From 976d657808581e2d75b3b088466a86e64396f842 Mon Sep 17 00:00:00 2001 From: sylph-decoder Date: Mon, 31 Aug 2026 03:33:00 +0000 Subject: [PATCH] tools: the register now prints its own scope, not just its verdict sylpheed-port found audit-kinds auditing 16 of 71 authored justifications and never saying so -- a checker that fails correctly while describing a sixth of the corpus. Their line is the one that generalises: 'I checked and it was fine' and 'I checked the part that declared itself' read identically in a log, and only one of them is what gets quoted. Measured here: of 86 refutation-shaped bullets in REFUTED.md, 83 are in the registered form. 97 %, which is better than their 16/71 but was equally unstated. The three gaps are deliberate, not a bug. They quote their claim in backticks and are bare identifiers -- +0x29d0, position = instance - 0x12c -- so registering them would match every live mention of the same offset and train the check to be ignored. Reported rather than forced to 100 %, for the same reason they report the ratio instead of demanding it: forcing a counter invites mislabelling, which is worse than the gap. Selftest and the real run both still exit 0. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_01Wuu56cE8vJGTBtn1ppsk8v --- tools/re-capture/check_refuted.py | 25 ++++++++++++++++++++++++- 1 file changed, 24 insertions(+), 1 deletion(-) diff --git a/tools/re-capture/check_refuted.py b/tools/re-capture/check_refuted.py index 64e1e740..0e0f5b27 100755 --- a/tools/re-capture/check_refuted.py +++ b/tools/re-capture/check_refuted.py @@ -135,7 +135,30 @@ for line in ref.read_text().splitlines(): if m: claims.append(m.group(1)) -print(f"{len(claims)} quoted claims in REFUTED.md\n") +# ── SCOPE, printed before the verdict ──────────────────────────────────────── +# sylpheed-port found `audit-kinds` auditing 16 of 71 authored justifications and +# never saying so -- a checker that FAILS CORRECTLY while describing a sixth of +# the corpus. Their line is the one that matters: "I checked and it was fine" and +# "I checked the part that declared itself" read identically in a log. +# +# Measured here 2026-08-31: of 86 refutation-shaped bullets in REFUTED.md, 83 are +# in the registered `* "claim"` form -- 97 %. +# +# ⚠️ THE THREE GAPS ARE NOT A BUG AND ARE NOT REGISTERED ON PURPOSE. They quote +# their claim in BACKTICKS and are bare identifiers -- `+0x29d0`, +# `position = instance - 0x12c`. Registering those would match every live mention +# of the same offset, producing permanent false hits and training the check to be +# ignored -- the unregistrable-claim limit this corpus already records for the +# 0.32 collision. Reported, not forced to 100 %. +_bul = [l for l in ref.read_text().splitlines() if l.lstrip().startswith("* ")] +_kill = [l for l in _bul if re.search(r"(→|->|—|--)\s*\*{0,2}(refuted|withdrawn|retracted" + r"|wrong|dead|no\b|it is)", l, re.I)] +_unreg = [l for l in _kill if not re.match(r'\s*\*\s*~?~?"([^"]{25,})"', l)] +_cov = 100 * (len(_kill) - len(_unreg)) / max(len(_kill), 1) +print(f"{len(claims)} quoted claims in REFUTED.md") +print(f"SCOPE: {len(_kill) - len(_unreg)} of {len(_kill)} refutation-shaped bullets are " + f"registered ({_cov:.0f}%); {len(_unreg)} quote their claim in backticks and are " + f"deliberately unregistrable\n") # 🔴 A register that loaded NOTHING cannot fail, and would report clean forever -- # the same shape as sylpheed-port's stub that prints "everything is fine" and