docs(changelog): backfill 0.1.0 through 1.5.0 and file the scaffold where it belongs - #58
docs(changelog): backfill 0.1.0 through 1.5.0 and file the scaffold where it belongs#58lopadova wants to merge 1 commit into
Conversation
…here it belongs The file claimed everything since the W6 scaffold was unreleased, and recorded exactly one version: 1.3.0. Meanwhile the package shipped 1.0, 1.1.0, 1.2.0, 1.4.0, 1.5.0 and 1.6.0. Somebody installing 1.6 and opening CHANGELOG.md was handed the pre-release scaffold and told it was still unreleased. Every entry is reconstructed from a published GitHub release note or from the diff for that tag range, and the details were checked against the code rather than paraphrased: the offline metrics are named by their real aliases (`bertscore-like`, not "BERTScore"), the 1.6 features by their real flags and command names, and `DatasetBuilder::withRepetitions()`, `eval-harness:baseline`, `--budget-usd` and the seven trajectory metric classes were each confirmed to exist. Where a release note and the diff disagreed, the diff won. Two findings fell out of doing it: - 1.5.0 contains no package change at all — a single Dependabot bump of linkify-it, a transitive dev dependency of docs-site/. Recorded as such, because a reader seeing a version bump between 1.4.0 and 1.6.0 would otherwise assume there was one. - The long-standing `[Unreleased]` section was the 0.1.0 content, and its test count (87 unit) was a mid-development figure; 0.1.0 shipped 109 unit plus 3 architecture, per its release note. `[Unreleased]` now holds the README quickstart fix from #57 and this change. The 1.0 tag has no `v` prefix, unlike every tag before and after it, so the compare links account for that. Two LESSON entries. Unit 1136, Architecture 3, Pint and composer validate --strict clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EZ4u82zdoa8naYm75kppMC
Deploying laravel-patent-box-tracker with
|
| Latest commit: |
3b411fe
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://9faf400c.laravel-patent-box-tracker.pages.dev |
| Branch Preview URL: | https://claude-changelog-backfill.laravel-patent-box-tracker.pages.dev |
Deploying eval-harness with
|
| Latest commit: |
3b411fe
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://3c7889ed.eval-harness.pages.dev |
| Branch Preview URL: | https://claude-changelog-backfill.eval-harness.pages.dev |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3b411fec91
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| - This file: 0.1.0–1.5.0 backfilled, and the long-standing `[Unreleased]` | ||
| section — which described the pre-release scaffold — filed under `[0.1.0]` | ||
| where it belongs. |
There was a problem hiding this comment.
Record the changelog backfill in progress
This is a meaningful release-documentation step, but the commit leaves docs/PROGRESS.md unchanged, so the repository's handoff log ends at the 1.6 implementation and gives the next agent no record of the backfill, validation gate, or remaining remote action. Add a progress entry for this work as required by the repository guide.
AGENTS.md reference: AGENTS.md:L22-L22
Useful? React with 👍 / 👎.
Sottotask
ID: follow-up. Documentation debt found while checking which repos needed a release tag.
Plan ref: none.
Summary
CHANGELOG.mdclaimed everything since the W6 scaffold was unreleased and recorded exactly one version — 1.3.0 — while the package shipped 1.0, 1.1.0, 1.2.0, 1.4.0, 1.5.0 and 1.6.0. Someone installing 1.6 and opening the file was handed the pre-release scaffold and told it was still unreleased.Changes
File modificati:
CHANGELOG.md(rewritten),docs/LESSON.md. No code, no migration, no test changes.Where each entry came from
Every entry is reconstructed from a published GitHub release note or from the diff for that tag range — the commit subjects could not have produced it (one range is forty commits of "address PR #37 Nth-round Copilot review comments"). Where a release note and the diff disagreed, the diff won.
Details were checked against the code rather than paraphrased:
rouge-l,bertscore-like(not "BERTScore"),refusal-quality,citation-groundedness; all four files added 2026-05-03/04, inside the rangesrc/Metrics/Trajectory/, excluding the interfacewithRepetitions(),eval-harness:baseline,eval-harness:brief,eval-harness:promote-online,--budget-usd,--comparison-out,--confident-onlyeach confirmed presentTwo findings that fell out of doing it
linkify-it, a transitive dev dependency ofdocs-site/. Nothing undersrc/,config/orresources/differs from 1.4.0. Recorded as such, because a reader seeing a version between 1.4.0 and 1.6.0 would otherwise assume there was one.[Unreleased]section was the 0.1.0 content, and its test count (87 unit) was a mid-development figure. 0.1.0 shipped 109 unit plus 3 architecture, per its release note — corrected, and the section filed under[0.1.0] - 2026-05-02.Housekeeping
[Unreleased]now holds the README quickstart fix from #57 and this change. The header states plainly that 0.1.0–1.5.0 were reconstructed on 2026-08-25 and how, so the file does not pretend to a continuity it never had. The1.0tag has novprefix, unlike every tag before and after it, so the compare links account for that.Test gate
composer validate --strictvalidArchitecture impact
None. Documentation only.
Security impact
None.
Risk
Low. The risk it removes is a reader being misinformed about what they installed.
Rollback plan
Revert the merge commit.
Generated by Claude Code