fluidfix 0.1.0__tar.gz

fluidfix-0.1.0/CHANGELOG.md

@@ -0,0 +1,44 @@
+# Changelog
+
+## 0.1.0 — 2026-08-30 (r3)
+
+- `fluidfix guard`: commit-and-forget maintenance. Mechanical fault-file
+  discovery (traceback frames, else failing-test coverage ranking), one-shot
+  CI mode (exit 2 on refusal) or `--interval` watch mode, opt-in `--commit`
+  of each restoration, and a machine-readable `.fluidfix/last_refusal.json`
+  teach-me signal on refusal. Library API: `guard_once`,
+  `find_candidate_files`, `commit_repair`, `write_refusal`.
+
+## 0.1.0 — 2026-08-30 (r2, pre-release review applied)
+
+- `register()`: teach a new fault class from one registration — observation
+  contract + transform; the router infers the act code from the same worked
+  example. Mechanical observer now covers registered kinds generically.
+- Review fixes (all reproduced before fixing): site-local zero simplification
+  in `_reduce_literal` (was able to eat an unrelated `-0.5` elsewhere on the
+  line); byte-preserving file handling (CRLF/FF-safe, no write on refusal);
+  path-boundary coverage matching (same-suffix sibling packages can no longer
+  shadow the defect file); root-level modules get a correct default `--cov`
+  target; CLI timeout flags (`--suite-timeout`, `--test-timeout`,
+  `--candidate-timeout`, defaulting consistently); observer errors are
+  diagnosable (truncation, missing text, wrong echoed ids raise instead of
+  masquerading as refusals); injected observer clients work without the SDK;
+  `acts_tried` counts only real candidates.
+
+## 0.1.0 — 2026-08-30
+
+Initial release.
+
+- Vendored routing kernel (fluid-router, verbatim, `minimal in D∩I`) and
+  EMIT/ADVANCE/HALT loop discipline (fluid-router2, verbatim).
+- Mechanical localisation: traceback frames ∪ failing-test coverage
+  (pytest-cov `--lf`, no `-x`) ∪ AST statement-span expansion. 33/33 true
+  lines captured on the benchmark corpus at 0 tokens.
+- Observers: `MechanicalObserver` (zero-token) and `ClaudeObserver`
+  (Claude Opus 5, batched, schema-forced, with server-side refusal fallbacks).
+- Clear-data applier pointers: `literal_value`/`literal_occurrence` (17/27 →
+  26/27 in-vocabulary byte-exact on the corpus) and `op_occurrence` (closes
+  the last measured miss).
+- Safety: green-suite refusal, out-of-vocabulary refusal, byte-exact rollback,
+  candidate timeouts, defended oracle (five encoded harness-defect classes).
+- `fluidfix` CLI: `repair`, `packet`, `selfcheck`.

fluidfix-0.1.0/COMMERCIAL.md

@@ -0,0 +1,28 @@
+# Commercial licensing
+
+fluidfix is dual-licensed.
+
+**Open source.** Under AGPL-3.0-or-later you may use, modify, and
+redistribute fluidfix freely — including commercially — provided you comply
+with the AGPL, including section 13: if you run a modified version as a
+network service, you must offer the complete corresponding source to that
+service's users.
+
+**Commercial.** If the AGPL does not suit your use — embedding fluidfix in a
+proprietary product, offering it inside a closed SaaS, or any deployment where
+the source-offer obligation is unacceptable — a commercial license removes
+those obligations.
+
+Contact: kanchetiindu9@gmail.com with:
+
+1. how fluidfix will be deployed (product, internal tool, SaaS),
+2. approximate scale (developers, repositories, or repair volume),
+3. whether you need the act-dictionary compilation service (a per-repository
+   vocabulary trained on your codebase's own bug history — measured at +21
+   points on the candidate-containment ceiling for real one-line fixes, 62%
+   vs 41% for a general dictionary, with the buggy line supplied; that is a
+   ceiling on what the dictionary can reach, not a realized repair rate).
+
+The copyright holder for fluidfix and the vendored kernels
+(fluid-router, fluid-router2) is devkancheti4-design; dual licensing is
+offered by the copyright holder, which the AGPL fully permits.