Release v0.4.0 by 24601 · Pull Request #32 · 24601/Augustus

24601 · 2026-09-20T14:39:47Z

Prepare the v0.4.0 cut from latest main after merged #30 (Merve §112). Does not reopen or amend open #31 (0743 fold — that branch is left alone).

Does not create the GitHub Release or tag. After merge, parent cuts:

gh release create v0.4.0 \
  --repo 24601/Augustus \
  --title "v0.4.0" \
  --notes-file docs/release-notes-v0.4.0.md

Head SHA: 30cd61a3896e20c4c0ec4d08dd521b34f47dc71e

Peg: typesafe-ai/skills v0.5.7 (65a39f3). Live HEAD of that repo is still this commit.

High-level summary

Twenty-eight commits after the v0.3.0 tag, grouped (not every hourly §):

Class breadth. Jev is the exemplar, not the monopoly — Laya, kev, OpenJev, TypeAR, GLiNER species, DGP, encoder zero-shot. Merve: BERTForXYZ → DeBERTa → ModernBERT; Jev vs GPT-5.6 bakeoffs are a category error.
Measurement honesty. Harbor / jevals practice. Ranking ≠ calibration. Score is 0..n−1. Noul has no confidence field. ECE ≠ an edge. 0.85 / minProbability theater. Soft Noul ≠ hard gate. VERIFY needs discriminating evidence.
Composition / placement. Decision Graph Protocol, meaning-grep, gut cost-of-error, fail-open vs fail-closed, compaction 0.5 theater, prune ≠ deny.
Formal methods pillar unchanged in spirit (soundness theater: a Noul is a SENSOR).
CHANGELOG restored to Keep a Changelog. Hourly uniqueness dump moved to research/changelog-hourly.md.

No invented metrics. No soft-Noul-as-hard-gate. No pipeline() install recipes.

Version pins

.claude-plugin/marketplace.json 0.3.0 → 0.4.0
.agents/skills/augustus/SKILL.md YAML metadata.version 0.3.0 → 0.4.0

Security

Added lean SECURITY.md (supported: 0.4.x; report via GitHub Security Advisories / private vulnerability reporting — no invented email).
Added CONTRIBUTING.md (docs/skill PRs; adversarial review; no secret keys; fold uniqueness pointer; branch protection is parent-owned).
Added brief CODE_OF_CONDUCT.md.
Added .github/workflows/scorecard.yml (official ossf/scorecard-action v2.4.4). Do not invent a numeric score — the Scorecard API is currently 404 / unscored until this publishes from main. Optional badge after the first successful default-branch run: https://api.scorecard.dev/projects/github.com/24601/Augustus/badge.
.gitignore now covers common secret filenames.
Dependabot security updates, secret scanning, and push protection stay enabled.

Discoverability (in-repo only)

README: one-liner above the fold; release + Pages + MIT badges; homepage; Install for marketplace / skills.sh / copy path; Not a TypeSafe product; rh-guard companion.
docs/index.md + _config.yml: v0.4.0 blurb, last-updated, SEO (System One, decision judgment, Jev-class). Pages gate strings preserved.
CITATION.cff for academic citation.
Already listed on awesomejev.com and several awesome-jev catalogs. This PR does not spam those lists.

Topics for parent (`gh repo edit`)

Current topics already include formal-methods, agentic-ai, zero-shot-classification. Suggest adding if still missing:

gh repo edit 24601/Augustus --add-topic decision-making

Do not invent stars.

Adversarial test results — GATE PASS

Ran 2026-09-20T14:40:59Z locally on this tip. GitHub Check Pages site / build is green on run 35517248654.

Check	Result
`python3 .agents/skills/augustus/scripts/evaluate_decisions.py --self-test`	`self-test ok` (local scoring math; no model call)
YAML `safe_load` SKILL frontmatter	`name=augustus` `license=MIT` `version=0.4.0` `typesafe_skill=v0.5.7 @ 65a39f3` `description_chars=46154`
Uniqueness fragments in SKILL description	Merve category-error / BERTForXYZ lineage / soft scores ≠ hard gates / soft Noul ≠ hard safety all present
Fragment counts (skill / notes / CHANGELOG / hourly archive)	category-error 6/1/0/1; ranking ≠ calibration 39/15/1/2; CHANGELOG 382 lines (was a ~3.5k-line dump)
Version pins	marketplace / CITATION.cff / README = 0.4.0
Pages gate strings in `docs/index.md`	`TypeSafe Jev Choice/Score/Noul`, `Install the skill`, `LICENSE` present; `ecosystem.md` present
GitHub Pages workflow	SUCCESS
Refresh dry-run (no keys, no log write)	docs.typesafe.ai/llms.txt 200; evals.typesafe.ai 200; openrouter jev-1.13 200; Pages homepage 200; typesafe-ai/skills HEAD still `65a39f393687`
Secret scan	no `.env` / `*.pem` / assigned `API_KEY=` literals in tree
Release-notes honesty	no `pipeline()` / no invented Scorecard number; ranking ≠ calibration and soft Noul ≠ hard gate present
`scorecard.yml`	`permissions: read-all`; official `ossf/scorecard-action@2d1146689` (v2.4.4)

scripts/hourly-refresh.sh was not run (it commits and pushes to main). scripts/refresh-jev-research.sh was not run as-is (it appends research/refresh-log.md); the public HTTP checks above are the safe dry-run.

Did not add npm/pip dependencies.

Out of scope

Open Fold hourly 0743 HIGH: ywchiu Harbor-jevals, Verdict-open-jev, DecisionOps #31 / cursor/fold-hourly-0743-high-794b — not touched
Creating v0.4.0 tag or GitHub Release
Merging this PR
Org/branch-protection settings

Bump the skill and marketplace pin to 0.4.0. Restore Keep a Changelog by moving the hourly uniqueness dump to research/changelog-hourly.md. Add SECURITY, CONTRIBUTING, a brief CODE_OF_CONDUCT, CITATION.cff, and the official ossf/scorecard-action workflow. Refresh README and Pages SEO without inventing metrics. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Make ranking ≠ calibration and soft Noul ≠ hard gate greppable in the release notes without changing the meaning. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Keep notes.md §113 / composition items 273–288 / findings batch #96. Do not reopen or amend #23–#30/#32. Do not bump 0.4.0. Live REST relock: Verdict HEAD 30f15564 (README c6f317ae UNCHANGED); tanay HEAD bfd2be53 (README 33bc46e3 UNCHANGED); SemIf 2237★; jevlike 1049★; yibie 521★; Laya likes 861; ywchiu 1★ size 173; tracker likes 66 / lastModified UNCHANGED. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Keep notes.md §113 / items 273–288 / batch #96. Do not reopen #23–#30/#32/#33. Do not bump 0.4.0. Keep #33 scannable README; uniqueness lock is one overlay line, not a restored dump. Live REST relock: tanay HEAD 96c90cdd (README 33bc46e3 UNCHANGED); SemIf 2239★; yibie 522★; Laya likes 862. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

SemIf 2240★; jevlike 1050★; tracker likes 67. lastModified UNCHANGED. Do not reopen #23–#30/#32/#33. Do not bump 0.4.0. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Tracker likes 67 / lastModified UNCHANGED. Do not reopen #23–#30/#32/#33. Do not bump 0.4.0. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

HEADs unchanged. Do not reopen #23–#30/#32/#33. Do not bump 0.4.0. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…onOps (#31) * Fold hourly 0743 HIGH: ywchiu Harbor-jevals, Verdict-open-jev, DecisionOps Docs-only fold of 2026-09-20 ~07:43 Boise System One watch into Augustus (notes.md §112, items 268–283, findings batch #95). Backend-agnostic categorization/scoring class; Jev is exemplar not monopoly. Skip Archer rewrite. Quote live REST. Soft Noul ≠ hard safety. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Remap 0743 HIGH off merged #30 onto §113 / items 273–288 / batch #96 #30 owns notes §112, composition 268–272, and findings batch #95. This fold now uses the next free IDs and uniqueness fragments name do-not-reopen #23–#30 / notes.md §113. Relock live REST: ywchiu 4322c350 / 75e6a338, Verdict 465d542f / e2932455, DecisionOps fe4e7aff / 4688009e, SemIf 2225, yibie 516, Laya likes 852. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Relock 0743 HIGH live REST census after #33 merge SemIf 2240★; jevlike 1050★; tracker likes 67. lastModified UNCHANGED. Do not reopen #23–#30/#32/#33. Do not bump 0.4.0. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Relock 0743 HIGH census: SemIf 2241, jevlike 1051, yibie 523, Laya 863 Tracker likes 67 / lastModified UNCHANGED. Do not reopen #23–#30/#32/#33. Do not bump 0.4.0. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Relock 0743 HIGH census: yibie 525★, Laya likes 864 HEADs unchanged. Do not reopen #23–#30/#32/#33. Do not bump 0.4.0. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> --------- Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs/skill only. notes.md §115 / composition 303–308 / findings #98. Quote theirs; invented_signal false. Do not reopen #31/#32/#33/#35. Reserve §114 for open #35 remap. Skip Archer. Do not merge. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

* Fold TianyuCodings/NanoJev unified-games-v1 densify Docs/skill only. notes.md §115 / composition 303–308 / findings #98. Quote theirs; invented_signal false. Do not reopen #31/#32/#33/#35. Reserve §114 for open #35 remap. Skip Archer. Do not merge. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Relock NanoJev live REST 1284★ (+1 vs fold 1283) Adversarial review head-check: GitHub TianyuCodings/NanoJev is 1284★. Keep uniqueness lock consecutive across 21 overlays. Quote docs/TYPESAFE_CONTRACT.md (README path), not the repo root 404. Do not merge. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Keep NanoJev densify on §115 after merged #35 Rebased onto main (0189825 / #35). Merged #35 owns notes §114 / items 289–302 / batch #97. This fold stays §115 / 303–308 / #98. Reorder composition-algebra so 289–302 precede 303–308. Drop open-#35 reservation language. Do not merge. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Relock NanoJev live REST 1288★ and Hub likes 58 Adversarial head-check after rebase: GitHub TianyuCodings/NanoJev is 1288★ (+4 vs 1284). Hub C-Tianyu/NanoJev likes 58 (+1 vs 57). Keep the 0915 uniqueness lock consecutive across 21 overlays. Do not merge. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Fix leftover Hub likes 57 next to 1288★ lock Adversarial review on 295e6e2: consecutive 0915 lock already has Hub likes 58, but validation.md prose still said 57. Live REST still 1288★ / likes 58. Do not mutate uniqueness_gate.py. Do not merge. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Relock NanoJev live REST 1289★ (+1 vs 1288) Adversarial head-check after leftover-likes fix: GitHub TianyuCodings/NanoJev is 1289★. Hub likes 58 held. Keep the 0915 uniqueness lock consecutive across 21 overlays. Do not mutate uniqueness_gate.py. Do not merge. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> --------- Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

cursoragent and others added 2 commits September 20, 2026 14:39

Keep 0.4.0 honesty phrases on one line

30cd61a

Make ranking ≠ calibration and soft Noul ≠ hard gate greppable in the release notes without changing the meaning. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

24601 marked this pull request as ready for review September 20, 2026 14:41

24601 merged commit 486e93e into main Sep 20, 2026
1 check passed

24601 deleted the cursor/release-v0.4.0-3a04 branch September 20, 2026 14:43

cursor Bot mentioned this pull request Sep 20, 2026

Fold hourly 0743 HIGH: ywchiu Harbor-jevals, Verdict-open-jev, DecisionOps #31

Merged

cursor Bot pushed a commit that referenced this pull request Sep 20, 2026

Relock 0743 HIGH census: yibie 525★, Laya likes 864

c049e22

HEADs unchanged. Do not reopen #23–#30/#32/#33. Do not bump 0.4.0. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

This was referenced Sep 20, 2026

Fold hourly 0843 HIGH: hysteresis, instruct-tuning ECE, calibration does not compose #35

Merged

Fold TianyuCodings/NanoJev unified-games-v1 densify #36

Merged

24601 mentioned this pull request Sep 20, 2026

Fold hourly 0947 HIGH: jev-as-judge, OneForward candidate_mass, catalogs and RLCD heads #41

Merged

Navigation Menu

Release v0.4.0 - #32

Release v0.4.0#32
24601 merged 2 commits into
mainfrom
cursor/release-v0.4.0-3a04

24601 commented Sep 20, 2026 •

edited by cursor Bot

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

24601 commented Sep 20, 2026 • edited by cursor Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

High-level summary

Version pins

Security

Discoverability (in-repo only)

Topics for parent (gh repo edit)

Adversarial test results — GATE PASS

Out of scope

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

24601 commented Sep 20, 2026 •

edited by cursor Bot

Loading

Topics for parent (`gh repo edit`)