New Additions

Run large pre-registered candidate campaigns (A6+) through a certified declarative engine with per-candidate execution certification, integrity auditing and multiple-testing accounting.

Source

Internal

Original implementation / inspiration

docs/STRATEGY_FACTORY_ARCHITECTURE.md

TradeRep location

lib/strategy-engine/research/strategyFactory/ — missing here: lib/strategy-engine/research/strategyFactory/

Licence / reuse constraints

INTERNAL — First-party code.

Tests performed

• · factory suite under tests/unit/strategy-engine/research/strategyFactory/ (owned by the multi-market lane)

Overlaps with existing functionality

• Strategy Lab campaign runner (lib/strategy-engine/research/strategyLab/campaign/) → BOTH RETAINED — the Strategy Lab carries hand-written A0-A5 strategies through its gates; the Factory carries machine-generated declarative candidates. The Strategy Registry indexes both and labels which engine produced each record.

Rollback: Campaign artifacts are frozen on disk and remain readable even if the runner is removed; the registry reads the artifacts, not the runner.

Provide strictly-before-Discovery data that no pre-Holdout gate can see, handed out one chronological slice at a time.

Source

Internal

Original implementation / inspiration

research-output/strategy-lab/sealed-holdout-bank/

TradeRep location

research-output/strategy-lab/sealed-holdout-bank/, lib/strategy-engine/research/strategyLab/campaign/holdoutGate.ts

Licence / reuse constraints

INTERNAL — Derived market data, not third-party code.

Tests performed

• · tests/unit/registry/holdoutPolicy.test.ts (surface honesty)
• · campaign-runner gate tests under tests/unit/strategy-engine/

Overlaps with existing functionality

• per-campaign forward Holdout partition → BOTH RETAINED — the sealed bank is pre-Discovery data; the forward partition withholds post-boundary bars inside a campaign. They answer different questions.

Rollback: Not rollbackable in any meaningful sense: consumed slices are permanently spent. This entry exists so that fact stays visible.

Long historical replays exceeded the serverless request ceiling; the durable worker runs the same runBacktestJob on a GitHub-hosted runner with cooperative cancellation, TTL advisory locks and bounded re-dispatch.

Source

Internal

Original implementation / inspiration

lib/backtest/jobs/* + .github/workflows/backtest-job-worker.yml

TradeRep location

lib/backtest/jobs/dispatch.ts, lib/backtest/jobs/runBacktestJob.ts, lib/backtest/jobs/recovery.ts

Licence / reuse constraints

INTERNAL — First-party code.

Tests performed

• · tests/unit/backtest/jobRecovery.test.ts
• · tests/unit/backtest/runBacktestJobCancellation.test.ts
• · tests/unit/backtest/progressPersistence.test.ts

Overlaps with existing functionality

• in-process backtest execution → RETAINED as the fallback when no worker token is configured (local dev); identical compute path.

Rollback: Unsetting BACKTEST_WORKER_GITHUB_TOKEN/BACKTEST_WORKER_REPO falls back to in-process execution with no code change.

Collect genuinely prospective evidence on frozen specifications, with no order execution.

Source

Internal

Original implementation / inspiration

lib/strategy-engine/research/{family6Shadow,family2ArmCShadow,a2Dot10Shadow,pbEliteOutcome}/

TradeRep location

wired into lib/live/webhookProcessor.ts; read back via /api/live/f6-tally and /api/live/a2dot10-tally

Licence / reuse constraints

INTERNAL — First-party code.

Tests performed

• · tests/phase4/07_a2dot10_shadow.test.ts
• · tests/integration/webhookProcessor.test.ts
• · tests/unit/live/strategyRosterArming.test.ts

Overlaps with existing functionality

• backtest engine → BOTH RETAINED — a shadow runner cannot observe history it did not live through; the backtest engine cannot produce prospective evidence.

Rollback: Each runner is standalone and additive; removing one stops its observation writes and affects no other runner or V1.

Classify live decisions in shadow against the approved candidate architecture without changing V1 behaviour, sizing or risk.

Source

Internal

Original implementation / inspiration

docs/CANDIDATE_ARCHITECTURE_V2.md

TradeRep location

lib/strategy-engine/research/architectureV2/

Licence / reuse constraints

INTERNAL — First-party code.

Tests performed

• · tests/unit/architectureV2/decisionIntegrity.test.ts
• · tests/integration/architectureV2LiveShadowIntegration.test.ts

Overlaps with existing functionality

• AV2 challenger program (research-output/av2-planning/) → DISTINCT PROGRAMS — Architecture V2 is a candidate architecture for V1 setups; AV2 is challenger testing against the F6/A2.10 champions. Kept separate in every surface to prevent misreporting either.

Rollback: Shadow-only and non-executing; disabling the classifier stops classification writes only.