paperlint 9.0.1 → 10.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CONTRIBUTING.md +16 -4
- package/README.md +68 -128
- package/dist/build.d.ts +1 -1
- package/dist/build.js +1 -1
- package/dist/cli.d.ts +5 -1
- package/dist/cli.d.ts.map +1 -1
- package/dist/cli.js +52 -6
- package/dist/cli.js.map +1 -1
- package/dist/domain/layout-tree.d.ts +19 -0
- package/dist/domain/layout-tree.d.ts.map +1 -0
- package/dist/domain/layout-tree.js +32 -0
- package/dist/domain/layout-tree.js.map +1 -0
- package/dist/domain/lookup-cache.d.ts +2 -2
- package/dist/domain/lookup-cache.js +2 -2
- package/dist/domain/paper-layout.d.ts +81 -0
- package/dist/domain/paper-layout.d.ts.map +1 -0
- package/dist/domain/paper-layout.js +259 -0
- package/dist/domain/paper-layout.js.map +1 -0
- package/dist/domain/repro-layout.d.ts +80 -0
- package/dist/domain/repro-layout.d.ts.map +1 -0
- package/dist/domain/repro-layout.js +247 -0
- package/dist/domain/repro-layout.js.map +1 -0
- package/dist/domain/scaffold-layout.d.ts +28 -0
- package/dist/domain/scaffold-layout.d.ts.map +1 -0
- package/dist/domain/scaffold-layout.js +81 -0
- package/dist/domain/scaffold-layout.js.map +1 -0
- package/dist/eslint-rules/js-needs-reason.d.ts +55 -0
- package/dist/eslint-rules/js-needs-reason.d.ts.map +1 -0
- package/dist/eslint-rules/js-needs-reason.js +35 -0
- package/dist/eslint-rules/js-needs-reason.js.map +1 -0
- package/dist/eslint-rules/prereg-coverage.d.ts +98 -0
- package/dist/eslint-rules/prereg-coverage.d.ts.map +1 -0
- package/dist/eslint-rules/prereg-coverage.js +274 -0
- package/dist/eslint-rules/prereg-coverage.js.map +1 -0
- package/dist/folder-layout-hook.d.ts +21 -0
- package/dist/folder-layout-hook.d.ts.map +1 -0
- package/dist/folder-layout-hook.js +117 -0
- package/dist/folder-layout-hook.js.map +1 -0
- package/dist/folder-layout-rule.d.ts +22 -0
- package/dist/folder-layout-rule.d.ts.map +1 -0
- package/dist/folder-layout-rule.js +107 -0
- package/dist/folder-layout-rule.js.map +1 -0
- package/dist/init.js +1 -1
- package/dist/init.js.map +1 -1
- package/dist/new-paper.d.ts +4 -1
- package/dist/new-paper.d.ts.map +1 -1
- package/dist/new-paper.js +40 -6
- package/dist/new-paper.js.map +1 -1
- package/dist/package-dirs.d.ts +6 -0
- package/dist/package-dirs.d.ts.map +1 -1
- package/dist/package-dirs.js +6 -0
- package/dist/package-dirs.js.map +1 -1
- package/dist/references.d.ts +1 -1
- package/dist/references.d.ts.map +1 -1
- package/dist/references.js +2 -2
- package/dist/references.js.map +1 -1
- package/dist/repro-layout-rule.d.ts +16 -0
- package/dist/repro-layout-rule.d.ts.map +1 -0
- package/dist/repro-layout-rule.js +81 -0
- package/dist/repro-layout-rule.js.map +1 -0
- package/dist/skills/label-agent-runs/scripts/labels-commands.d.ts +13 -0
- package/dist/skills/label-agent-runs/scripts/labels.d.ts +2 -0
- package/dist/skills/run-agent-grid/scripts/grid-guard-commands.d.ts +20 -0
- package/dist/skills/run-agent-grid/scripts/grid-guard.d.ts +2 -0
- package/docs/ci.md +60 -0
- package/docs/configuration.md +2 -0
- package/docs/e2e.md +2 -2
- package/docs/install.md +20 -16
- package/docs/layout.md +154 -0
- package/docs/optional-rules.md +5 -4
- package/docs/references.md +5 -2
- package/docs/rules/paper/allowed-files-only.md +192 -0
- package/docs/rules/repro/allowed-files-only.md +128 -0
- package/docs/rules/study/declare-sample.md +146 -0
- package/docs/rules.md +53 -50
- package/docs/skills.md +8 -3
- package/docs/testing.md +13 -0
- package/docs/venues.md +94 -0
- package/eslint-rules/js-needs-reason.test.ts +141 -0
- package/eslint-rules/js-needs-reason.ts +77 -0
- package/eslint-rules/prereg-coverage.test.ts +448 -0
- package/eslint-rules/prereg-coverage.ts +419 -0
- package/eslint.config.mjs +22 -2
- package/lib/skill-trigger-cases.harness.mjs +51 -5
- package/lib/skill-trigger-cases.mjs +53 -15
- package/lib/skill-trigger-cases.test.ts +211 -0
- package/package.json +8 -4
- package/plugin/hooks/hooks.json +9 -0
- package/scripts/coverage-config.test.ts +5 -3
- package/skills/analyze-sibling-paper/SKILL.md +2 -2
- package/skills/analyze-sibling-paper/SKILL.md.spec.ts +1 -1
- package/skills/build-benchmark/SKILL.md +17 -1
- package/skills/build-benchmark/SKILL.md.spec.ts +16 -0
- package/skills/build-label-page/SKILL.md +120 -0
- package/skills/build-label-page/SKILL.md.spec.ts +131 -0
- package/skills/build-label-page/build-label-page.eval.mjs +15 -0
- package/skills/build-label-page/build-label-page.test.ts +959 -0
- package/skills/build-label-page/references/checklist.md +148 -0
- package/skills/build-label-page/references/example-generator.py +803 -0
- package/skills/find-venue/SKILL.md +5 -3
- package/skills/find-venue/SKILL.md.spec.ts +4 -2
- package/skills/grade-paper-writing/SKILL.md +7 -7
- package/skills/grade-paper-writing/SKILL.md.spec.ts +6 -6
- package/skills/label-agent-runs/SKILL.md +103 -0
- package/skills/label-agent-runs/SKILL.md.spec.ts +106 -0
- package/skills/label-agent-runs/label-agent-runs.eval.ts +55 -0
- package/skills/label-agent-runs/label-agent-runs.harness.ts +123 -0
- package/skills/label-agent-runs/references/formats.md +64 -0
- package/skills/label-agent-runs/scripts/labels-commands.js +13 -0
- package/skills/label-agent-runs/scripts/labels-commands.ts +13 -0
- package/skills/label-agent-runs/scripts/labels.js +396 -0
- package/skills/label-agent-runs/scripts/labels.test.ts +538 -0
- package/skills/label-agent-runs/scripts/labels.ts +501 -0
- package/skills/map-prior-work/SKILL.md +2 -2
- package/skills/map-prior-work/SKILL.md.spec.ts +1 -1
- package/skills/paper-pipeline/SKILL.md +15 -6
- package/skills/paper-pipeline/SKILL.md.spec.ts +14 -5
- package/skills/paper-pipeline/grade-paper-writing-ablation.eval.mjs +19 -9
- package/skills/paper-pipeline/references/acceptance-gate.md +19 -19
- package/skills/paper-pipeline/references/artifact-checklist.md +5 -0
- package/skills/paper-pipeline/references/pipeline-status-template.md +30 -1
- package/skills/paper-pipeline/repro/2026-08-07-fork-check.log +1 -1
- package/skills/paper-pipeline/repro/2026-08-07-fork-check2.log +1 -1
- package/skills/paper-pipeline/repro/2026-08-07-language-eval-pilot.log +1 -1
- package/skills/paper-pipeline/repro/2026-08-07-language-eval-raw.log +1 -1
- package/skills/paper-pipeline/repro/2026-08-08-framing-vs-vocabulary-oracle.log +48 -48
- package/skills/paper-pipeline/repro/2026-08-08-grade-paper-writing-ablation-A6-oracle.log +8 -8
- package/skills/paper-pipeline/repro/2026-08-08-grade-paper-writing-ablation-oracle.log +48 -48
- package/skills/paper-pipeline/repro/2026-08-08-parent-replication-gpw.log +2 -2
- package/skills/paper-pipeline/repro/2026-08-08-parent-replication.mjs +3 -3
- package/skills/paper-pipeline/repro/2026-10-13-grade-paper-writing-ablation-preflight-after.log +81 -0
- package/skills/paper-pipeline/repro/2026-10-13-grade-paper-writing-ablation-preflight-before.log +81 -0
- package/skills/paper-pipeline/repro/2026-10-13-grade-paper-writing-ablation-setupdiff.log +59 -0
- package/skills/paper-pipeline/repro/2026-10-13-grade-paper-writing-trigger.jsonl +1 -0
- package/skills/paper-pipeline/repro/2026-10-13-grade-paper-writing-trigger.log +22 -0
- package/skills/paper-pipeline/repro/README.md +4 -0
- package/skills/paper-pipeline/scripts/extract-ref-facts.harness.mjs +2 -2
- package/skills/paper-pipeline/scripts/extract-ref-facts.mjs +2 -2
- package/skills/paper-pipeline/scripts/extract-ref-facts.test.mjs +4 -2
- package/skills/paper-pipeline/scripts/generated-code.harness.mjs +7 -2
- package/skills/paper-pipeline/scripts/generated-code.mjs +8 -4
- package/skills/paper-pipeline/scripts/generated-code.test.ts +15 -1
- package/skills/paper-status/SKILL.md +2 -2
- package/skills/paper-status/SKILL.md.spec.ts +1 -1
- package/skills/run-agent-grid/SKILL.md +181 -0
- package/skills/run-agent-grid/SKILL.md.spec.ts +184 -0
- package/skills/run-agent-grid/references/failure-catalog.md +36 -0
- package/skills/run-agent-grid/references/registry-and-ledger.md +142 -0
- package/skills/run-agent-grid/run-agent-grid.eval.ts +59 -0
- package/skills/run-agent-grid/run-agent-grid.harness.ts +115 -0
- package/skills/run-agent-grid/scripts/grid-guard-commands.js +18 -0
- package/skills/run-agent-grid/scripts/grid-guard-commands.ts +21 -0
- package/skills/run-agent-grid/scripts/grid-guard.js +700 -0
- package/skills/run-agent-grid/scripts/grid-guard.test.ts +1090 -0
- package/skills/run-agent-grid/scripts/grid-guard.ts +869 -0
- package/skills/study-accepted-papers/SKILL.md +163 -161
- package/skills/study-accepted-papers/SKILL.md.spec.ts +167 -165
- package/skills/study-accepted-papers/references/base-rates.md +69 -0
- package/skills/study-accepted-papers/references/mechanisms.md +181 -0
- package/skills/verify-citations/SKILL.md +2 -2
- package/skills/verify-citations/SKILL.md.spec.ts +1 -1
- package/src/build.ts +1 -1
- package/src/cli.harness.mjs +42 -0
- package/src/cli.test.ts +90 -6
- package/src/cli.ts +78 -5
- package/src/domain/layout-tree.test.ts +32 -0
- package/src/domain/layout-tree.ts +47 -0
- package/src/domain/lookup-cache.ts +2 -2
- package/src/domain/paper-layout.test.ts +264 -0
- package/src/domain/paper-layout.ts +336 -0
- package/src/domain/repro-layout.test.ts +222 -0
- package/src/domain/repro-layout.ts +311 -0
- package/src/domain/scaffold-layout.ts +104 -0
- package/src/folder-layout-hook.test.ts +214 -0
- package/src/folder-layout-hook.ts +141 -0
- package/src/folder-layout-rule.test.ts +235 -0
- package/src/folder-layout-rule.ts +151 -0
- package/src/hooks-settings.harness.ts +9 -8
- package/src/init.ts +1 -1
- package/src/layout-docs.test.ts +52 -0
- package/src/new-paper.harness.mjs +23 -1
- package/src/new-paper.test.ts +299 -6
- package/src/new-paper.ts +51 -5
- package/src/package-dirs.ts +8 -0
- package/src/paper-files.test.ts +9 -6
- package/src/references.test.ts +5 -5
- package/src/references.ts +2 -2
- package/src/repro-layout-rule.test.ts +114 -0
- package/src/repro-layout-rule.ts +121 -0
- package/src/rule-cases.test.ts +98 -0
- package/templates/paper/PIPELINE-STATUS.md +41 -7
- package/tsconfig.base.json +22 -0
- package/tsconfig.json +6 -24
- package/tsconfig.pkg.json +3 -1
- package/tsconfig.skills.json +14 -0
- package/tsconfig.src.json +12 -0
|
@@ -18,106 +18,106 @@ A5 renamed (description identical) 1174 94 0 0 0.85 0.78
|
|
|
18
18
|
oracle: 6 arms x 8 closed-book picks over 37 descriptions
|
|
19
19
|
|
|
20
20
|
=== A0 baseline (verbatim)
|
|
21
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
21
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
22
22
|
AH HIT grade-paper-writing grade the writing craft in this draft and name the offendi
|
|
23
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
23
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
24
24
|
AH HIT grade-paper-writing run the blind multi-grader panel over the prose clarity an
|
|
25
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
25
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
26
26
|
AH HIT grade-paper-writing score the title, the abstract and the intro architecture o
|
|
27
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
27
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
28
28
|
AH HIT grade-paper-writing grade how well this reads, hedge stacking included, and gi
|
|
29
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
29
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
30
30
|
SH HIT grade-paper-writing the abstract reads like a wall of jargon
|
|
31
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
31
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
32
32
|
SH HIT grade-paper-writing the intro architecture is fine but the prose clarity is no
|
|
33
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
33
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
34
34
|
SH miss tighten-paper there is hedge stacking everywhere and the point never lan
|
|
35
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
35
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
36
36
|
SH HIT grade-paper-writing the writing does not read well even though the draft is fi
|
|
37
37
|
=== A1 short (sentences 1-2 only)
|
|
38
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
38
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
39
39
|
AH HIT grade-paper-writing grade the writing craft in this draft and name the offendi
|
|
40
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
40
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
41
41
|
AH HIT grade-paper-writing run the blind multi-grader panel over the prose clarity an
|
|
42
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
42
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
43
43
|
AH HIT grade-paper-writing score the title, the abstract and the intro architecture o
|
|
44
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
44
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
45
45
|
AH HIT grade-paper-writing grade how well this reads, hedge stacking included, and gi
|
|
46
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
46
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
47
47
|
SH HIT grade-paper-writing the abstract reads like a wall of jargon
|
|
48
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
48
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
49
49
|
SH HIT grade-paper-writing the intro architecture is fine but the prose clarity is no
|
|
50
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
50
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
51
51
|
SH miss tighten-paper there is hedge stacking everywhere and the point never lan
|
|
52
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
52
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
53
53
|
SH HIT grade-paper-writing the writing does not read well even though the draft is fi
|
|
54
54
|
=== A2 no-subordination
|
|
55
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
55
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
56
56
|
AH HIT grade-paper-writing grade the writing craft in this draft and name the offendi
|
|
57
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
57
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
58
58
|
AH HIT grade-paper-writing run the blind multi-grader panel over the prose clarity an
|
|
59
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
59
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
60
60
|
AH HIT grade-paper-writing score the title, the abstract and the intro architecture o
|
|
61
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
61
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
62
62
|
AH HIT grade-paper-writing grade how well this reads, hedge stacking included, and gi
|
|
63
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
63
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
64
64
|
SH HIT grade-paper-writing the abstract reads like a wall of jargon
|
|
65
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
65
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
66
66
|
SH HIT grade-paper-writing the intro architecture is fine but the prose clarity is no
|
|
67
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
67
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
68
68
|
SH HIT grade-paper-writing there is hedge stacking everywhere and the point never lan
|
|
69
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
69
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
70
70
|
SH HIT grade-paper-writing the writing does not read well even though the draft is fi
|
|
71
71
|
=== A3 no-negation
|
|
72
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
72
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
73
73
|
AH HIT grade-paper-writing grade the writing craft in this draft and name the offendi
|
|
74
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
74
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
75
75
|
AH HIT grade-paper-writing run the blind multi-grader panel over the prose clarity an
|
|
76
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
76
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
77
77
|
AH HIT grade-paper-writing score the title, the abstract and the intro architecture o
|
|
78
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
78
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
79
79
|
AH HIT grade-paper-writing grade how well this reads, hedge stacking included, and gi
|
|
80
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
80
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
81
81
|
SH HIT grade-paper-writing the abstract reads like a wall of jargon
|
|
82
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
82
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
83
83
|
SH HIT grade-paper-writing the intro architecture is fine but the prose clarity is no
|
|
84
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
84
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
85
85
|
SH miss cold-read-diff there is hedge stacking everywhere and the point never lan
|
|
86
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
86
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
87
87
|
SH HIT grade-paper-writing the writing does not read well even though the draft is fi
|
|
88
88
|
=== A4 no-machinery
|
|
89
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
89
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
90
90
|
AH HIT grade-paper-writing grade the writing craft in this draft and name the offendi
|
|
91
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
91
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
92
92
|
AH HIT grade-paper-writing run the blind multi-grader panel over the prose clarity an
|
|
93
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
93
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
94
94
|
AH HIT grade-paper-writing score the title, the abstract and the intro architecture o
|
|
95
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
95
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
96
96
|
AH HIT grade-paper-writing grade how well this reads, hedge stacking included, and gi
|
|
97
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
97
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
98
98
|
SH HIT grade-paper-writing the abstract reads like a wall of jargon
|
|
99
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
99
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
100
100
|
SH HIT grade-paper-writing the intro architecture is fine but the prose clarity is no
|
|
101
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
101
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
102
102
|
SH HIT grade-paper-writing there is hedge stacking everywhere and the point never lan
|
|
103
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
103
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
104
104
|
SH HIT grade-paper-writing the writing does not read well even though the draft is fi
|
|
105
105
|
=== A5 renamed (description identical)
|
|
106
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
106
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
107
107
|
AH miss ?grade-paper-writing grade the writing craft in this draft and name the offendi
|
|
108
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
108
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
109
109
|
AH miss ?grade-paper-writing run the blind multi-grader panel over the prose clarity an
|
|
110
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
110
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
111
111
|
AH miss ?grade-paper-writing score the title, the abstract and the intro architecture o
|
|
112
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
112
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
113
113
|
AH miss ?grade-paper-writing grade how well this reads, hedge stacking included, and gi
|
|
114
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
114
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
115
115
|
SH miss ?grade-paper-writing the abstract reads like a wall of jargon
|
|
116
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
116
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
117
117
|
SH miss ?grade-paper-writing the intro architecture is fine but the prose clarity is no
|
|
118
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
118
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
119
119
|
SH miss tighten-paper there is hedge stacking everywhere and the point never lan
|
|
120
|
-
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["
|
|
120
|
+
Ignoring 24 permissions.allow entries from .claude/settings.json and .claude/settings.local.json: this workspace has not been trusted. Run Claude Code interactively here once and accept the trust dialog, or set projects["<project>"].hasTrustDialogAccepted: true in /root/.claude.json.
|
|
121
121
|
SH miss ?grade-paper-writing the writing does not read well even though the draft is fi
|
|
122
122
|
|
|
123
123
|
ORACLE ACCURACY BY ARM (closed book, 37 descriptions in context, 1 pick each)
|
|
@@ -2,8 +2,8 @@ patched: filter the MAIN LOOP to grade-paper-writing, cells AH+SH (guards untouc
|
|
|
2
2
|
patched: redirect the JSON output so the parent's own result file survives
|
|
3
3
|
|
|
4
4
|
running the PARENT'S code, grade-paper-writing AH+SH only, 3 trials = 24 runs
|
|
5
|
-
parent :
|
|
6
|
-
patched:
|
|
5
|
+
parent : <project>/.claude/skills/paper-pipeline/framing-vs-vocabulary.eval.mjs
|
|
6
|
+
patched: <scratch>/parent-repl/replication.mjs
|
|
7
7
|
|
|
8
8
|
|
|
9
9
|
37 skills installed → 36 competitors per run
|
|
@@ -38,13 +38,13 @@
|
|
|
38
38
|
|
|
39
39
|
import { readFileSync, writeFileSync, mkdirSync, symlinkSync, existsSync, rmSync } from "node:fs";
|
|
40
40
|
import { join } from "node:path";
|
|
41
|
+
import { tmpdir } from "node:os";
|
|
41
42
|
import { spawnSync } from "node:child_process";
|
|
42
43
|
|
|
43
|
-
const ROOT = process.env.CLAUDE_PROJECT_DIR ??
|
|
44
|
+
const ROOT = process.env.CLAUDE_PROJECT_DIR ?? process.cwd();
|
|
44
45
|
const PARENT = join(ROOT, ".claude", "skills", "paper-pipeline", "framing-vs-vocabulary.eval.mjs");
|
|
45
46
|
const SCRATCH =
|
|
46
|
-
process.env.ABLATION_SCRATCH ??
|
|
47
|
-
"/tmp/claude-0/-home-user-mine/8268acb1-66a2-55ac-858e-2b9e3c81f84e/scratchpad/parent-repl";
|
|
47
|
+
process.env.ABLATION_SCRATCH ?? join(tmpdir(), "parent-repl");
|
|
48
48
|
|
|
49
49
|
const src = readFileSync(PARENT, "utf-8");
|
|
50
50
|
|
package/skills/paper-pipeline/repro/2026-10-13-grade-paper-writing-ablation-preflight-after.log
ADDED
|
@@ -0,0 +1,81 @@
|
|
|
1
|
+
|
|
2
|
+
26 skills installed → 25 competitors per run
|
|
3
|
+
|
|
4
|
+
ARMS — description length, word-set delta vs A0, mean prompt overlap
|
|
5
|
+
arm label chars words − + ovAH ovSH
|
|
6
|
+
A0 baseline (verbatim) 1371 103 0 0 0.85 0.78
|
|
7
|
+
A1 short (sentences 1-2 only) 618 51 52 0 0.71 0.78
|
|
8
|
+
A2 no-subordination 1189 94 9 0 0.85 0.78
|
|
9
|
+
A3 no-negation 1141 92 11 0 0.85 0.78
|
|
10
|
+
A4 no-machinery 670 45 58 0 0.44 0.35
|
|
11
|
+
A5 renamed (description identical) 1371 103 0 0 0.85 0.78
|
|
12
|
+
A6 persona-only (A1's mirror) 298 28 75 0 0.23 0.42
|
|
13
|
+
|
|
14
|
+
⚠ ovAH/ovSH FALL AS TEXT IS DELETED — caveat 2 in the header, quantified. The deletion arms carry
|
|
15
|
+
a vocabulary handicap against the parent's dominant factor. A rising deletion arm rose DESPITE it;
|
|
16
|
+
a flat one may be a real gain cancelled by it, and this design cannot tell. A5 is the only arm
|
|
17
|
+
with zero handicap (identical word set).
|
|
18
|
+
|
|
19
|
+
WORD-SET DELTAS (stemmed content words removed from A0 — none may be added)
|
|
20
|
+
|
|
21
|
+
A0 baseline (verbatim)
|
|
22
|
+
removed (0): —
|
|
23
|
+
added (0): —
|
|
24
|
+
|
|
25
|
+
A1 short (sentences 1-2 only)
|
|
26
|
+
removed (52): run cold stall pas committ persona non academic engineer per section workshop reviewer skimm introduction conclusion inventory rubric number readability gate skill consume panel review adversarial harden default blind multi grader mandate claim preservation diff row content defect structural cut plan tighten first bloat strength reward study compose call generative counterpart render
|
|
27
|
+
added (0): —
|
|
28
|
+
|
|
29
|
+
A2 no-subordination
|
|
30
|
+
removed (9): inventory rubric number readability gate skill consume harden call
|
|
31
|
+
added (0): —
|
|
32
|
+
|
|
33
|
+
A3 no-negation
|
|
34
|
+
removed (11): content defect structural cut plan tighten first bloat strength reward study
|
|
35
|
+
added (0): —
|
|
36
|
+
|
|
37
|
+
A4 no-machinery
|
|
38
|
+
removed (58): title abstract intro architecture structure prose clarity discipline land point figure economy honesty without hedge stack register distance peyton jone mcenerney gopen swan exemplar trust carlini greshake cold stall pas committ persona non academic engineer per section workshop reviewer skimm introduction conclusion inventory rubric number readability gate skill consume default blind multi grader mandate claim preservation diff row
|
|
39
|
+
added (0): —
|
|
40
|
+
|
|
41
|
+
A5 renamed (description identical)
|
|
42
|
+
removed (0): —
|
|
43
|
+
added (0): —
|
|
44
|
+
|
|
45
|
+
A6 persona-only (A1's mirror)
|
|
46
|
+
removed (75): grade craft ten dimension title intro architecture structure prose clarity discipline land point figure economy honesty without hedge stack register distance target venue best written peyton jone mcenerney gopen swan exemplar trust carlini greshake accept nam offend sentence fix inventory rubric number readability gate skill consume panel review adversarial harden default blind multi grader mandate claim preservation diff row content defect structural cut plan tighten first bloat strength reward study compose call generative counterpart render
|
|
47
|
+
added (0): —
|
|
48
|
+
|
|
49
|
+
RESULTING DESCRIPTIONS
|
|
50
|
+
|
|
51
|
+
── A0 (grade-paper-writing, 1371 chars)
|
|
52
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion) whose stall inventory — not the rubric number — is the readability gate other skills consume (pc-panel-review, paper-adversarial-review, harden-paper). Defaults to a blind multi-grader panel; after fixes, mandates the claims-preservation diff (row `claims`). NOT a content/defect review (paper-adversarial-review / pc-panel-review), NOT a structural cut plan (tighten-paper — run that first on a bloated draft), NOT content strength against what the venue rewards (study-accepted-papers). Compose with harden-paper (which calls it), draft-paper (generative counterpart), render-paper.
|
|
53
|
+
|
|
54
|
+
── A1 (grade-paper-writing, 618 chars)
|
|
55
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each.
|
|
56
|
+
|
|
57
|
+
── A2 (grade-paper-writing, 1189 chars)
|
|
58
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion). Defaults to a blind multi-grader panel; after fixes, mandates the claims-preservation diff (row `claims`). NOT a content/defect review (paper-adversarial-review / pc-panel-review), NOT a structural cut plan (tighten-paper — run that first on a bloated draft), NOT content strength against what the venue rewards (study-accepted-papers). Compose with draft-paper (generative counterpart), render-paper.
|
|
59
|
+
|
|
60
|
+
── A3 (grade-paper-writing, 1141 chars)
|
|
61
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion) whose stall inventory — not the rubric number — is the readability gate other skills consume (pc-panel-review, paper-adversarial-review, harden-paper). Defaults to a blind multi-grader panel; after fixes, mandates the claims-preservation diff (row `claims`). Compose with harden-paper (which calls it), draft-paper (generative counterpart), render-paper.
|
|
62
|
+
|
|
63
|
+
── A4 (grade-paper-writing, 670 chars)
|
|
64
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 against how the best-written papers read AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. NOT a content/defect review (paper-adversarial-review / pc-panel-review), NOT a structural cut plan (tighten-paper — run that first on a bloated draft), NOT content strength against what the venue rewards (study-accepted-papers). Compose with harden-paper (which calls it), draft-paper (generative counterpart), render-paper.
|
|
65
|
+
|
|
66
|
+
── A5 (paper-readability-review, 1371 chars)
|
|
67
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion) whose stall inventory — not the rubric number — is the readability gate other skills consume (pc-panel-review, paper-adversarial-review, harden-paper). Defaults to a blind multi-grader panel; after fixes, mandates the claims-preservation diff (row `claims`). NOT a content/defect review (paper-adversarial-review / pc-panel-review), NOT a structural cut plan (tighten-paper — run that first on a bloated draft), NOT content strength against what the venue rewards (study-accepted-papers). Compose with harden-paper (which calls it), draft-paper (generative counterpart), render-paper.
|
|
68
|
+
|
|
69
|
+
── A6 (grade-paper-writing, 298 chars)
|
|
70
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" on a draft. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion).
|
|
71
|
+
|
|
72
|
+
building each arm's skills dir under <scratch>/gpw-ablation (no real file is written) …
|
|
73
|
+
A0 → <scratch>/gpw-ablation/arm-A0
|
|
74
|
+
A1 → <scratch>/gpw-ablation/arm-A1
|
|
75
|
+
A2 → <scratch>/gpw-ablation/arm-A2
|
|
76
|
+
A3 → <scratch>/gpw-ablation/arm-A3
|
|
77
|
+
A4 → <scratch>/gpw-ablation/arm-A4
|
|
78
|
+
A5 → <scratch>/gpw-ablation/arm-A5
|
|
79
|
+
A6 → <scratch>/gpw-ablation/arm-A6
|
|
80
|
+
|
|
81
|
+
ALL PREFLIGHT GUARDS PASSED. No tokens spent. Run --mode main next.
|
package/skills/paper-pipeline/repro/2026-10-13-grade-paper-writing-ablation-preflight-before.log
ADDED
|
@@ -0,0 +1,81 @@
|
|
|
1
|
+
|
|
2
|
+
26 skills installed → 25 competitors per run
|
|
3
|
+
|
|
4
|
+
ARMS — description length, word-set delta vs A0, mean prompt overlap
|
|
5
|
+
arm label chars words − + ovAH ovSH
|
|
6
|
+
A0 baseline (verbatim) 1371 103 0 0 0.85 0.78
|
|
7
|
+
A1 short (sentences 1-2 only) 618 51 52 0 0.71 0.78
|
|
8
|
+
A2 no-subordination 1189 94 9 0 0.85 0.78
|
|
9
|
+
A3 no-negation 1141 92 11 0 0.85 0.78
|
|
10
|
+
A4 no-machinery 670 45 58 0 0.44 0.35
|
|
11
|
+
A5 renamed (description identical) 1371 103 0 0 0.85 0.78
|
|
12
|
+
A6 persona-only (A1's mirror) 298 28 75 0 0.23 0.42
|
|
13
|
+
|
|
14
|
+
⚠ ovAH/ovSH FALL AS TEXT IS DELETED — caveat 2 in the header, quantified. The deletion arms carry
|
|
15
|
+
a vocabulary handicap against the parent's dominant factor. A rising deletion arm rose DESPITE it;
|
|
16
|
+
a flat one may be a real gain cancelled by it, and this design cannot tell. A5 is the only arm
|
|
17
|
+
with zero handicap (identical word set).
|
|
18
|
+
|
|
19
|
+
WORD-SET DELTAS (stemmed content words removed from A0 — none may be added)
|
|
20
|
+
|
|
21
|
+
A0 baseline (verbatim)
|
|
22
|
+
removed (0): —
|
|
23
|
+
added (0): —
|
|
24
|
+
|
|
25
|
+
A1 short (sentences 1-2 only)
|
|
26
|
+
removed (52): run cold stall pas committ persona non academic engineer per section workshop reviewer skimm introduction conclusion inventory rubric number readability gate skill consume panel review adversarial harden default blind multi grader mandate claim preservation diff row content defect structural cut plan tighten first bloat strength reward study compose call generative counterpart render
|
|
27
|
+
added (0): —
|
|
28
|
+
|
|
29
|
+
A2 no-subordination
|
|
30
|
+
removed (9): inventory rubric number readability gate skill consume harden call
|
|
31
|
+
added (0): —
|
|
32
|
+
|
|
33
|
+
A3 no-negation
|
|
34
|
+
removed (11): content defect structural cut plan tighten first bloat strength reward study
|
|
35
|
+
added (0): —
|
|
36
|
+
|
|
37
|
+
A4 no-machinery
|
|
38
|
+
removed (58): title abstract intro architecture structure prose clarity discipline land point figure economy honesty without hedge stack register distance peyton jone mcenerney gopen swan exemplar trust carlini greshake cold stall pas committ persona non academic engineer per section workshop reviewer skimm introduction conclusion inventory rubric number readability gate skill consume default blind multi grader mandate claim preservation diff row
|
|
39
|
+
added (0): —
|
|
40
|
+
|
|
41
|
+
A5 renamed (description identical)
|
|
42
|
+
removed (0): —
|
|
43
|
+
added (0): —
|
|
44
|
+
|
|
45
|
+
A6 persona-only (A1's mirror)
|
|
46
|
+
removed (75): grade craft ten dimension title intro architecture structure prose clarity discipline land point figure economy honesty without hedge stack register distance target venue best written peyton jone mcenerney gopen swan exemplar trust carlini greshake accept nam offend sentence fix inventory rubric number readability gate skill consume panel review adversarial harden default blind multi grader mandate claim preservation diff row content defect structural cut plan tighten first bloat strength reward study compose call generative counterpart render
|
|
47
|
+
added (0): —
|
|
48
|
+
|
|
49
|
+
RESULTING DESCRIPTIONS
|
|
50
|
+
|
|
51
|
+
── A0 (grade-paper-writing, 1371 chars)
|
|
52
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion) whose stall inventory — not the rubric number — is the readability gate other skills consume (pc-panel-review, paper-adversarial-review, harden-paper). Defaults to a blind multi-grader panel; after fixes, mandates the claims-preservation diff (row `claims`). NOT a content/defect review (paper-adversarial-review / pc-panel-review), NOT a structural cut plan (tighten-paper — run that first on a bloated draft), NOT content strength against what the venue rewards (study-accepted-papers). Compose with harden-paper (which calls it), draft-paper (generative counterpart), render-paper.
|
|
53
|
+
|
|
54
|
+
── A1 (grade-paper-writing, 618 chars)
|
|
55
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each.
|
|
56
|
+
|
|
57
|
+
── A2 (grade-paper-writing, 1189 chars)
|
|
58
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion). Defaults to a blind multi-grader panel; after fixes, mandates the claims-preservation diff (row `claims`). NOT a content/defect review (paper-adversarial-review / pc-panel-review), NOT a structural cut plan (tighten-paper — run that first on a bloated draft), NOT content strength against what the venue rewards (study-accepted-papers). Compose with draft-paper (generative counterpart), render-paper.
|
|
59
|
+
|
|
60
|
+
── A3 (grade-paper-writing, 1141 chars)
|
|
61
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion) whose stall inventory — not the rubric number — is the readability gate other skills consume (pc-panel-review, paper-adversarial-review, harden-paper). Defaults to a blind multi-grader panel; after fixes, mandates the claims-preservation diff (row `claims`). Compose with harden-paper (which calls it), draft-paper (generative counterpart), render-paper.
|
|
62
|
+
|
|
63
|
+
── A4 (grade-paper-writing, 670 chars)
|
|
64
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 against how the best-written papers read AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. NOT a content/defect review (paper-adversarial-review / pc-panel-review), NOT a structural cut plan (tighten-paper — run that first on a bloated draft), NOT content strength against what the venue rewards (study-accepted-papers). Compose with harden-paper (which calls it), draft-paper (generative counterpart), render-paper.
|
|
65
|
+
|
|
66
|
+
── A5 (paper-readability-review, 1371 chars)
|
|
67
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" / "grade the writing" on a draft. Grades WRITING CRAFT only — ten dimensions 1–5 (Title, Abstract, Intro architecture, Structure, Prose clarity, Jargon discipline, Landing-the-point, Figure economy, Honesty-without-hedge-stacking, Register distance to the target venue) against how the best-written papers read (Peyton Jones, McEnerney, Gopen & Swan; exemplars Trusting Trust, Carlini, Greshake) AND against 2–3 accepted papers of the target venue, naming the offending sentence and the fix for each. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion) whose stall inventory — not the rubric number — is the readability gate other skills consume (pc-panel-review, paper-adversarial-review, harden-paper). Defaults to a blind multi-grader panel; after fixes, mandates the claims-preservation diff (row `claims`). NOT a content/defect review (paper-adversarial-review / pc-panel-review), NOT a structural cut plan (tighten-paper — run that first on a bloated draft), NOT content strength against what the venue rewards (study-accepted-papers). Compose with harden-paper (which calls it), draft-paper (generative counterpart), render-paper.
|
|
68
|
+
|
|
69
|
+
── A6 (grade-paper-writing, 298 chars)
|
|
70
|
+
Use when asking "does this paper read well?" / "is the writing any good?" / "it reads like a wall of text and jargon" on a draft. Runs the cold-read stall pass with TWO committed personas (a non-academic engineer per section, and a workshop reviewer skimming abstract, introduction and conclusion).
|
|
71
|
+
|
|
72
|
+
building each arm's skills dir under <scratch>/gpw-ablation (no real file is written) …
|
|
73
|
+
A0 → <scratch>/gpw-ablation/arm-A0
|
|
74
|
+
A1 → <scratch>/gpw-ablation/arm-A1
|
|
75
|
+
A2 → <scratch>/gpw-ablation/arm-A2
|
|
76
|
+
A3 → <scratch>/gpw-ablation/arm-A3
|
|
77
|
+
A4 → <scratch>/gpw-ablation/arm-A4
|
|
78
|
+
A5 → <scratch>/gpw-ablation/arm-A5
|
|
79
|
+
A6 → <scratch>/gpw-ablation/arm-A6
|
|
80
|
+
|
|
81
|
+
ALL PREFLIGHT GUARDS PASSED. No tokens spent. Run --mode main next.
|
|
@@ -0,0 +1,59 @@
|
|
|
1
|
+
|
|
2
|
+
26 skills installed → 25 competitors per run
|
|
3
|
+
|
|
4
|
+
ARMS — description length, word-set delta vs A0, mean prompt overlap
|
|
5
|
+
arm label chars words − + ovAH ovSH
|
|
6
|
+
A0 baseline (verbatim) 1371 103 0 0 0.85 0.78
|
|
7
|
+
A1 short (sentences 1-2 only) 618 51 52 0 0.71 0.78
|
|
8
|
+
A2 no-subordination 1189 94 9 0 0.85 0.78
|
|
9
|
+
A3 no-negation 1141 92 11 0 0.85 0.78
|
|
10
|
+
A4 no-machinery 670 45 58 0 0.44 0.35
|
|
11
|
+
A5 renamed (description identical) 1371 103 0 0 0.85 0.78
|
|
12
|
+
A6 persona-only (A1's mirror) 298 28 75 0 0.23 0.42
|
|
13
|
+
|
|
14
|
+
⚠ ovAH/ovSH FALL AS TEXT IS DELETED — caveat 2 in the header, quantified. The deletion arms carry
|
|
15
|
+
a vocabulary handicap against the parent's dominant factor. A rising deletion arm rose DESPITE it;
|
|
16
|
+
a flat one may be a real gain cancelled by it, and this design cannot tell. A5 is the only arm
|
|
17
|
+
with zero handicap (identical word set).
|
|
18
|
+
|
|
19
|
+
HARNESS OPTIONS — this file vs the parent (extracted from both sources)
|
|
20
|
+
option ablation parent
|
|
21
|
+
allowedTools ["Skill" ["Skill"
|
|
22
|
+
trials TRIALS TRIALS
|
|
23
|
+
concurrency 3 3
|
|
24
|
+
spacingSec 2 2
|
|
25
|
+
timeoutMs 180000 180000
|
|
26
|
+
minPrompts 4 4
|
|
27
|
+
fixture /.exec(src)?.[0] ?? "" FIXTURE
|
|
28
|
+
skillsDir dir SKILLS_DIR
|
|
29
|
+
|
|
30
|
+
fixture sha256/16 ablation 89d1b720c4cedde6
|
|
31
|
+
FIXTURE source block identical to the parent's: YES
|
|
32
|
+
fired predicate identical (modulo the id it matches): YES
|
|
33
|
+
firedSkills() helper identical: YES
|
|
34
|
+
|
|
35
|
+
ARM DIR INTEGRITY — sha256/16 of each SKILL.md frontmatter vs the real .claude/skills
|
|
36
|
+
A0 26 skills · unintended frontmatter diffs: none
|
|
37
|
+
A1 26 skills · unintended frontmatter diffs: none
|
|
38
|
+
A2 26 skills · unintended frontmatter diffs: none
|
|
39
|
+
A3 26 skills · unintended frontmatter diffs: none
|
|
40
|
+
A4 26 skills · unintended frontmatter diffs: none
|
|
41
|
+
A5 26 skills · unintended frontmatter diffs: none
|
|
42
|
+
A6 26 skills · unintended frontmatter diffs: none
|
|
43
|
+
|
|
44
|
+
ROW BY ROW — parent (8 prompts) vs this file's A0 (8 prompts)
|
|
45
|
+
cell parent A0 Δ prompt
|
|
46
|
+
AH 0/3 1/3 +1 grade the writing craft in this draft and name the off
|
|
47
|
+
AH 3/3 3/3 0 run the blind multi-grader panel over the prose clarit
|
|
48
|
+
AH 0/3 3/3 +3 score the title, the abstract and the intro architectu
|
|
49
|
+
AH 0/3 0/3 0 grade how well this reads, hedge stacking included, an
|
|
50
|
+
SH 0/3 0/3 0 the abstract reads like a wall of jargon
|
|
51
|
+
SH 0/3 2/3 +2 the intro architecture is fine but the prose clarity i
|
|
52
|
+
SH 0/3 2/3 +2 there is hedge stacking everywhere and the point never
|
|
53
|
+
SH 2/3 2/3 0 the writing does not read well even though the draft i
|
|
54
|
+
5/24 13/24 +8 TOTAL
|
|
55
|
+
|
|
56
|
+
3 of 8 prompts moved by ≥2 of 3 trials. Spread across many prompts ⇒ a global shift
|
|
57
|
+
(session, load, model routing). Concentrated in one or two ⇒ something prompt-specific.
|
|
58
|
+
|
|
59
|
+
SETUPDIFF COMPLETE. No tokens spent.
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"skill":"grade-paper-writing","model":"sonnet","trials":1,"rate":0.875,"falsePositiveRate":0,"precision":1,"competitors":25,"perPrompt":[{"prompt":"оцени как написано — читается как стена жаргона","rate":1},{"prompt":"is the writing any good or does it read like shit","rate":1},{"prompt":"abstract звучит криво хотя по смыслу всё на месте, дай оценку прозе","rate":1},{"prompt":"grade the craft: title, abstract, sentence clarity, hedge stacking","rate":1},{"prompt":"прочитай черновик и скажи честно, нормально ли он читается","rate":0},{"prompt":"the reviewer says our paper reads like a blog post — is he right about the prose","rate":1},{"prompt":"каждое предложение по отдельности понятно, а абзацы всё равно сливаются в сплошную простыню","rate":1},{"prompt":"grade the prose in section 3 and give me the sentences to rewrite","rate":1}]}
|
|
@@ -0,0 +1,22 @@
|
|
|
1
|
+
Spent: 1,807,963 tokens (88 in · 12.6k out · 1562.4k cache) · ~$1.30 API-equivalent
|
|
2
|
+
Billed to: your Claude subscription — $0 metered ✅
|
|
3
|
+
|
|
4
|
+
grade-paper-writing — prose craft — the sentence is the unit, NOT the section [model: sonnet]
|
|
5
|
+
trigger-rate: 88% (8 runs)
|
|
6
|
+
1.00 оцени как написано — читается как стена жаргона
|
|
7
|
+
1.00 is the writing any good or does it read like shit
|
|
8
|
+
1.00 abstract звучит криво хотя по смыслу всё на месте, дай оценк
|
|
9
|
+
1.00 grade the craft: title, abstract, sentence clarity, hedge st
|
|
10
|
+
0.00 прочитай черновик и скажи честно, нормально ли он читается
|
|
11
|
+
1.00 the reviewer says our paper reads like a blog post — is he r
|
|
12
|
+
1.00 каждое предложение по отдельности понятно, а абзацы всё равн
|
|
13
|
+
1.00 grade the prose in section 3 and give me the sentences to re
|
|
14
|
+
false-positive: 0% precision: 100%
|
|
15
|
+
0.00 [irrelevant] статья раздулась, середина провисает — что резат
|
|
16
|
+
0.00 [irrelevant] does the paper actually carry a reader to one co
|
|
17
|
+
0.00 [irrelevant] поправил формулировки в 4.2, прогони свежим чита
|
|
18
|
+
0.00 [irrelevant] what is our desk reject risk and where do we ove
|
|
19
|
+
0.00 [irrelevant] какая вероятность принятия у этой статьи, если ч
|
|
20
|
+
0.00 [irrelevant] статья готова к сабмиту или нет
|
|
21
|
+
0.00 [irrelevant] посмотри что принимали на этой площадке и чем те
|
|
22
|
+
whole-harness: measured against 25 competing skill(s)
|
|
@@ -25,6 +25,10 @@ the session.
|
|
|
25
25
|
| `2026-08-08-parent-replication.mjs` | the replication driver | Runs `framing-vs-vocabulary.eval.mjs` **itself** with two asserted patches (main loop → this skill + AH/SH; output path), everything else untouched, against the real `.claude/skills`. Written because the ablation's control contradicted the parent. |
|
|
26
26
|
| `2026-08-08-parent-replication-gpw.log` | that driver's run | 🔴 **The parent's own code returns AH 9/12 + SH 3/12 = 12/24 (50%)** hours after the same code returned 3/12 + 2/12 = 5/24 (21%). |
|
|
27
27
|
| `2026-08-08-parent-replication-gpw.json` | per-run rows for the above | The parent's own row schema, so it diffs directly against `2026-08-08-framing-vs-vocabulary.json`. |
|
|
28
|
+
| `2026-10-13-grade-paper-writing-trigger.log` | `grade-paper-writing/grade-paper-writing.eval.mjs`, the rewritten 499-char description (ledger `description_sha` dc41f0994474b368) | 1 trial, sonnet, 25 competing skills. Recall **7/8 (88%)**: the four measured prompts 4/4, the four new ones 3/4 (missed: «прочитай черновик и скажи честно, нормально ли он читается»). False positives **0/7**, one prompt from each collider. One trial is a smoke run, not a rate. |
|
|
29
|
+
| `2026-10-13-grade-paper-writing-trigger.jsonl` | per-prompt rows for the above | The kit's `VIGILES_RESULTS` row: rate per should-fire prompt. |
|
|
30
|
+
| `2026-10-13-grade-paper-writing-ablation-preflight-before.log`, `…-after.log` | `grade-paper-writing-ablation.eval.mjs --mode preflight`, **free**, before and after its A0 was pinned | Byte-identical: every arm builds from the same text after the live description was rewritten. Scratch paths replaced by `<scratch>`. |
|
|
31
|
+
| `2026-10-13-grade-paper-writing-ablation-setupdiff.log` | the same file's `--mode setupdiff`, **free**, after the pin | Passes; the parent-vs-A0 row-by-row is the recorded 2026-08-08 data, unchanged. |
|
|
28
32
|
|
|
29
33
|
## How to read these, and the trap in them
|
|
30
34
|
|
|
@@ -400,7 +400,7 @@ process.on("exit", () => rmSync(TMP, { recursive: true, force: true }));
|
|
|
400
400
|
{
|
|
401
401
|
const { spawnSync } = await import("node:child_process");
|
|
402
402
|
const dir = join(TMP, "e2e");
|
|
403
|
-
mkdirSync(join(dir, "repro"), { recursive: true });
|
|
403
|
+
mkdirSync(join(dir, "repro", "data"), { recursive: true });
|
|
404
404
|
const paper =
|
|
405
405
|
"@misc{lovelace2025,\n author={Lovelace, Ada and Hopper, Grace},\n title={A Study},\n" +
|
|
406
406
|
" note={arXiv:2503.09002},\n year={2025}}\n";
|
|
@@ -411,7 +411,7 @@ process.on("exit", () => rmSync(TMP, { recursive: true, force: true }));
|
|
|
411
411
|
);
|
|
412
412
|
built(dir);
|
|
413
413
|
writeFileSync(
|
|
414
|
-
join(dir, "repro", "refs-cache.json"),
|
|
414
|
+
join(dir, "repro", "data", "refs-cache.json"),
|
|
415
415
|
JSON.stringify({
|
|
416
416
|
"arxiv:2503.09002": {
|
|
417
417
|
httpStatus: 200,
|
|
@@ -46,7 +46,7 @@
|
|
|
46
46
|
* only one that returns a STRUCTURED name; with any other the surname would have to be cut out by
|
|
47
47
|
* hand, that is, we would be writing ourselves the very half that was broken.
|
|
48
48
|
*
|
|
49
|
-
* 🔴 THE CACHE IS EVIDENCE, NOT A SECOND DECLARATION. `<paper>/repro/refs-cache.json` keeps the RAW
|
|
49
|
+
* 🔴 THE CACHE IS EVIDENCE, NOT A SECOND DECLARATION. `<paper>/repro/data/refs-cache.json` keeps the RAW
|
|
50
50
|
* BODIES of the registries' responses and is committed (for `<paper-a>` it has been in git since
|
|
51
51
|
* 16.08). It cannot be made up — only refreshed from the registry. The facts in `_build/` are
|
|
52
52
|
* derived from it and from the bibliography, which is why `_build/` is in `.gitignore` while the
|
|
@@ -450,7 +450,7 @@ async function main(argv) {
|
|
|
450
450
|
const { paperDir, texts } = found;
|
|
451
451
|
const cachePath = opt("cache")
|
|
452
452
|
? resolve(opt("cache"))
|
|
453
|
-
: join(paperDir, "repro", "refs-cache.json");
|
|
453
|
+
: join(paperDir, "repro", "data", "refs-cache.json");
|
|
454
454
|
const out = opt("out")
|
|
455
455
|
? resolve(opt("out"))
|
|
456
456
|
: join(paperDir, "_build", "refs.facts.json");
|
|
@@ -236,7 +236,9 @@ test("online: hits are fetched and cached, a transport failure is named and stay
|
|
|
236
236
|
stdout: r.stdout,
|
|
237
237
|
stderr: r.stderr,
|
|
238
238
|
cached: Object.keys(
|
|
239
|
-
JSON.parse(
|
|
239
|
+
JSON.parse(
|
|
240
|
+
readFileSync(join(root, "paper/repro/data/refs-cache.json")),
|
|
241
|
+
),
|
|
240
242
|
),
|
|
241
243
|
},
|
|
242
244
|
{
|
|
@@ -265,7 +267,7 @@ test("the paper file itself as the target, with --cache, --out and --offline", (
|
|
|
265
267
|
const r = cli([
|
|
266
268
|
join(root, "paper/refs.bib"),
|
|
267
269
|
"--offline",
|
|
268
|
-
`--cache=${join(root, "paper/repro/refs-cache.json")}`,
|
|
270
|
+
`--cache=${join(root, "paper/repro/data/refs-cache.json")}`,
|
|
269
271
|
`--out=${join(root, "out/facts.json")}`,
|
|
270
272
|
]);
|
|
271
273
|
assert.deepEqual(r, {
|
|
@@ -79,8 +79,13 @@ function fixture({ scripts = {}, allow = null, bundle = {}, at = "repro" }) {
|
|
|
79
79
|
mkdirSync(join(d, at, "artifact-anon", dirname(p)), { recursive: true });
|
|
80
80
|
writeFileSync(join(d, at, "artifact-anon", p), src);
|
|
81
81
|
}
|
|
82
|
-
if (allow !== null)
|
|
83
|
-
|
|
82
|
+
if (allow !== null) {
|
|
83
|
+
mkdirSync(join(d, at, "tools"), { recursive: true });
|
|
84
|
+
writeFileSync(
|
|
85
|
+
join(d, at, "tools", "generated-code-grandfathered.txt"),
|
|
86
|
+
allow,
|
|
87
|
+
);
|
|
88
|
+
}
|
|
84
89
|
return d;
|
|
85
90
|
}
|
|
86
91
|
|