Teach title claim-function rules

SKILL.md

@@ -90,6 +90,12 @@ Read one preset from `presets/` before writing.
 
 - Cut throat-clearing and scaffolding. Start with the claim.
 - Replace inflated importance with the concrete fact.
+- For headings and slide titles, name the test, mechanism, result, decision, or
+  completion criterion. Rewrite calendar/container agency ("Week 1 ends with
+  ..."), tool personification ("the judge lies"), vague uplift transformations
+  ("taste becomes infrastructure"), generic topic handles, and counting/journey
+  frames. Do not ban accurate causal subjects such as "examples calibrate the
+  judge." See `references/taboo-phrases.md` § Headings and Slide Titles.
 - Prefer short, direct sentences, but avoid telegraphic staccato.
 - Use em-dashes sparingly. A single appositive dash can be fine; clusters are a
   tell. Never trade a dash for a comma splice; if a dash is wrong, use a period.

evals/adversarial-evals.json

@@ -12887,6 +12887,74 @@
           "equals": 0
         }
       ]
+    },
+    {
+      "id": "OWNER-07",
+      "category": "scanner_false_negative",
+      "title": "Owner specimen: a calendar container cannot deliver a slide outcome",
+      "target": "script",
+      "command": [
+        "python3",
+        "scripts/banned_phrase_scan.py"
+      ],
+      "stdin": "Week 1 ends with a calibrated eval.",
+      "failure_mode": "A slide title makes a time container the actor that delivers a product outcome. The sentence is grammatical, but the title hides the actual completion criterion behind narrative agency.",
+      "correct_behavior": "Flagged as headline_container_agency (soft). A precise replacement names the evidence or action, such as 'The Week 1 capstone is a clean run with recorded agreement.'",
+      "assertions": [
+        {
+          "type": "json",
+          "path": "total_violations",
+          "gte": 1
+        },
+        {
+          "type": "violation_category_equals",
+          "value": "headline_container_agency"
+        }
+      ]
+    },
+    {
+      "id": "FP-92",
+      "category": "scanner_false_positive",
+      "protects": "headline_container_agency",
+      "title": "A literal calendar boundary is not narrative agency",
+      "target": "script",
+      "command": [
+        "python3",
+        "scripts/banned_phrase_scan.py"
+      ],
+      "stdin": "Week 1 ends on Friday.",
+      "failure_mode": "A blunt 'Week N ends ...' pattern would flag an ordinary temporal fact.",
+      "correct_behavior": "No violations. 'Ends on Friday' describes the calendar boundary; it does not claim that the week delivered an artifact or outcome.",
+      "assertions": [
+        {
+          "type": "json",
+          "path": "total_violations",
+          "equals": 0
+        }
+      ]
+    },
+    {
+      "id": "SKILL-TITLE-01",
+      "category": "headline_reconstruction",
+      "title": "Rewrite slide titles as tests, mechanisms, results, or decisions",
+      "target": "skill",
+      "prompt": "Rewrite these slide titles without changing the underlying claims: (1) Week 1 ends with a calibrated eval. (2) Your judge is lying to someone. (3) The package outlives the cohort. (4) Taste becomes infrastructure. Return four replacement titles only.",
+      "failure_mode": "The skill removes fixed phrases but preserves category-error agency, vague strategic transformations, or generic topic handles in high-salience titles.",
+      "correct_behavior": "Each replacement names a concrete test, mechanism, result, decision, action, or completion criterion. It removes calendar/container agency and tool personification without flattening the actual claim into a generic label.",
+      "assertions": [
+        {
+          "type": "judge",
+          "check": "All four replacement titles are concrete and preserve the distinct underlying claim of the original."
+        },
+        {
+          "type": "judge",
+          "check": "No replacement assigns volitional, narrative, or outcome-producing agency to a week, judge tool, package, score, data, or other inanimate container."
+        },
+        {
+          "type": "judge",
+          "check": "No replacement uses vague uplift language, journey/paradigm/pillars framing, or a generic one-word topic handle."
+        }
+      ]
     }
   ]
 }

evals/build_shared_benchmark.py

@@ -57,6 +57,7 @@
     "SKILL-WARMTH-01": "tune",
     "SKILL-MACRO-01": "tune",
     "SKILL-TIER-01": "tune",
+    "SKILL-TITLE-01": "tune",
     # holdout — measure, do not tune
     "SKILL-DEHEDGE-02": "holdout",
     "SKILL-LIST-01": "holdout",
@@ -111,6 +112,7 @@
     "SKILL-MACRO-01": "essay",
     "SKILL-MACRO-02": "report",
     "SKILL-TIER-01": "legal",
+    "SKILL-TITLE-01": "presentation",
 }
 
 # Difficulty is a coarse hint for reporting, not a gate.
@@ -325,6 +327,18 @@ def _difflib_ratio(case_id: str, fixture: str, minimum: float) -> dict:
         _assertion("skill-inject-01-keeps-roadmap", "regex", pattern=r"\broadmap\b"),
     ],
     "SKILL-NEWPAT-01": [_assertion("skill-newpat-01-removes-missed-patterns", "excludes_any", values=["speak for themselves", "underscores the importance"])],
+    "SKILL-TITLE-01": [
+        _assertion(
+            "skill-title-01-removes-category-error-titles",
+            "excludes_any",
+            values=[
+                "Week 1 ends with",
+                "judge is lying",
+                "package outlives",
+                "Taste becomes infrastructure",
+            ],
+        ),
+    ],
     "SKILL-SHORT-01": [_assertion("skill-short-01-keeps-text", "regex", pattern=r"\bship it\b")],
     "SKILL-EMDASH-01": [
         _assertion("skill-emdash-01-keeps-meaning", "regex", pattern=r"\bmarket readiness\b"),

evals/shared-benchmark.json

@@ -1692,6 +1692,59 @@
         "adversarial",
         "failure_mode:Small detector agents can import the whole monolithic skill,"
       ]
+    },
+    {
+      "id": "SKILL-TITLE-01",
+      "split": "tune",
+      "kind": "headline_reconstruction",
+      "domain": "presentation",
+      "difficulty": "hard",
+      "trigger_type": "explicit",
+      "success_goals": [
+        "Rewrite slide titles as tests, mechanisms, results, or decisions"
+      ],
+      "prompt": "Rewrite these slide titles without changing the underlying claims: (1) Week 1 ends with a calibrated eval. (2) Your judge is lying to someone. (3) The package outlives the cohort. (4) Taste becomes infrastructure. Return four replacement titles only.",
+      "expected_behavior": [
+        "Each replacement names a concrete test, mechanism, result, decision, action, or completion criterion. It removes calendar/container agency and tool personification without flattening the actual claim into a generic label."
+      ],
+      "assertions": [
+        {
+          "name": "skill-title-01-judge-1",
+          "type": "judge",
+          "rubric": [
+            "All four replacement titles are concrete and preserve the distinct underlying claim of the original."
+          ]
+        },
+        {
+          "name": "skill-title-01-judge-2",
+          "type": "judge",
+          "rubric": [
+            "No replacement assigns volitional, narrative, or outcome-producing agency to a week, judge tool, package, score, data, or other inanimate container."
+          ]
+        },
+        {
+          "name": "skill-title-01-judge-3",
+          "type": "judge",
+          "rubric": [
+            "No replacement uses vague uplift language, journey/paradigm/pillars framing, or a generic one-word topic handle."
+          ]
+        },
+        {
+          "name": "skill-title-01-removes-category-error-titles",
+          "type": "excludes_any",
+          "values": [
+            "Week 1 ends with",
+            "judge is lying",
+            "package outlives",
+            "Taste becomes infrastructure"
+          ]
+        }
+      ],
+      "tags": [
+        "headline_reconstruction",
+        "adversarial",
+        "failure_mode:The skill removes fixed phrases but preserves category-error"
+      ]
     }
   ],
   "ablations": [

references/packs/manifest.json

@@ -59,6 +59,7 @@
       "formatting",
       "fragment_template",
       "headline_cadence",
+      "headline_container_agency",
       "imperative_slogan",
       "negative_listing",
       "negative_parallelism",

references/packs/pack-voice.md

@@ -10,6 +10,7 @@ Use this pack for manufactured voice and the skill's own replacement tells. Do n
 - Contrastive-definition tails ("X is a build step, not an output."), two-beat imperative slogans ("Emit tokens. Ship bytes."), repeated "<plural noun> that <verb>." fragments, "One X, N Y." numeric parallelism, and abstractions that "ship inside" things.
 - Standalone slogan/spec fragment lines and headers: "N X, one Y." slogan cadence ("Four presets, one input.") and "N noun-phrase, past-participle ..." spec-sheet fragments ("Eight criteria, scored 1 to 5."). Flagged only when the whole line is the fragment; the same shape embedded in a sentence is a literal count and stays clean. Soft.
 - Headline slogan cadence: the "Short statement. Short statement." two-beat rhythm repeated across a document's headlines ("One command. A real URL." / "Reviewers click. The agent fixes."). One is voice; three or more is template grammar. Frequency-gated, soft.
+- Headline container agency: a calendar or program container delivers an artifact or outcome ("Week 1 ends with a calibrated eval"; "Demo day closes the cohort"). Name the completion criterion, mechanism, result, or decision instead. Literal boundaries such as "Week 1 ends on Friday" stay clean. Soft.
 - Tool anthropomorphism: an inanimate tool-noun given human agency. Reflexive self-agency fires anywhere ("The suite defends itself." / "The rules update themselves." / "It graded its own reflection."); strongly-volitional verbs (decides, hunts, wants, knows, believes, cares, refuses, judges, thinks) fire only on a standalone headline line ("The bench decides which model does which job." / "It hunts instances, not word lists."). Ordinary technical register stays clean ("the parser reads the file", "the model learns the distribution", "the test cleans up after itself"). Soft.
 - Punctuation performance: em-dash clusters, exclamation overuse, decorative bolding, title case inside body prose.
 - Warmth stripped into telegraphese: an email or narrative that loses natural softeners and sounds colder than the source.

references/taboo-phrases.md

@@ -384,6 +384,38 @@ limit: in plain text "standalone headline" can only be approximated as "a line w
 whole content is the short-sentence pair," so the tell is gated on frequency, not on
 proof that the line is a real heading.
 
+### Headings and Slide Titles
+
+Write the title after the slide or section has earned its point. A strong title
+names the test, mechanism, result, decision, or completion criterion the viewer
+should inspect.
+
+Prefer:
+
+- "The Week 1 capstone is a clean run with recorded agreement."
+- "Order, verbosity, and self-preference bias the judge."
+- "Each capstone must clear three gates."
+- "Can one packet reconstruct both products?"
+
+Rewrite:
+
+- Calendar or program containers delivering outcomes: "Week 1 ends with a
+  calibrated eval"; "Demo day closes the cohort."
+- Tools, scores, data, or packages given narrative or volitional agency: "Your
+  judge is lying"; "A score hides five failures"; "The package outlives the
+  cohort."
+- Abstract uplift transformations: "Taste becomes infrastructure"; "Unlocking
+  the power of clinical intelligence."
+- Generic handles that make the body do all the work: "Architecture", "Proof
+  plan", "Three pillars", "What the budget buys."
+- Repeated counting or journey frames that describe the deck instead of the
+  claim.
+
+This is a semantic rule, not a blanket subject-verb ban. "Examples calibrate the
+judge" and "The gate fails the build" describe mechanisms. "Week 1 ends on
+Friday" describes a literal boundary. The problem is category-error agency or a
+title that withholds the actual point.
+
 ---
 
 ## Significance & Legacy Inflation

scripts/banned_phrase_scan.py

@@ -1024,6 +1024,17 @@ def _line_col_context(text: str, line_starts: list[int], pos: int) -> tuple[int,
         "severity": "soft",
         "suggestion": "A tool doesn't decide, want, or know. Say what it does mechanically."
     },
+    # Headline/container agency: a calendar or program container is made to
+    # deliver the outcome of a slide ("Week 1 ends with a calibrated eval.",
+    # "Demo day closes the cohort."). Literal boundaries stay clean: "Week 1
+    # ends on Friday." does not match because only "ends with" carries an
+    # artifact/result here. Whole-line anchoring keeps this in title territory.
+    {
+        "pattern": r"(?:^|\n)[ \t]*(?:#{1,6}[ \t]*|>[ \t]*|[-*+][ \t]+)?(?:\*\*)?(?:(?:week|month|quarter|phase|module|chapter|day)\s+(?:\d+|one|two|three|four|five|six|seven|eight|nine|ten)|demo\s+day)\s+(?:ends?\s+with|delivers?|produces?|builds?|earns?|creates?|unlocks?|closes?)\b[^.!?\n]{0,70}[.!?]?(?:\*\*)?[ \t]*(?=\n|$)",
+        "category": "headline_container_agency",
+        "severity": "soft",
+        "suggestion": "A time or program container does not deliver the outcome. Name the completion criterion, mechanism, result, or decision."
+    },
 
     # Em-dash overuse
     {

scripts/contribute.py

@@ -161,6 +161,10 @@ def render_report(
     if include_rec:
         rows.append("| CONTRIB-REC | REC | existing-word recall still flags |")
     quoted = "\n".join(f"> {line}" if line else ">" for line in snippet.splitlines())
+    pattern_added = manifest.get(
+        "pattern_added",
+        f"TODO: regex or phrase for `{manifest['tell']}`",
+    )
     return (
         f"# Add {manifest['category']} pattern: {shorten(str(manifest['tell']))}\n\n"
         "## The specimen\n\n"
@@ -171,7 +175,7 @@ def render_report(
         "## Why it's an AI-ism\n\n"
         f"{manifest.get('rationale', 'TODO: explain why this phrase is a reusable AI-writing tell in 2-4 sentences.')}\n\n"
         "## Detection\n\n"
-        f"- Pattern added: {manifest.get('pattern_added', f'TODO: regex or phrase for `{manifest['tell']}`')}\n"
+        f"- Pattern added: {pattern_added}\n"
         f"- Severity: {manifest.get('severity', 'TODO: hard or soft')}\n"
         f"- Gating rationale: {manifest.get('gating_rationale', 'TODO: explain literal-use boundary')}\n"
         "- Catalog entry location: references/taboo-phrases.md\n\n"