{ "gate": "hostile-read", "utc": "20260905T224559Z", "verdict": "READ — the page may NOT release as read: fourteen must-fixes, one of them a publication-law hit in the public pre-registration the page links to", "detail": "it named fourteen must-fixes in its own order, the first two of them defects rather than sentences — the public pre-registration the page links to as data/prereg.md still prints a live production port in a file whose own opening line says that port is described rather than printed, and the round's blindness reading is computed with an identity test against a boolean where one of the seven seats answers with a string, so 33 of the 103 recognised head-to-head cells and 4 more on the groundedness reading were read as not-recognised, the registered sensitivity cut dropped 70 cells instead of 103 (0.549 recomputed, against the 0.553 printed), and three of the panel's correct identifications of the rival arm went missing (16 becomes 19) — beside G4's headline counts printing with no aggregation rule and no file behind them, the row trees for both judged legs listed in neither files nor withheld while the index says nothing is absent, a deck that upgrades \"did not separate at this sample size\" into \"could not tell the two apart\", a window sentence and three section stamps that exclude the page's own probes and its pen scan, a no-mode denominator on the author's own arm that the page's own truncation rule forbids, a pen scan called measured four hours before eight of the kit files it covers existed, a NOT-RUN leg that was never registered and whose stated reason appears in no artifact, and a human read of this page asserted in the past tense before the page has gone out", "evidence": { "findings_raised": 54, "must_fixes": 14, "must_fixes_landed": 14 }, "class_counts": { "A": 30, "B": 14, "C": 10, "D": 0, "E": 0 }, "must_fixes_named": [ { "name": "the public prereg copy prints a live production port (a production port) its own opening line says is described", "section": "How to check our work / results/kit/prereg.md §12 A1" }, { "name": "recognition read with `is True` against a string: 70 of 103 cells cut, 0.553 should read 0.549, 16 should read 19", "section": "Head-to-head — How blind was the blind, here / The rules desk — G4 recognition table" }, { "name": "G4's arm-level GROUNDED prints with no aggregation rule and no re-derivable source", "section": "The rules desk — G4" }, { "name": "\"the seal manifest with the shas of everything we did not publish\" — it holds seven keys, and five row trees are unindexed", "section": "How to check our work" }, { "name": "the deck claims the judges could not tell the two apart; no test was registered and the rows disagree", "section": "the deck" }, { "name": "\"every call on this page was made\" inside the window, and three section stamps mis-date their contents", "section": "Three weeks earlier / The ledger / What this page does not say / The bill" }, { "name": "the no-mode cell gives the author's own arm a denominator of 60 where its case set is 54", "section": "The rules desk — what came back, call by call" }, { "name": "\"No personal data ... measured\" rests on a scan taken before eight of the kit's files existed", "section": "The ledger, and what left our machines" }, { "name": "\"Four more records ship under receipts/\" — nine do", "section": "The ledger — the round's own probes" }, { "name": "\"most of them narrowly\" — 5 of 36 votes sit within a tenth of one half; 6 were unanimous", "section": "Head-to-head" }, { "name": "the outside reader rated two findings DECISIVE; the page names the one it folded", "section": "Who judged, and who wrote this" }, { "name": "the cove leg is printed NOT-RUN but was never registered, and its stated reason is in no artifact", "section": "Two strangers at the door" }, { "name": "an operator's stranger read of this page asserted in the past tense, unreceipted, before release", "section": "Who ran this, and thanks" }, { "name": "NOT-COLLECTED — TRANSPORT prints twice on the page and is absent from the state table", "section": "The words this page leans on" } ], "v3_must_fixes_landed_as_verified_by_this_reader": 10, "version_read": 5, "supersedes": [ "20260905T181859Z-hostile-read.json", "20260905T195339Z-hostile-read.json", "20260905T195711Z-hostile-read.json", "20260905T203537Z-hostile-read.json", "20260905T210832Z-hostile-read.json" ], "critique": "article/panel-v5/CRITIQUE-hostile.md", "critique_sha256": "97ca1917c186f2c82a2f0b3d34a17a03fc1246e0ebb9699f05d37261bd48fdca", "must_fixes_landed_in_version": 6, "landed_note": "all fourteen landed in draft v6 (PREREG A19): the recognition flag read as the seats write it and both judged legs rescored; G4's aggregation rule printed with the per-case verdict file; the judged legs' row trees under withheld; the deck, the timeline sentence, the no-mode denominator, the pen scan re-run over the built kit, the unrowed receipts counted, the per-case shape counted, both DECISIVE findings named, the cove leg's recorded reason, an operator's read receipted, TRANSPORT and BLIND-LEAK in the state table, and the public copy's second port described", "verify_pass": { "version": 11, "utc": "20260906T011845Z", "must_fixes_verified_landed": 1, "new_must_fixes": 0, "notes": [ "delta verify of v11 (md5 3353c4d296ecab98a8d1f3094308dbca, 588 lines, author copy still v0.15): the one must-fix outstanding, V10-1, is landed; no new must-fix; CLEAR TO RELEASE", "V10-1 landed: 'before it went out' is gone (0 hits) and an operator clause reads 'read a draft of this page on the drafts shelf on 2026-09-05, and found working notes at the foot of the page; it was fixed' - the receipted locator restored, the tense claim dropped, and still no draft number anywhere in the credits", "the diff is five line groups and nine changed lines: the byline (5) and the credits paragraph's operator clause (541); the ledger stamp (367) and the pen-scan sentence (396) re-stamped 01:07Z -> 01:15Z, and one new G-PEN ledger row (420) - the kit stamps this pour's own re-scan regenerates", "everything re-checked and reconciling: 27 ledger rows = 22 shipped + 5 withheld with none unresolved, 33 shipped receipts minus 22 rowed = the 11 unrowed the clause prints, 6 hostile-read receipts shipping against the page's 6, 5 withheld against its Five, amendments 21 / 15 / 6, and my receipt still PUBLISHED with the CURRENT label and both port classes described and counted under scrubbed", "no measurement moved: every scorer file's mtime predates the 01:15Z pour; G1 36/34/34, G3a 7/7/10, G5a 0/0/2, legC 18/18 and 18/18 and 12/12, calibration 102 readings at 0.983, 103 of 449 recognised, the cut 0.549 [0.447, 0.644], the headline 0.544, G4 58/19 and 57/7 and 34/0, the bill 23.05 / 17.79 / 5.25 / 31.44, 42 Wilson intervals correct and none on N < 30, 37 fills all verbatim but the hub front-matter stamp", "publication law clean: 0 hits on v11; across results/kit/ the strict port pattern returns only the two Python slices in scorers the round writes, and no bare port digits survive", "should-fix, in the fill rather than by hand: 'It then read a later draft against those changes: 0 of 0 landed as required, 1 new must-fix raised' is a faithful projection of the v10 block, but '0 of 0 landed as required' reads as vacuous or as a failure to a stranger. When the denominator is zero, render 'carried no outstanding must-fix' and print the new count after it", "note: that same clause is one pass behind, so it reports the v10 must-fix without saying what became of it - it is V10-1 and v11 is its remedy, which the head already states ('the changes it required are in this one'). With this block (version 11, 1 of 1 landed, 0 new) a re-pour would print 'It then read a later draft against those changes: 1 of 1 landed as required, 0 new must-fixes raised'. That re-pour is optional, not required: nothing on v11 is false without it. I considered raising the rendering as a must-fix and did not, because doing so would print a new open count on the next pour and start the loop this page has already been round twice", "note: the pour again writes the page about a second after the final kit build (index 01:15:26.206Z, page 01:15:27.233Z), as at v8 to v10; substance verified - all 27 ledger rows resolve and the page's pen stamps read this pour's own 01:15Z scan", "verify report: article/panel-v7/VERIFY-hostile.md, section 'V11 DELTA' (appended)" ], "history": [ { "version": 6, "utc": "20260905T233309Z", "must_fixes_verified_landed": 14, "new_must_fixes": 5, "notes": [ "all 14 verified landed in v6 against the page, the rebuilt kit (index built 23:13:56Z), the re-run scorers and PREREG A19; evidence.must_fixes_landed 14 and must_fixes_landed_in_version 6 stand", "re-derived from the row files, all matching: 103 of 449 recognised cells; the cut 0.549 with bootstrap [0.447, 0.644] and cluster_sd 0.302; G4 58 claims / 19 right on GPT-6 Astra and 57 / 17 on Claude Fable 5.1; the per-case shape 5 within a tenth of one half, 3 unanimous for GPT-6 Astra and 1 for Claude Fable 5.1", "nothing else moved: every gate cell, the calibration's 102 readings, the bill and all three Leg C rows unchanged; all 42 Wilson intervals correct and none on N < 30", "the publication-law grep is clean on the page and on results/kit/prereg.md (both ports described)", "NEW must-fix: results/kit/receipts/20260905T224559Z-hostile-read.json ships with the port literal in must_fixes_named[0]; the kit screen applies 45 rules and none is a port, so screening.clean is true of its rules and false of the round's law", "NEW must-fix: the amendments bullet carries a dangling fragment — 'what the kit ships made while the round ran and after it'", "NEW must-fix: the probe ledger prints 'not listed' for the 23:13:55Z pen receipt that index.json lists PUBLISHED; fills.json was stamped one second before the index that ships, so the A19 (6) pour order was not followed", "NEW must-fix: the bill's stamp dates its records to 2026-09-05 04:39Z; the earliest record the bill counts is 14:03:58Z", "NEW must-fix: the shipped freeze index says index_shas.py --check disagrees on nine rows where it prints eleven (sample_record and report_build unnamed), and index.json's post_seal_reporting_additions names A16 where the shipped scorers are post-A19", "should-fix: the shelf roster read (04:37:29Z) postdates the 01:54Z co-sign, so '4 records from before the round's co-sign' is right for three of four; the GROUNDED paragraph's unanimity gloss reads as a ratio; Limits and the takeaways are now undated and the scaffolding delta with them", "verify report: article/panel-v6/VERIFY-hostile.md (page md5 1740a1cef295d0795e874af38a6471cc)" ] }, { "version": 7, "utc": "20260906T002732Z", "must_fixes_verified_landed": 5, "new_must_fixes": 0, "notes": [ "all 5 new must-fixes and all 3 should-fixes of the v6 verify pass landed in v7 under PREREG A20; nothing new must not publish", "the port is gone from the kit entirely: grep -rn 'a production port|a production port' results/kit/ -> 0 (v6: 1, in this receipt's own shipped copy); kit_build.PORT_RE imports public_copy.OTHER_PORT_RE so one definition serves the prereg cut and the screen; a written kit file carrying a port refuses the build; a receipt or reader copy has it described and counted under scrubbed as 'port-shaped literal -> \"a production port\"'; the bench receipt still carries the literal once and the bench critique four times", "the pour order is self-enforcing: article_fill.py:2329 refuses the fill when a probe receipt is in neither files nor withheld; the ledger's 23 rows read 18 shipped and 5 withheld, 'not listed' -> 0, all four G-PEN rows shipped", "the bill emits records_window and the stamp reads its earliest (14:03Z); my own recompute over the same globs reproduces it: arms 985 records 14:03:58Z-15:27:59Z, seats 1232 14:57:12Z-18:14:35Z, 2217 together", "index_shas.py --check reports 11 rows and results/kit/prereg-index.md names all 11; post_seal_reporting_additions names A16-A20; I verified the stronger claim under it — each of the nine scorer rows' recorded sha is the file's sha at the seal commit 11475ac, and 7d65882b is the sealed prereg at 6a47f48 which carries A1-A15, so the drift begins at A16", "the dangling fragment is gone; the unrowed-records sentence splits its eleven correctly (6 reader passes + 1 operator read + 3 records from 2026-09-04 + the shelf roster read); the GROUNDED unanimity gloss has its own sentence; both missing stamps are back, the scaffolding delta dated to its receipt's 14:04Z", "the lane's three deviations are ACCEPTED on evidence I reproduced: a Python slice is not a port (the (? 00:32Z and one new G-PEN ledger row (417), all three the kit stamps this pour's own re-scan regenerates; and the unrowed-records sentence (428), which is kit-derived and carries the new must-fix", "the new standfirst and opening paragraph are author prose, not fills, and introduce no number beyond the registered thirty-six cases; the citation contract they state matches legA-gates.json G1's floor_source (cite_survival AND stripped_count == 0, forged markers counted separately) and the abstention half matches G3a's floor; the example marker [2] is the shape the quoted reply on ofl-ans-0001 uses", "NEW must-fix: the kit now WITHHOLDS the current hostile-read receipt (this file) because my own v7 verify note quoted the two production port numbers inside a grep command, which fires the screen's bare-number rule 77 rather than the port-shape rule the A20 receipts scrub applies. The screen was right to refuse it. But the page's clause still reads 'the 5 receipts of the adversarial reader's passes - one per pass, append-only, the newest the current one', when the newest that ships is the v3 pass's receipt and the current one is withheld; and all five shipped rows carry a SUPERSEDED label naming a path the same index lists under withheld, without saying so. The counts are right (29 shipped receipts - 19 rowed = 10; 5+1+3+1). Fix the clause the way the page already handles the withheld critique, and have a SUPERSEDED label name the withholding when its target is not in files; a bare-port-number scrub on receipt copies is the durable alternative. The bench bytes must not move, which is why the offending note is preserved in history rather than rewritten", "note: the credits head still reads 'this is version 8, which carries the changes it required' and not 'through this version of the page' (0 hits) - article_fill.py:2690 sets verified_this_one = (verify_pass.version == FILLED_VERSION and new_must_fixes == 0) and the block it read named version 7. The verify sentence itself closed correctly on that block ('read version 7 ... 5 of 5 landed as required, 0 new must-fixes raised'). This block names 8 with 1 new must-fix, so the head will stay in its current honest form until that must-fix is folded and a pass records 0 new; nothing on the page is false as it stands", "note: the pour gate the brief set is not literally met - index.json mtime 00:32:26.961Z against the page's 00:32:27.993Z, so the .md is written about a second after the final kit build. The substance holds and I checked it: all 24 ledger rows resolve against the shipped index and the page's pen stamps read this pour's own 00:32Z scan", "note: a kit count did move, against the brief's 'no kit count moves' - withheld 31 -> 32, with the shipped receipt set turning over one for one (this receipt out, the pour's new pen receipt in), so files stayed 61 and shipped receipts stayed 29. Both are consequences of the must-fix above and of the registered re-scan, not of the prose", "no measurement moved: every scorer file's mtime predates the pour; G1 36/34/34, G3a 7/7/10, G5a 0/0/2, calibration 102 readings at 0.983, 103 of 449 recognised, the cut 0.549 [0.447, 0.644], G4 58/19 and 57/7 and 34/0, the bill 23.05 / 17.79 / 5.25 / 31.44, 20 amendments, 42 Wilson intervals correct and none on N < 30", "publication law clean: 0 hits on v8 and, across results/kit/, only pinned :0813 model tags, public-copy.py's own port regex and comments, and two Python slices in scorers the round writes", "verify report: article/panel-v7/VERIFY-hostile.md, section 'V8 DELTA' (appended, not a new file)" ] }, { "version": 9, "utc": "20260906T010606Z", "must_fixes_verified_landed": 1, "new_must_fixes": 0, "notes": [ "delta verify of v9 (md5 e5186c30c417a2ae0ed9260b03a57d95, 586 lines, author copy unchanged at v0.15): the ONE must-fix raised on v8 is landed and closed on all three of its surfaces; nothing new must not publish. must_fixes_verified_landed reads 1 of 1 because one must-fix was outstanding, not five", "the diff is seven line groups: the byline (5) and the credits sentence (539) as expected; the ledger stamp (367) and the pen-scan sentence (396) re-stamped 00:32Z -> 00:59Z plus one new G-PEN ledger row (418), the kit stamps this pour's own re-scan regenerates; the ledger clause (429) '10 more records ... the 5 receipts' -> '11 ... the 6 receipts', which is the fix; and the amendments bullet (526) '20 dated amendments, 5 after the seal' -> '21 ... 6', which the brief's expected list did not name but which registering A21 requires", "A21 verified where it lands: my receipt is PUBLISHED in index.json (state PUBLISHED, sha adfae098), no longer in withheld, and its scrubbed block counts both classes - one colon-form from the v6 note and two bare-number forms from the v7 note, each described as the words 'a production port'; no bare port digits survive anywhere in the kit; the round's own strict port rule sees zero in the file (its remaining colon-digit strings are two article_fill.py line references and the pinned model tag)", "the ledger clause now reconciles from the index rather than by hand: 25 ledger rows = 20 shipped + 5 withheld with none unresolved, 31 shipped receipts - 20 rowed = 11 unrowed (the page says 11), 6 hostile-read receipts ship (the page says 6), 5 withheld receipts (the page says Five), and my receipt carries the CURRENT label while the five older ones say SUPERSEDED - so 'the newest the current one' is true for the first time since v7", "the credits check against both receipts to the digit: version_read 5, 54 findings, 14 changes every one landed, the critique withheld as readers/v5-hostile.md; the verify sentence 'read version 8 ... 5 of 5 landed as required, 1 new must-fix raised' is the v8 block exactly; and an operator sentence's version 4 and 'landed in version 6' are an operator-read receipt's version_read and found[0].landed_in", "no measurement moved: every scorer file's mtime predates the 00:59Z pour; G1 36/34/34, G3a 7/7/10, G5a 0/0/2, legC 18/18 and 18/18 and 12/12, calibration 102 readings at 0.983, 103 of 449 recognised, the cut 0.549 [0.447, 0.644], the headline 0.544 [0.444, 0.638], G4 58/19 and 57/7 and 34/0, the bill 23.05 / 17.79 / 5.25 / 31.44, records_window.earliest 14:03:58Z, 42 Wilson intervals correct and none on N < 30, 37 fills all verbatim but the hub front-matter stamp", "publication law clean: 0 hits on v9; across results/kit/ the strict port pattern returns only the two Python slices in scorers the round writes, and screening reports 46 rules, clean, no literal hits", "note, and a miss of my own: 'Five receipts are withheld from the kit' was FALSE in v8 - once the screen withheld my receipt there were six - and my v8 pass checked the ledger clause's counts without re-reading this sentence. v9 makes it true again as a side effect of the fold, not because anyone caught it. When a file moves between files and withheld, every sentence counting either set has to be re-read", "note: the scrub collapses the two production ports to the same description, so the shipped copy of my v7 note reads as a doubled phrase in one grep command; disclosed under scrubbed, in a receipt and not on the page. A numbered description would read better if it recurs", "note: the credits head still does not read 'through this version of the page' (0 hits), because the block it read named version 8 with one new must-fix outstanding. harness/article_fill.py already reads FILLED_VERSION 10 while article/pour/ ends at v9, so a v10 is expected; the planned credits rewrite with no draft numbers in the prose would retire this head/version loop rather than iterate it. The poll for v10 is armed", "verify report: article/panel-v7/VERIFY-hostile.md, section 'V9 DELTA' (appended)" ] }, { "version": 10, "utc": "20260906T011239Z", "must_fixes_verified_landed": 0, "new_must_fixes": 1, "notes": [ "delta verify of v10 (md5 8ecd58c38993c80812729d2103d949b4, 587 lines, author copy still v0.15): no must-fix was outstanding going in, because the v9 pass raised none, so must_fixes_verified_landed reads 0 of 0 and is not a failure; the credits rewrite raises ONE new must-fix", "the diff is five line groups: the byline (5) and the credits paragraph (540) as expected; the ledger stamp (367) and the pen-scan sentence (396) re-stamped 00:59Z -> 01:07Z, and one new G-PEN ledger row (419) - the kit stamps this pour's own re-scan regenerates. Nine changed lines and nothing else, and I checked why the ledger clause and the amendments bullet did not need to move: the pour added one receipt and one row together, so 32 shipped receipts minus 21 rowed still leaves the 11 unrowed the clause prints", "the credits rewrite carries zero draft numbers, as asked, and every receipted claim survives it: the co-sign before the first call; the egress ruling; an operator's read on 2026-09-05 with the working notes found and fixed; three reader passes shipping under readers/; 'a hostile reader has been through a draft of this page; the changes it required are in this one'; the late draft's 54 findings and 14 changes every one landed; the critique withheld with its sha; and 'It then read a later draft against those changes: 1 of 1 landed as required, 0 new must-fixes raised', which is the verify_pass version 9 block exactly", "NEW must-fix: 'read a draft of this page on 2026-09-05 BEFORE IT WENT OUT' asserts a completed publication that has not happened - the byline still reads Draft v10 and release/cut40.sh times the pour for 03:37. Drafts v6 to v9 located the same read where the receipt locates it, 'on the drafts shelf'; v10 dropped the draft number as asked but swapped the locator for a tense the receipt does not support (an operator receipt contains no 'went out' string, and its own note cites must-fix 13, 'never a review it has no receipt for', as the rule it was written under). Fix while keeping the no-numbers instruction: 'and read a draft of this page on the drafts shelf on 2026-09-05, and found working notes at the foot of the page; it was fixed'", "note: dropping the critique's filename leaves two withheld hostile critiques in the index, distinguished in the sentence only by 'its sha beside it'. The chain holds because the receipt's critique_sha256 matches the right row, but the page now needs the receipt to disambiguate what it used to name", "note: the rewrite retires by design the head/version loop I raised twice - the head no longer depends on verify_pass.version matching FILLED_VERSION, so it can state the truth on the pour that carries the block instead of trailing it by one. That is the right way to have closed it", "note: the pour again writes the page about a second after the final kit build (index 01:07:36.046Z, page 01:07:37.057Z). Substance holds and I checked it: all 26 ledger rows resolve and the page's pen stamps read this pour's own 01:07Z scan", "no measurement moved: every scorer file's mtime predates the 01:07Z pour; G1 36/34/34, G3a 7/7/10, G5a 0/0/2, legC 18/18 and 18/18 and 12/12, calibration 102 readings at 0.983, 103 of 449 recognised, the cut 0.549 [0.447, 0.644], the headline 0.544, G4 58/19 and 57/7 and 34/0, the bill 23.05 / 17.79 / 5.25 / 31.44, 42 Wilson intervals correct and none on N < 30, 37 fills all verbatim but the hub front-matter stamp", "publication law clean: 0 hits on v10; across results/kit/ the strict port pattern returns only the two Python slices in scorers the round writes, and no bare port digits survive; my receipt is still PUBLISHED with the CURRENT label and both port classes described and counted under scrubbed", "FILLED_VERSION in harness/article_fill.py reads 10, so this verify is the one the stop rule points at; the must-fix above will likely produce a v11, and a 20-minute poll for it is armed", "verify report: article/panel-v7/VERIFY-hostile.md, section 'V10 DELTA' (appended)" ] } ] } }