DSG Governed Self‑Evolution

Interactive 3D view of the deterministic search and governance path. Simulation may run frequently; only measurably better candidates can enter review, and this repository never becomes its own promotion authority.

Code checkpoint CI: PASSCandidate: SIMULATION_ONLYPromotion: DSG_CONTROL_PLANESelf promotion: forbiddenProduction provider: system-level UNBOUND

1 · Technology Intake

Allowlisted official feeds → normalized observation → deterministic adaptation score.

ACTIVE DAILY · 02:23 UTC

2 · Simulation Search

Run bounded deterministic evolution repeatedly over approved parameter/capability space.

HOURLY POLICY · ACTIVATION PENDING

3 · Admission Gate

Composite must improve, efficiency must improve, protected metrics must not regress.

IMPLEMENTED + CI VERIFIED

4 · Candidate Pool

Only independently testable winners enter the pool. Rejected experiments remain evidence.

POOL POLICY CONFIGURED

5 · Batch Review

Compare admitted candidates; remove dominated or conflicting proposals.

TWICE-DAILY POLICY

6 · Tests / Build / Security

TypeScript, Rust, WASM, integration, deterministic and contract checks.

CI VERIFIED

7 · Cinema Raw Proof

Independent verifier binds raw metric/test/build evidence to the exact candidate commit.

RUNTIME BINDING REQUIRED

8 · Control Plane Gate

Canonical authority evaluates plan, scope, constraints, proof, metrics and commit binding.

EXTERNAL ALLOW / BLOCK

9 · Batch Merge

Only coherent ALLOWed set may merge. Merge event alone is not deployment authority.

PROMOTION RECEIPT REQUIRED

10 · One Deploy / Batch

Bound provider adapter executes once after the reviewed aggregate is merged.

BLOCKED WHILE PROVIDER UNBOUND

11 · Health + Canary

Readiness and canary workloads measure the combined deployed result before baseline promotion.

LIVE INTEGRATION REQUIRED

12 · Monitoring → Feedback

Regression → rollback. Improvement → new baseline → feed measured production evidence into the next simulation cycle.

TARGET CLOSED LOOP
SimulationProofControlMerge/DeployMonitoring/Feedback

Candidate admission invariants

Must improve

candidate.composite > baseline.composite and at least one efficiency dimension improves: latency, throughput or cost efficiency.

Must not regress

Latency, error rate, throughput, cost efficiency, audit completeness, human override rate, SLA compliance and zero-data-loss are protected.

Experiment ≠ candidate

A rejected result may be preserved as evidence but cannot enter review, merge or deploy.

Candidate ≠ authority

Every candidate stays SIMULATION_ONLY. Cinema verifies; Control Plane promotes.

Function status

FunctionStatusMeaning
Exact QUBO / Ising shard solverCI VERIFIEDInteger-only deterministic finite search and independently checkable witness.
Deterministic evolution simulatorCI VERIFIEDBounded deterministic genome search.
Technology evolution intakeACTIVEDefault-branch read-only daily GitHub Action, evidence artifacts only.
Candidate admission gateCI VERIFIEDOnly measurable improvement without protected regression is admitted.
Hourly search cadenceCONFIGURED POLICYDeclared in evolution-cadence config; active schedule must be updated before claiming hourly runtime.
Candidate pool / batch reviewCONFIGURED POLICYPool max 24, review max 12, quality-first Pareto target; end-to-end persistent orchestration is pending.
Cinema OIDC raw proofBINDING REQUIREDImplementation fails closed without a live Cinema attestation endpoint.
Control Plane promotionEXTERNAL AUTHORITYThis repository cannot self-promote.
Automatic production deployBLOCKEDRequires approved bound provider, health probe, rollback path, credentials and deployment evidence contract.
Canary / rollback feedbackINTEGRATION REQUIREDRequires live deployment and monitoring bindings.

Event model

Observe

Daily technology evidence refresh. News never mutates production directly.

Explore

Simulation may run much more frequently than review or deployment.

Review

Slower batch review compares multiple admitted winners together.

Deploy

One deployment per approved batch; canary decides rollback or next baseline.