amos 9d633e9e67 Batch 15 (20261023-C-hwe-scale-15): 320 HWE-derived repair tasks
One user-space repair task per eligible HWE debug-diag / compile-deploy base
ID: 320 admitted (C 218, C++ 97, build scripts 5), 1 skipped by the content
filter. Each ships a public spec with a check-to-basis table, at least two
defects with single-fix fixtures, a symptom-masking fixture and an alternative
correct implementation. Claude re-checked all 320: structure, the HWE quality
items and trace6 Docker calibration receipts against current digests
(320/320 USABLE). Difficulty is unmeasured. Known follow-up: 212 tasks use 120
ad-hoc family names; the next batch maps them to the canonical list.
The batch's 1,256-line progress log stays in its factory report and the
36-local branch batch15-hwe.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 01:55:56 +08:00

6.8 KiB

resource-rule-programs specification

Route C original C11 software task inspired by hwe/gi-leak-regex-regfree-missing. The source regcomp/regfree ownership issue is retained as a deterministic compiler-port lifetime problem. No source scripts executed. Source admission/hash proof was completed by root, not repeated here. No real incident, hardware, timing, network, model sampling or training claim. Difficulty UNMEASURED; initial status DRAFT.

The public engineering contract will be template/docs/contract.md. The replacement compiler implements a documented bounded byte regular-expression language; each successful compile returns a separately owned program, observable through the public port ledger and real matching behavior. A bank supports atomic replacement, priority selection and repeated clear. Two independent ownership defects: installed bank retirement leaks programs; rollback after partial compilation leaves temporary programs alive. The latter requires a documented rejected batch, outside the default demo. Necessary compiler, public driver and header support are counted as ordinary unique engineering code, with no repeated fixtures counted.

Acceptance requires all documented rc, bank generation/count, real match hit count, selected identifier/priority/input index, live program/term ledger, unchanged bank after rejected replacement, and final lifetime closure. Python oracle will compile its own public grammar to Python byte regex and independently model bank state; the C public driver only prints raw observations. Exact per-clause file:line mapping will be frozen after implementation. Candidate is engine.c. Gold array and alternative priority-sorted linked implementation must both pass. Negative fixtures: initial, noop, public-demo hardcode, private-grader reader, each isolated repair and symptom masking.

Execution: clean scratch, all material before first child, candidate-identity Linux subreaper supervisor with deadlines and fresh PID/start checks, strict cleanup then independent lstat absence. Exceptions and inconsistent metadata reject. stdout numeric protocol accepts optional signs, arbitrary leading zeroes and horizontal whitespace as documented; stderr arbitrary bytes without cap. Final production-image diagnostics and local/Docker five-per-class calibration required. Root owns full repository unit suite and review.

Frozen behavior-to-public-evidence mapping

Acceptance behavior Exact public basis
C11 build, frozen compiler/driver, candidate file template/docs/contract.md:3 (C01)
Initial bank, reusable repeated clear, EOF closure and generation template/docs/contract.md:5 (C02)
Complete list prevalidation, rule count/identifier/priority/length bounds and duplicates template/docs/contract.md:7 (C03)
Literal, quotes, anchors, arbitrary-byte dot, empty expression template/docs/contract.md:9 (C04)
Classes, negation, ranges, escaped members, syntax errors template/docs/contract.md:11 (C05)
Quantifiers, exact end anchors, term ledger and single ownership template/docs/contract.md:13 (C06)
Deterministic budget, syntax-before-memory error, budget consumption template/docs/contract.md:15 (C07)
Input-order compile, partial-error cleanup, transactional publication/retirement, copied patterns template/docs/contract.md:17 (C08)
Real all-rule hits, priority/input-order selection, empty/no-hit and oversized-message sentinels template/docs/contract.md:19 (C09)
Raw public driver command protocol and EOF observation template/docs/contract.md:21 (C10)
Decimal rows, arbitrary signs/leading zeroes/horizontal whitespace, stderr permitted template/docs/contract.md:23 (C11)
Repeated replacement/error/empty/lifetime behavior and defined single-threaded scope template/docs/contract.md:25 (C12)

All hidden matrix rows carry their clause label; no hidden business assertions run inside the C driver. The Python expectation uses byte regex translation and an independent bank model, with a fixed random seed. The compiler-port support is a public deterministic allocator/matcher rather than host RSS/call counting. Objects must actually match patterned messages and produce priority selections; resource-ledger checks additionally require release.

Gold uses a contiguous fixed-capacity installation and scans entries to select the winner. valid_priority_list builds heap nodes, inserts them in priority order, publishes a linked chain and selects its first matching node while counting all hits. The list validation uses pairwise forward duplicate checks; errors dispose the chain. Their observable contract is identical, with different data structures and matching-selection control flow.

Defect 1: retirement visits installed entries without releasing port objects, affecting normal demo, replacement, clear and EOF. only_fix_retirement repairs that loop alone and still leaks staged programs after partial compile failure. Defect 2: on a compiler error only the temporary container is freed, leaving previously compiled entries live. only_fix_rollback adds staged-program release, while normal retirement is still defective. It requires syntax/memory failure after at least one successful compile, outside --demo. mask_symptom turns failed batches into apparent success and clears the bank; snapshots and match behavior reject it. cheat hardcodes demo winner 12 and fails non-demo match/priority checks. noop lacks compilation and hits. cheat_read_grader needs private check readability to enable gold behavior; formal UID isolation must reject it, while local lack of isolation may allow it.

Ordinary unique engineering lines (nonblank, noncomment; braces/directives counted; fixtures counted once only): rules.h 44, compiler.c 212, engine.c 116, driver.c 131, Makefile 10 = 513. These files are required public interfaces/compiler/business/driver/build support. No candidate-only or strict brace-stripped line gate is used. Verify is a byte copy of rollout/bos_verify.py. Source attribution: original HWE prompt/reference read; no source script executed, no source replication/hash repeated. Root proof root-next-medium-source-proof-06.json establishes source admission.

Status is READY_FOR_CALIBRATION until actual final-digest gates are recorded. Production diagnostics inject only trusted copies and retain complete loadable task packages. No real Agent rollout, native tool trajectory, physical hardware, training, or empirical task difficulty has been measured.

Final status CALIBRATED / UNMEASURED. Frozen digest 82843ff9628e02ee8542114656e452a26897217674fbe45e03285cdd67881747. Docker45/45 expected, cleanup confirmed. Local only permitted reader5PASS exception. 22 fixed-image production diagnostic packages retained, 7 tests pass. Root full suite/review pending. Exact receipts and limitations: factory/20261023-C-hwe-scale-15/subagents/gi-leak-regex-regfree-missing/report.json and evidence-audit.json.