poster_out/GATE_REPORT.json

{
  "schema_version": 1,
  "skill": "posterly",
  "timestamp": "2026-07-19T19:17:47Z",
  "poster_html": "/repro/poster/poster.html",
  "canvas": {
    "source": "page-rule",
    "width_cm": 152.4,
    "height_cm": 91.44,
    "orientation": "landscape",
    "source_url": null
  },
  "overall": "FAIL",
  "hard_failures": 1,
  "warnings": 1,
  "soft_advisories": 8,
  "gates": [
    {
      "name": "preflight",
      "severity": "hard",
      "status": "PASS",
      "command": [
        "/usr/bin/python",
        "/repro/poster/tools/poster_check.py",
        "preflight",
        "/repro/poster/poster.html"
      ],
      "summary": {
        "exit_code": 0,
        "tail": "[preflight] /repro/poster/poster.html\n  problems: 0   warnings: 0\n[preflight] PASS"
      },
      "artifacts": []
    },
    {
      "name": "style",
      "severity": "hard",
      "status": "PASS",
      "command": [
        "/usr/bin/python",
        "/repro/poster/tools/style_check.py",
        "/repro/poster/poster.html",
        "--disable",
        "4,5",
        "--json",
        "/repro/poster_out/style_check.json"
      ],
      "summary": {
        "gate": "style",
        "status": "PASS",
        "rules": [
          {
            "id": 1,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          },
          {
            "id": 2,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          },
          {
            "id": 3,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          },
          {
            "id": 4,
            "severity": "hard",
            "status": "SKIPPED",
            "detail": "disabled via --disable (rule 4)"
          },
          {
            "id": 5,
            "severity": "hard",
            "status": "SKIPPED",
            "detail": "disabled via --disable (rule 5)"
          },
          {
            "id": 6,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          },
          {
            "id": 7,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          },
          {
            "id": 8,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          },
          {
            "id": 9,
            "severity": "warn",
            "status": "PASS",
            "detail": "9 --fs-* token(s) defined"
          },
          {
            "id": 10,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          },
          {
            "id": 11,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          },
          {
            "id": 12,
            "severity": "warn",
            "status": "PASS",
            "detail": "dark area = 0.0% of poster (<= 8%)"
          },
          {
            "id": 13,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          },
          {
            "id": 14,
            "severity": "hard",
            "status": "PASS",
            "detail": "ok"
          }
        ]
      },
      "artifacts": [
        "/repro/poster_out/style_check.json"
      ]
    },
    {
      "name": "asset",
      "severity": "hard",
      "status": "NOT_RUN",
      "command": [],
      "summary": {
        "not_run": "no --manifest: real-figure provenance gate opted out (NOT verified)"
      },
      "artifacts": []
    },
    {
      "name": "measure",
      "severity": "hard",
      "status": "FAIL",
      "command": [
        "/usr/bin/python",
        "/repro/poster/tools/poster_check.py",
        "measure",
        "/repro/poster/poster.html"
      ],
      "summary": {
        "exit_code": 1,
        "tail": "[measure] suggested adjustments:\n  shared passing band: 3079.69..3084.18 px (EVERY column bottom must land in\n  this one band; then gap and spread both pass. Anchor: footer-strip top 3122 px)\n  col0     2345.97 px -> grow ~736 px [safe +734..+738]\n  col1     2465.19 px -> grow ~617 px [safe +615..+619]\n  col2     2564.86 px -> grow ~517 px [safe +515..+519]\n  col3     2217.81 px -> grow ~864 px [safe +862..+866]\n  Tip: a body paragraph adds/removes ~25 px per wrapped line, a callout ~60-90 px,\n       a small figure ~80-150 px. Prefer trimming the tallest column first.\n[measure] edit targets -- cards per column, top-to-bottom.\n  Locate by source line (L<n>) or by grepping the quoted anchor;\n  do NOT re-read the whole file. Anchors are section-title text\n  with math stripped -- confirm uniqueness before editing.\n  col0 (grow ~736 px [safe +734..+738]):\n    card#0   L1078   h=   845px  \"1Why this reproduction\"\n    card#1   L1092   h=   488px  \"2Claim 1 \\xb7 Benchmark scale\"   <- bottom card (sets the column bottom)\n  col1 (grow ~617 px [safe +615..+619]):\n    card#2   L1109   h=   690px  \"3Claim 4 \\xb7 Script complexity\"\n    card#3   L1119   h=   762px  \"4Claim 2 \\xb7 Best model \\u2605 KEY\"   <- bottom card (sets the column bottom)\n  col2 (grow ~517 px [safe +515..+519]):\n    card#4   L1136   h=   769px  \"5Claim 3 \\xb7 Model rankings \\u2605 Headline\"\n    card#5   L1163   h=   782px  \"6Our toy substitute experiment\"   <- bottom card (sets the column bottom)\n  col3 (grow ~864 px [safe +862..+866]):\n    card#6   L1180   h=   711px  \"7Claim-by-claim verdict\"\n    card#7   L1205   h=   494px  \"8Reproduction bundle\"   <- bottom card (sets the column bottom)\n[measure] consecutive failed measurements: 2/30 (resets on PASS or --reset-budget)\nFAIL: spread 347.05 >= max 5.0\nFAIL: max gap 904.12 > 50.0\n[measure] FAIL -- alignment gate not met"
      },
      "artifacts": []
    },
    {
      "name": "polish",
      "severity": "soft",
      "status": "WARN",
      "command": [
        "/usr/bin/python",
        "/repro/poster/tools/poster_check.py",
        "polish",
        "/repro/poster/poster.html",
        "--strict"
      ],
      "summary": {
        "exit_code": 1,
        "tail": "  WARN: WIDOW: <p class='body-text fs-3 text-secondary mb-1'> wraps to a stranded last line that fills only 25% of the typeset width ('Squirrel-Semantic'), a runt (SKILL.md Gate B). Fix by REWORDING first: expand or trim a few words so the break lands on a phrase boundary and the last line carries more of the measure (if alignment is already tuned, swap a word for a longer synonym to keep the line count). On a CENTERED heading/title use text-wrap: balance instead. &nbsp;-glue is a LAST RESORT for a leading marker or a tight stat cell only -- glue at most two tokens; never chain more (a fused multi-word unit wraps early and tears a hole in the line above -- the GLUE-CHAIN gate flags it). Context: '...~30 LLMs evaluated on Squirrel-Syntax / Squirrel-Semantic'.\n  WARN: WIDOW: <p class='body-text'> wraps to a stranded last line that fills only 29% of the typeset width ('Inference Providers.'), a runt (SKILL.md Gate B). Fix by REWORDING first: expand or trim a few words so the break lands on a phrase boundary and the last line carries more of the measure (if alignment is already tuned, swap a word for a longer synonym to keep the line count). On a CENTERED heading/title use text-wrap: balance instead. &nbsp;-glue is a LAST RESORT for a leading marker or a tight stat cell only -- glue at most two tokens; never chain more (a fused multi-word unit wraps early and tears a hole in the line above -- the GLUE-CHAIN gate flags it). Context: '...omains) and ran 4 open models via HF Inference Providers.'.\n  WARN: WIDOW: <div class='callout mt-3'> wraps to a stranded last line that fills only 15% of the typeset width ('985 tasks.'), a runt (SKILL.md Gate B). Fix by REWORDING first: expand or trim a few words so the break lands on a phrase boundary and the last line carries more of the measure (if alignment is already tuned, swap a word for a longer synonym to keep the line count). On a CENTERED heading/title use text-wrap: balance instead. &nbsp;-glue is a LAST RESORT for a leading marker or a tight stat cell only -- glue at most two tokens; never chain more (a fused multi-word unit wraps early and tears a hole in the line above -- the GLUE-CHAIN gate flags it). Context: '... \\u2014 absolute scores aren't comparable at 20 vs. 985 tasks.'.\n  WARN: WIDOW: <p class='body-text fs-2 text-secondary mb-1'> wraps to a stranded last line that fills only 33% of the typeset width ('of arXiv:2601.18119'), a runt (SKILL.md Gate B). Fix by REWORDING first: expand or trim a few words so the break lands on a phrase boundary and the last line carries more of the measure (if alignment is already tuned, swap a word for a longer synonym to keep the line count). On a CENTERED heading/title use text-wrap: balance instead. &nbsp;-glue is a LAST RESORT for a leading marker or a tight stat cell only -- glue at most two tokens; never chain more (a fused multi-word unit wraps early and tears a hole in the line above -- the GLUE-CHAIN gate flags it). Context: '...t Tables 1-2 / abstract / Section 3-5 of arXiv:2601.18119'.\n  WARN: CONTRAST: <span class='key-mark'> '\\u2605 Headline' renders #c9a24a on #e8f1f8 = 2.1:1 (floor 3.0:1) (+1 more run(s) of this class/color pair). The usual cause: an inline emphasis/highlight class sets a background but INHERITS its text color from a different ground -- any class that paints a background must declare its own `color`. Fix the token pairing (use the matching *-ink token), then re-render and look.\n  WARN: CONTRAST: <strong class=''> 'Q:' renders #c9a24a on #2d5f8b = 2.8:1 (floor 3.0:1). The usual cause: an inline emphasis/highlight class sets a background but INHERITS its text color from a different ground -- any class that paints a background must declare its own `color`. Fix the token pairing (use the matching *-ink token), then re-render and look.\n  WARN: IDENTITY/WOVEN: no woven identity mark (data-ps-mark=\"woven\") -- add one registration-mark glyph riding EXISTING content (a best/target marker, a terminal period, an inline bullet, ...); it is the second identity layer beyond the always-on corner signature. Never let it solely carry a scientific claim (Ours/Best/target must stay readable from text or table styling).\n[polish] FAIL -- --strict and warnings present"
      },
      "artifacts": [],
      "advisories": 8,
      "advisory_lines": [
        "WARN: WIDOW: <p class='body-text'> wraps to a stranded last line that fills only 26% of the typeset width ('kept (paper Eq. 2):'), a runt (SKILL.md Gate B). Fix by REWORDING first: expand or trim a few words so the break lands on a phrase boundary and the last line carries more of the measure (if alignment is already tuned, swap a word for a longer synonym to keep the line count). On a CENTERED heading/title use text-wrap: balance instead. &nbsp;-glue is a LAST RESORT for a leading marker or a tight stat cell only -- glue at most two tokens; never chain more (a fused multi-word unit wraps early and tears a hole in the line above -- the GLUE-CHAIN gate flags it). Context: '...mposite complexity score before being kept (paper Eq. 2):'.",
        "WARN: WIDOW: <p class='body-text fs-3 text-secondary mb-1'> wraps to a stranded last line that fills only 25% of the typeset width ('Squirrel-Semantic'), a runt (SKILL.md Gate B). Fix by REWORDING first: expand or trim a few words so the break lands on a phrase boundary and the last line carries more of the measure (if alignment is already tuned, swap a word for a longer synonym to keep the line count). On a CENTERED heading/title use text-wrap: balance instead. &nbsp;-glue is a LAST RESORT for a leading marker or a tight stat cell only -- glue at most two tokens; never chain more (a fused multi-word unit wraps early and tears a hole in the line above -- the GLUE-CHAIN gate flags it). Context: '...~30 LLMs evaluated on Squirrel-Syntax / Squirrel-Semantic'.",
        "WARN: WIDOW: <p class='body-text'> wraps to a stranded last line that fills only 29% of the typeset width ('Inference Providers.'), a runt (SKILL.md Gate B). Fix by REWORDING first: expand or trim a few words so the break lands on a phrase boundary and the last line carries more of the measure (if alignment is already tuned, swap a word for a longer synonym to keep the line count). On a CENTERED heading/title use text-wrap: balance instead. &nbsp;-glue is a LAST RESORT for a leading marker or a tight stat cell only -- glue at most two tokens; never chain more (a fused multi-word unit wraps early and tears a hole in the line above -- the GLUE-CHAIN gate flags it). Context: '...omains) and ran 4 open models via HF Inference Providers.'.",
        "WARN: WIDOW: <div class='callout mt-3'> wraps to a stranded last line that fills only 15% of the typeset width ('985 tasks.'), a runt (SKILL.md Gate B). Fix by REWORDING first: expand or trim a few words so the break lands on a phrase boundary and the last line carries more of the measure (if alignment is already tuned, swap a word for a longer synonym to keep the line count). On a CENTERED heading/title use text-wrap: balance instead. &nbsp;-glue is a LAST RESORT for a leading marker or a tight stat cell only -- glue at most two tokens; never chain more (a fused multi-word unit wraps early and tears a hole in the line above -- the GLUE-CHAIN gate flags it). Context: '... \\u2014 absolute scores aren't comparable at 20 vs. 985 tasks.'.",
        "WARN: WIDOW: <p class='body-text fs-2 text-secondary mb-1'> wraps to a stranded last line that fills only 33% of the typeset width ('of arXiv:2601.18119'), a runt (SKILL.md Gate B). Fix by REWORDING first: expand or trim a few words so the break lands on a phrase boundary and the last line carries more of the measure (if alignment is already tuned, swap a word for a longer synonym to keep the line count). On a CENTERED heading/title use text-wrap: balance instead. &nbsp;-glue is a LAST RESORT for a leading marker or a tight stat cell only -- glue at most two tokens; never chain more (a fused multi-word unit wraps early and tears a hole in the line above -- the GLUE-CHAIN gate flags it). Context: '...t Tables 1-2 / abstract / Section 3-5 of arXiv:2601.18119'.",
        "WARN: CONTRAST: <span class='key-mark'> '\\u2605 Headline' renders #c9a24a on #e8f1f8 = 2.1:1 (floor 3.0:1) (+1 more run(s) of this class/color pair). The usual cause: an inline emphasis/highlight class sets a background but INHERITS its text color from a different ground -- any class that paints a background must declare its own `color`. Fix the token pairing (use the matching *-ink token), then re-render and look.",
        "WARN: CONTRAST: <strong class=''> 'Q:' renders #c9a24a on #2d5f8b = 2.8:1 (floor 3.0:1). The usual cause: an inline emphasis/highlight class sets a background but INHERITS its text color from a different ground -- any class that paints a background must declare its own `color`. Fix the token pairing (use the matching *-ink token), then re-render and look.",
        "WARN: IDENTITY/WOVEN: no woven identity mark (data-ps-mark=\"woven\") -- add one registration-mark glyph riding EXISTING content (a best/target marker, a terminal period, an inline bullet, ...); it is the second identity layer beyond the always-on corner signature. Never let it solely carry a scientific claim (Ours/Best/target must stay readable from text or table styling)."
      ]
    }
  ]
}