Derivative Forge v0.2.2

Derivative Forge v0.2.2

Release date: 2026-08-28

Summary

This patch release hardens blind benchmark production before the first Blind V8 execution. It does not expand Forge's domain surface or report new benchmark results.

Changes

• Missing or truncated structured model output is rejected inside the existing bounded production attempts instead of aborting the full transaction immediately.
• Requirement and oracle generation use the same fail-closed retry semantics, including their independent review steps.
• Missing provider text is represented by a typed MissingTextOutputError.
• Unrelated generator, configuration, and runtime failures propagate unchanged and are never reclassified by message text.
• Exhausted production attempts still publish no partial blind bundle.

Verification

• Full local suite: 478 passed, 2 skipped.
• Integrated blind producer/provider suite: 73 passed.
• Both GitHub Actions test-and-gate runs passed.
• Minimal runtime gates, CodeQL, and CodeRabbit passed.

Scope

Blind V7 remains a regression corpus. Blind V8 has not been generated or executed as part of this release. The v0.2.2 tag identifies the immutable implementation baseline that must be recorded by any subsequently frozen V8 bundle.

Derivative Forge v0.2.1

Derivative Forge v0.2.1

Release date: 2026-08-22

Summary

This patch release hardens two execution-critical boundaries without expanding Forge's supported domain surface.

Changes

• Candidate Compiler repairs are now compared against the last safe candidate state before selection.
• Regressive, incomplete, or omission-producing repair attempts are rejected instead of replacing safer code.
• Candidate files, semantic preflight evidence, routing phase, and active paths are restored atomically.
• Repair evidence records baseline quality, selected quality, regression attempts, and routing decisions.
• Native Qiskit probe execution is removed from the substrate worker pool.
• The Qiskit circuit probe is serialized and cached while the other cognitive lenses remain parallel.
• Deterministic concurrency tests cover native probe serialization and caller-thread execution.

Verification

• Full local suite: 475 passed, 2 skipped.
• PlannerStage regression module: 25 passed.
• Qiskit/lens/substrate focused tests: 8 passed.
• Repeated real substrate stress: 300 successful decompositions.
• Both independent GitHub Actions test-and-gate runs passed.
• CodeQL and CodeRabbit passed with no review findings.

Scope

This release does not add new synthesis domains or change the frozen Blind V7 evidence. V7 remains a regression corpus after execution. Generality must be evaluated on a newly frozen, independently prepared blind bundle.

See the v0.2.0 release notes and BENCHMARK_EVIDENCE.md for the broader architecture, limitations, and benchmark receipts.

Derivative Forge v0.2.0

Derivative Forge v0.2.0

Release date: 2026-08-22

Summary

This release turns Forge from an initial typed pipeline into an evidence-grounded, fail-closed software synthesis system for greenfield Python artifacts.

Forge compiles natural-language requirements into traceable contracts, plans through the Derivative substrate, generates deterministic or model-backed candidates, validates them in an isolated environment, performs bounded evidence-targeted repair, and packages only fully validated artifacts.

Highlights

• Atomic requirement preservation from source text through assertions and validation evidence.
• Typed acceptance, obligation, public import, implementation blueprint, and quality contracts.
• Deterministic CLI, REST service, data pipeline, and library adapters.
• Composable capability renderers with provenance and manifest enforcement.
• OpenAI-backed complete-candidate compilation for uncovered plans in hybrid and remote-only modes.
• Three independent validation layers: runtime, contracts, and adversarial checks.
• Requirement-level semantic assertion evidence and stronger anti-stub detection.
• Bounded repair driven by observed failure signatures and artifact deltas.
• Docker execution policy with no network, read-only root, bounded resources, and secret isolation.
• Fail-closed packaging with explicit verified, validation_failed, and infeasible_proven terminal states.
• Oracle contract preflight that rejects incoherent external harnesses before Forge or model execution.
• Capability-specific dependency profiles and lazy scientific/model runtime loading.
• A reorganized README, technical reference, and benchmark evidence ledger.

Evidence Checkpoint

The release repository passes 470 tests with 2 skips locally and is gated by minimal-runtime and full Docker-backed GitHub Actions jobs.

Blind V7 remains an honest limitation, not a marketing score:

• frozen baseline external Verified@1: 0/6 (0.000);
• frozen baseline legacy false-verified field: 0.000 with denominator 0; schema-v3 adjudicated value null;
• targeted post-fix replay: 0/5 externally accepted;
• one targeted oracle was independently invalid;
• four cases remained genuine fail-closed candidate-generation failures.

Known V5, V6, and V7 requirements are regression corpora after first execution. Generality must be measured on a newly frozen schema-v3 blind bundle.

See Benchmark Evidence for immutable receipts, hashes, denominators, and replay labels.

Scope

This release supports greenfield Python CLI, service, pipeline, and library synthesis.

It does not claim:

• arbitrary existing-repository modification;
• arbitrary language or frontend generation;
• formal verification of general software;
• proof of universal properties beyond implemented contracts;
• wheel/container distribution, SBOM, or supply-chain attestation.

Upgrade Notes

The minimal deterministic host is installed with:

python -m pip install -r requirements/forge.txt

Model-backed compilation and repair require:

python -m pip install -r requirements/model.txt

Production verification requires the sandbox image:

docker build --file Dockerfile.forge-sandbox --tag derivative-forge-sandbox:py311 .

requirements.txt remains a compatibility alias for the complete requirements/all.txt environment.

Forge Evidence Closure - Blind V5

Frozen research snapshot for the Blind V5 evidence-closure phase.

Verification:

• commit: 02765e4
• local suite: 407 passed, 2 skipped
• GitHub Actions test-and-gate: https://github.com/Daniele-Cangi/Derivative/actions/runs/32374585970
• Docker sandbox build, full suite, and 30-case internal regression gate passed

Evidence semantics:

• the immutable Blind V5 baseline remains 4/12 with status accuracy 0.417, external false-verified rate 1.000, and infeasibility detection 0.333
• post-fix receipts remain revision-specific regression evidence and are not combined into a synthetic current score
• externally accepted V5-002, V5-005, and V5-006 receipts are documented
• V5-001, V5-003, and V5-004 oracle-invalid adjudications are documented without modifying frozen inputs
• baseline false-positive incidents remain preserved
• V5-006 passed its external oracle 9/9 after two repairs, not at Verified@1

This release freezes V5 as a known regression corpus. The next blind must be newly authored, hidden from Forge during development, and frozen before execution. See docs/FORGE_V5_EVIDENCE_CLOSURE.md and docs/DERIVATIVE_FORGE_ARCHITECTURE.md for the evidence and architecture boundaries.

Derivative Forge v0.1.0

Derivative Forge v0.1.0

This is the first public release of Derivative Forge: an execution-grounded Python system that turns natural-language requirements into verified greenfield code artifacts or a formal infeasibility result.

What is included

• A typed Forge pipeline: requirement compilation, planning, code generation, three-layer validation, repair, and packaging.
• Atomic requirement preservation and traceability from source requirement to plan, generated files, tests, and validation evidence.
• Fail-closed terminal states: verified, validation_failed, and infeasible_proven.
• Syntax/import/run checks, obligation and acceptance checks, and adversarial validation against superficial implementations and tests.
• Quality contracts for security, persistence, rate limiting, audit trails, observability, and integration testing.
• Isolated execution through a Docker sandbox with controlled filesystem, network, timeout, and resource policies.
• Deterministic capability profiles for supported Python CLI, service, pipeline, and algorithmic build classes.
• Optional OpenAI-backed candidate compilation with deterministic validation and repair boundaries.
• The underlying Derivative reasoning substrate for symbolic, probabilistic, topological, circuit-backed, and contradiction-detection workloads.

Verification baseline

The release commit passed the complete GitHub Actions gate:

• 346 tests passed.
• Status accuracy: 1.000.
• Verified@1: 1.000.
• False-verified rate: 0.000.
• Infeasible detection rate: 1.000.
• Benchmark execution used the Docker sandbox backend.

Current boundaries

verified means the generated artifact satisfied its preserved requirements, declared obligations, executable tests, and adversarial checks inside the controlled validation environment. It is not a claim of universal correctness, production certification, or support for requirements outside the implemented capability and candidate-compilation boundaries.

The current Forge release targets greenfield Python artifacts. Existing-repository modification, additional languages, frontend generation, wheel/container product packaging, SBOM generation, and dependency auditing remain future work.

Entry points

python forge.py "Build a Python CLI that reads a CSV and includes tests."
python derivative.py "<computational problem>"

See the repository README for setup, architecture, validation semantics, supported surfaces, and benchmark evidence.