Changelog
v2.0.10
- Stop automatic Copilot code review on every push; reviews remain available on request.
- Drop Python 3.10 from the supported range at its upstream end of life; the core now requires Python 3.11–3.13.
v2.0.9
- Replace the README stability attribution with the dated, linked CSV result-labeling defect.
- Remove the Keras comparison image and retain the runnable example and first-sweep guide.
- State research citation use directly in the README and citation guide.
v2.0.8
- Publish each new reviewed Talos version automatically after successful protected-master CI, without a separate release approval.
- Reuse the exact CI-tested Talos distributions for signing and publication instead of repeating training and builds.
- Reject foreign or failed CI evidence and skip superseded master heads before creating a release.
- Require dependency-bot version and changelog bumps so every merge has a unique release identity.
- Update installation instructions for the published Talos 2 generation and record verified 2.0.7 release provenance.
- Record the attained OpenSSF Best Practices Silver badge and its dependency, signature and assurance evidence.
v2.0.7
- Backport reviewed legacy framework and documentation dependency security fixes while preserving original model predictions and explicit trusted deserialization.
- Reject vocabulary config reads when safe mode is unset; require a fresh report for each executed upstream audit.
- Require hash-pinned owned Keras/Protobuf wheels in the legacy extra and CI; retain upstream findings and reject unproved advisory dispositions.
- Monitor owned dependency upstream identities and documentation advisories in one bounded weekly job without increasing PR fan-out.
- Include the owned dependency wheels in signed GitHub release artifacts, keeping the Talos-only PyPI upload separate.
v2.0.6
- Avoid repeated per-file wheel metadata work during source snapshots while preserving dependency ownership and fresh resume checks.
- Record two maintainer-approved, exact-version documentation audit exceptions through November 1, 2026, expiring at 00:00 UTC on November 2; retain all findings and block every other advisory.
v2.0.5
- Consolidate CI coverage, installed-wheel checks and dependency audits while preserving scientific acceptance and required check names.
- Cancel superseded source checks, bound documentation execution and group and stagger dependency updates.
v2.0.4
- Enforce the documented Python style with zero findings across maintained source, tools, tests and examples, retaining existing quality and typing ratchets.
- Preserve scientific and archive behavior while correcting import ordering, whitespace, docstrings and explicit export lists.
- Record the six-month regression ledger, verified master coverage and network and credential trust boundaries for the OpenSSF assessment.
- Explain Sigstore verification and signing-key custody for consumers of future authorized release artifacts.
v2.0.3
-
Restore familiar README navigation and notebook introductions, preserve historical citations and systemize version-specific software and experiment references.
-
Replace the model animation with a runnable Keras-to-SFD comparison and verify an additional recovery notebook on all three frameworks.
-
Publish verified documentation and source-bound statement coverage from successful master CI, with live OpenSSF status badges.
-
Document the tested Talos 1.4 installation constraint and support commitment through at least 2028.
-
Verify committed manifest content against its immutable identity before execution or reindexing.
-
Reject malformed objective directions and execution settings before importing caller code.
-
Verify external sampler TLS certificates and hostname, reject transport downgrade, and require caller-owned service credentials.
-
Restore realized dictionary-sweep candidates before sampling and attest ordered parameter keys for new run identities.
v2.0.2
- Bank the core coverage gain at 70% lines and 59% branches under the existing ratchet.
- Verify complete-package statement coverage across framework acceptance and runnable documentation without narrowing the measured source.
- Add scientific cohort-selection, causal-scaling and shipped-framework-template regression tests.
- Keep the metric-rescaling regression away from mathematically tied medoids across supported sklearn versions.
- Attach immutable distribution assets, checksums and verified Sigstore build provenance to authorized GitHub releases while keeping PyPI publication explicitly gated.
- Tighten workflow token permissions and invoke the locked local browser tooling directly.
- Run feature-branch validation through pull requests and retain protected-master builds without duplicate push executions.
- Restore explicit contribution test/style requirements and document vulnerability handling and the annual maintenance horizon.
v2.0.1
- Designate zero-bang and bit-mis alongside mikkokotila as code owners for governance and supply-chain surfaces.
- Apply the documented bounded backoff when external documentation links return transient failures.
- Bind hash-locked official CPU framework builds to upstream advisory identities while auditing every installed dependency and recording lookup evidence.
- Repair hosted Linux packaging locks, CPU framework execution, documentation backup fixtures and protected-base version comparison; verify optional dependency failure diagnostics.
- Require native CodeQL security-result protection and document verified repository activation order, credential scope and automatic review behavior.
- Adopt repository governance, issue and review contracts, measured quality ratchets, supply-chain checks and controlled release tooling for Autonomio Talos.
- Preserve the public Talos interfaces, optional Keras, TensorFlow and PyTorch integrations, reproducible experiment records and trusted archive recovery contracts.
- Expand contributor, security, packaging and release documentation while retaining the existing README and carefully authored user guides.
v2.0.0
- Adopt an independent framework-neutral experiment core with the legacy
Scan, reporting, prediction, evaluation, deployment and restoration interfaces. - Add native Keras, TensorFlow and PyTorch SFD templates, CLI execution, manifests, resumable experiments and local queue control.
- Preserve reducers, archive readers and explicit framework factories; verify documentation examples against bundled real fixtures.
- Adopt the Autonomio documentation system and style with preserved source content and executable examples.
Earlier releases and contribution history remain in the repository history.