GPU jobs or placing broker orders.

Historical data decisions are reproducible: 75,381 valid equity bars are admitted, five malformed bars excluded, eight missing-volume observations remain missing. Forex conflicts and unsupported weekend observations are excluded, leaving one causal example—insufficient for validation. The public reasoning data and all 95 archived snapshot versions are quarantined for new training/promotion because provenance and temporal-validity gates fail. Existing database and recovery backups remain preserved; their entire contents have not been certified.

Validation includes 838 backend tests, 9 frontend tests, 12 gateway tests, static product build, TypeScript checks, exact source-hash audit and live deployment verification. A historical diagnostic generated 549 genuine baseline forecasts; validation rejected the candidate for insufficient slice sample, no confirmed calibration gain and robustness regression. This diagnostic is not forward-performance or production-promotion evidence.

Product: https://blum.signallayerlabs.workers.dev

Operations and recovery: Deployment. Data decisions: Data quality. Research boundaries: ASCENT.

Recovery verification retained the existing ASCENT generation and restored the financial database from its physical backup. The post-restart count comparison shows 20 fewer recent historical predictions and 60 fewer prediction outcomes than the pre-restart observation. Embedded PostgreSQL uses periodic backups and does not provide zero-loss container restarts; strict durability requires external PostgreSQL. This comparison is a count audit, not certification of every row.