Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
39 changes: 39 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,45 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0

## [Unreleased]

## [0.60.0] - 2026-07-06

The **provable-correctness deepening + gate-integrity** release. The proof corpus
climbs several kernels to strict **L4** and attention / Pillar-1 metrics / the
Pillar-2 cross-entropy backward gradient to **L5** (Lean/Mathlib-verified), and the
L4/L5 leveling is made reproducible on a fresh clone/CI. A 14-agent **Fable
architectural review** audits every quality gate for *theater* — gates that look
enforced but aren't — and turns the findings into an EV-ranked, falsifiable backlog,
fixing the first holes (a CI pass-check that merged failing runs green) and deleting
scaffold that never compiled.

### Added

- **Proof-corpus L4/L5 climb** (#2282, #2284, #2287, #2288) — 10 kernels raised to
strict L4; attention, Pillar-1 metrics, and the Pillar-2 cross-entropy backward
gradient proved to **L5** against Lean/Mathlib (prebuilt oleans via `lake env lean`).
- **Fable architectural review** (#2292) — a 14-agent audit producing a verified
claims ledger, a gate-integrity assessment, and an 18-item EV-ranked falsifiable
backlog (`docs/specifications/fable-architectural-review.md`).
- **apr code parity for Claude Code _or_ Google Antigravity** (#2277) — Pillar-5
prompt parity across either agentic coding host.

### Fixed

- **CI pass-check substring bypass** (#2298, PMAT-CI-PASSGREP-001) — the
aprender-compute gate grepped `"test result.*0 failed"`, which substring-matches
"10 failed", so a 10-failure run merged **green**. Anchored to require
`test result: ok.` and the absence of `test result: FAILED`.
- **Proof-status over-promotion** (#2285, #2281, #2283) — strict per-obligation
L4/L5 leveling, reproducible on a fresh clone by scanning the in-tree Lean tree;
honesty sweeps credit lost proofs and correct dangling-tag / L1→L2 levels.

### Removed

- **186 orphan contract-scaffold modules** (#2302, PMAT-THEATER-TRIAGE-001) —
`crates/aprender-core/src/format/*_NNN_NNN.rs` files never reached via
`mod`/`include!` (never compiled → dead `#[test]`s = enforcement theater).
Build- and test-neutral; honest-ifies the reported test count.

## [0.59.0] - 2026-07-04

The **Pillar-1 (scikit-learn) breadth + provable-parity** release. Three sklearn
Expand Down
Loading
Loading