Skip to content

HISTORY.md

Latest commit

 

History

History
25 lines (20 loc) · 2.88 KB

HISTORY.md

File metadata and controls

25 lines (20 loc) · 2.88 KB
Raw

HISTORY — errata & corrections log

Newest first. Every published INDICATIVE number is honest at the time of writing but may be revised as the corpus grows or an error is found. When a figure on the product page / README is corrected, log it here with the date, what changed, and why — so a reader can trust that corrections are tracked, not quietly overwritten. This is the trust-first counterpart to "we under-claim rather than over-claim".

2026-06-13 — Phase 6: Graphify language count VERIFIED at 25 (a self-corrected near-miss)

  • What: the product page claims DeepInit's AST parser (Graphify) reads "25 languages". During Phase-6 Track 0 an interim re-count briefly "corrected" this to 14 (from a too-narrow regex that only matched the explicit ts_module= configs in extract.py) and even briefly listed Go as grep-fallback-only. An empirical extraction test then re-verified the original 25: graphify update on Go/Rust/Elixir/Zig/Julia/Kotlin/Swift/Scala files all produced AST nodes; the extension-dispatch census in extract.py shows 25 tree-sitter language grammars (29 incl. JSON/SQL/Terraform; +Dart/Astro/Svelte/Apex/Razor extractors).
  • Why it mattered / outcome: the page's "25" was accurate — the brief "14" was the error, caught before any commit and reverted everywhere (detection.md + README + page + coverage-matrix + the A/B record). Only Crystal and OCaml are genuine no-grammar fallbacks. The trigger was an operator question about how Graphify was installed, which surfaced the package's full tree-sitter Requires list (Go/Rust/Elixir present) and prompted the re-check — a good example of the verify-don't-assume discipline catching a number before it shipped.
  • Reference: validation/graphify/ab-graphify-vs-fallback.json (grammars_real: 25, the correction_note) + the empirical test in test-plan Run 35.

2026-06-13 — Phase 6: harness check count is now self-deriving

  • What: the "150-check / 150/150" figure was hand-transcribed in several places and had begun to drift as the harness grew (now 162).
  • Fix: every page/README/CLAUDE.md count now derives from validation/STATS.json (built by tools/build_stats.py), and tools/check_stats_drift.py + the validate CI workflow fail on any future drift. The mutation harness (tests-fixtures-v1/_mutation_harness.py) proves the gated checks are load-bearing.

Template for future entries

## YYYY-MM-DD — <one-line summary>
- **What:** the figure / claim that changed (old → new).
- **Why:** the cause (corpus grew / measurement error / methodology refinement).
- **Reference:** the SHA-pinned record or run that supports the correction.

No corrections to the make-or-break gates have ever been needed: external metamorphic-FP has stayed 0 and the Mirror hard gate (deepinit_wrong_high) has stayed 0 across every measurement (see validation/_baseline.json).