Skip to content

fix(docs): sync version metadata, changelog, and roadmap to shipped reality#4

Merged
StressTestor merged 3 commits into
mainfrom
fix/version-doc-drift
Jun 16, 2026
Merged

fix(docs): sync version metadata, changelog, and roadmap to shipped reality#4
StressTestor merged 3 commits into
mainfrom
fix/version-doc-drift

Conversation

@StressTestor

Copy link
Copy Markdown
Owner

why

the repo docs drifted behind what actually shipped. 3.2.0 and 3.2.1 are both live on PyPI as promptpressure-evals, but VERSION still said 3.1.0, the CHANGELOG had no entries for either release, and the roadmap claimed v3.0 was current and listed already-shipped work as "next". for a project whose whole pitch is becoming a citable, trustworthy eval method, shipping releases with no changelog and a stale version file is an own-goal.

what

  • VERSION 3.1.0 -> 3.2.1 (now matches pyproject.toml and PyPI)
  • CHANGELOG backfilled the two missing releases, reconstructed from git history:
    • 3.2.0: pp launcher (provider/model/eval-set dropdowns, SSE status, Cancel) + PyPI packaging as promptpressure-evals
    • 3.2.1: pp --version reads __version__ instead of dist metadata
  • roadmap rewritten to reflect real shipped state (3.0/3.1/3.2.x), with a reality-check on the multi-turn drift gap (smoke/deep tiers still have zero generated sequences, judge has no published calibration) and v3.3 set as the credibility release: drift-corpus + judge-calibration v0.1
  • .gitignore ignores .debate/ scratch and outputs/*.zip
  • created the missing v3.2.0 and v3.2.1 git tags on their release commits

notes

  • docs/metadata only. no code changes, package still imports clean, __version__ = 3.2.1.
  • the v3.3 direction (calibrate the judge on real multi-turn sequences, not the single-turn corpus) came out of a multi-model debate. the README still says smoke/deep sequences are "coming" - narrowing that overclaim is a separate, public-facing change and not in this PR.

@github-actions

Copy link
Copy Markdown

codex review

Codex review is not configured for this repository yet.

Missing required secret: CODEX_ACCESS_TOKEN.

…0/3.2.1

VERSION file lagged at 3.1.0 while pyproject.toml and PyPI (promptpressure-evals)
were already at 3.2.1. CHANGELOG had no entries for the two shipped releases:
3.2.0 (pp launcher + PyPI packaging) and 3.2.1 (pp --version fix). reconstructed
both from git history.
roadmap claimed v3.0 current and listed already-shipped work as 'next'. rewrote to
reflect actual shipped state (3.0/3.1/3.2.x), add a reality-check on the multi-turn
vaporware gap, and set v3.3 = drift-corpus + judge-calibration v0.1 as the next
release. deprioritized '300+ prompts' (uncalibrated data is noise, not credibility).
@StressTestor StressTestor force-pushed the fix/version-doc-drift branch from 4e5eeec to d23cf73 Compare June 16, 2026 02:53
@StressTestor StressTestor merged commit 3d13647 into main Jun 16, 2026
5 checks passed
@StressTestor StressTestor deleted the fix/version-doc-drift branch June 16, 2026 02:54
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant