CarbonFactorResolver v0.14.2

CarbonFactorResolver v0.14.2

Portfolio-ready, reproducible research prototype. This is not a production carbon-accounting system.

Release commit: 5155a6829fcdd521f04359263f543245a1c6b03f
CI: https://github.com/kamjcx/CarbonFactorResolver/actions/runs/33658119864

Python 3.11/3.12/3.13: 360 passed on each; Python 3.11 branch coverage 87.15%. Ruff, mypy, compileall, package/archive isolation, FactorBench V3, Frozen Unit, Sealed v2/v4, Portfolio, BYOC contracts, container build and /healthz passed.

Includes a 20-record PUBLIC_SYNTHETIC BYOC example and direct, MORE_INPUT_NEEDED, and safe-refusal demos. No ecoinvent, customer, licensed catalogue, credential, or formal approved factor database is included. Synthetic values are not valid for accounting. Human review and immutable locking remain separate governed actions.

Autonomous Evaluation V1 raw evidence retains six frozen geography/year label disagreements under the published USABLE_WITH_ASSUMPTIONS adjudication; raw evidence was not rewritten. The 10k/50k performance attachment is descriptive synthetic evidence, not a production SLA.

Docker digest: sha256:7dc24bf490d63ff155265c7b859f1f40ebdb47a71c062949090adab4f0034646

See the tagged release notes, release-readiness report, and BYOC tutorial.

CarbonFactorResolver v0.14.1

CarbonFactorResolver v0.14.1 — Acceptance Closure

v0.14.1 closes the unit-system and FIN-05 acceptance work without changing any frozen
benchmark answer or formally admitting deployment factor records.

Runtime corrections

• Conditioned Nm3 requests remain in their stated reference state during normalization,
  allowing source-factor evidence to be applied in its declared direction.
• Every ambient/conditioned volume crossing requires explicit versioned evidence; scaled
  ambient units such as L cannot silently convert to or from Nm3.
• Otherwise-qualified candidates under an unresolved material-subtype request are exposed
  only as REFERENCE_ONLY. The terminal decision remains MORE_INPUT_NEEDED and standard
  approval cannot select them.

Preserved evaluation history

• Sealed Unit v2: first run 31/32; unchanged post-fix regression 32/32.
• Sealed Unit v3: first run 23/24 due to an incorrect frozen expected value; NO-GO retained.
• Sealed Unit v4: independent first run 21/21, with every reported check at 100%.
• Portfolio FIN-05: correct source preserved as reference-only while subtype input remains
  required; the frozen benchmark answer is unchanged.

This remains a portfolio-ready research prototype, not a production accounting system or
formal factor catalogue. Structured electricity records without complete canonical source
evidence remain blocked.

Verification summary

• Core package: 324 tests passed with 87.06% branch coverage on Python 3.11.
• Compatibility: 324 tests passed on both Python 3.12 and Python 3.13.
• Offline acceptance harness: 13 tests passed with 43.24% branch coverage.
• Evaluation critical path: 6 tests passed with 84.10% branch coverage.
• FactorBench v1/v2/v3, frozen unit v1, sealed unit v2 and sealed unit v4 passed.
• Ruff, mypy, source compilation, package build and 0.14.1 archive isolation passed.
• Container release eligibility is decided by the protected remote CI container job.

Reproducible release assets

The attached wheel and source archive were rebuilt from tagged commit 19485ee2b915fc8b721e2b549212adb537e24c44 and passed archive-isolation checks. The Docker digest comes from the successful tag CI run. Sealed Unit v4 includes its frozen query set, synthetic catalog, and preserved first-run result; SHA256SUMS.txt and release_manifest.json bind the artifacts to their hashes and source commit.

CarbonFactorResolver v0.14.0

CarbonFactorResolver v0.14.0

Stable portfolio release of the structured carbon-factor retrieval and qualification engine.

This release is runtime-equivalent to v0.14.0-rc.6: it changes version metadata and public documentation only. It does not change resolver runtime behavior, ranking, qualification, synthetic fixtures, frozen answers, or sealed evidence. RC3-RC5 remain preserved historical NO-GO results; RC6 is the unchanged sealed GO evidence.

Verified results

• Core package: 315 passed; 87.05% branch coverage
• Evaluator critical path: 84.10% branch coverage
• Developer-only offline harness: 13 passed; 43.24% branch coverage (reported separately)
• FactorBench V3: 57 cases; contract metrics passed
• Frozen Unit Regression Set: independent first run 24/28; post-fix regression 28/28
• Closed Portfolio Benchmark: 39/40 answerable Top-1; zero boundary and subject violations
• RC6 sealed first run: 48/48 full case contracts; 100% Top-1, recall, abstention, and deterministic replay; zero boundary, subject, unit, forbidden-candidate, and HTTP-500 failures
• Required GitHub test and container checks: passed on final commit 58ae427f763f3b7f708b9c396f75819aadead655

Scope and data boundary

CFR accepts structured resolution requests and returns explainable candidates, MORE_INPUT, or safe refusal for human review/locking. It does not parse documents, calculate full product footprints, generate reports, or auto-approve formal factors.

The repository and release contain only project-authored public-synthetic fixtures. No ecoinvent database, customer evidence, private credentials, or formal factor catalogue is included. Code licensing and factor-data licensing are independent; see DATA_LICENSE.md.

This is a portfolio-ready, reproducible research prototype—not a production carbon-accounting system and not a universal real-world accuracy claim.

CarbonFactorResolver v0.14.0-rc.6

CarbonFactorResolver v0.14.0-rc.6

Portfolio-ready, reproducible research prototype for structured carbon-factor retrieval and deterministic qualification. It is not a production carbon-accounting system and does not include or approve licensed/formal factor data.

RC6 sealed first run

• 48/48 full frozen case contracts
• Answerable Top-1: 100%
• Retrieval recall before gate: 100%
• Abstention correctness: 100%
• Deterministic replay: 100%
• Boundary / subject / unit-dimension violations: 0 / 0 / 0
• Forbidden candidate escapes: 0
• Unhandled HTTP 500: 0

RC3, RC4 and RC5 failures remain preserved in the repository. No failed answer set or raw first-run output was rewritten.

Engineering gates

• Core: 315 passed, 87.05% branch coverage
• Offline true-data harness: 13 passed, 43% branch coverage (separate developer-only metric)
• Sealed evaluator: 10 passed, 84% branch coverage
• FactorBench V3: 57 cases, all published contract metrics 100%
• Frozen Unit Regression: 28/28
• Portfolio Challenge: Full CFR 39/40 Top-1, zero boundary/subject violations
• Remote GitHub test and container jobs: PASS
• Container digest: sha256:45b0c6e847443b7e133064b8b1c6748c654f7b0c87098dc5a1f1c23dbdec1467

Scope and data safety

CFR accepts structured JSON factor requests and structured factor sources. PDF/DOCX handling exists only in a developer-only offline QA harness. The release contains public-synthetic fixtures only; no ecoinvent database, customer evidence, credentials, or formal factor catalogue is included.

v0.14.0-rc.2 — NO-GO evidence

NO-GO evidence release. Remote test/container checks are green and all safety gates passed, but the wholly new sealed evaluation reached 83.33% Answerable Top-1 against a 90% release threshold. Inputs, answers and raw first-run results remain unchanged. Do not promote to v0.14.0. This is a reproducible research-prototype evidence release, not a production carbon-accounting release.

v0.14.0-rc.1 — NO-GO evidence

NO-GO evidence release. The sealed run passed its metric gates, but the remote container workflow failed in the newly added image-digest evidence command after the image itself built. The result is preserved and was not reused for rc.2. This is not a production release and contains no licensed factor database.

CarbonFactorResolver v0.13.1

CarbonFactorResolver v0.13.1

CarbonFactorResolver resolves structured material, energy, transport, and process factor requests against local catalogues and structured external factor sources. This release hardens precision, lifecycle-boundary and subject qualification, safe abstention, evidence admission, evaluation, and the product/runtime boundary.

Validation

Core package

• 189 tests passed
• 85.59% branch coverage

Developer-only offline QA harness

• 13 tests passed
• 43.24% branch coverage

FactorBench

• PASS (57/57)

Independent Holdout

• 32 cases
• Recall@5: 100%
• Top-1: 100%
• Wrong-candidate rate: 0%
• Correct abstention: 100%
• MORE_INPUT: 100%

Product boundary

CFR accepts structured FactorQuery / ResolutionRequest data and returns reviewed or locked factor results. PDF, DOCX, Excel, image and OCR handling, enterprise-document interpretation, BOM/activity-data extraction, formal footprint calculation, and report generation are outside the production runtime.

DOCX/PDF handling exists only in a developer-only offline QA harness and is not a CarbonFactorResolver product capability.

No extracted customer factors, source reports, acceptance workbooks, licensed databases, or formal factor approvals are included in this release.