Upstream blader/humanizer PR blader#96 (erhanurgun). Replaces pattern 14's
"use sparingly" guideline with a hard ban plus replacement hierarchy
and a mandatory scan step. Aligns with writing-style.md rule.

Preserved local paired-bracketing content from PR blader#84.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Closes E2E meaning dimension breach (5/5 cases below 8.0 threshold at v3.4.0
baseline; 6/6 cases pass per-case 8.0+ on all 3 dims after iteration).

E2E aggregate: human_ness 8.667 / meaning 8.389 / length 8.667 across 6 cases
x 3 runs (sonnet skill + sonnet judge). Per-case:
  academic   8.00 / 8.33 / 9.00
  career     9.00 / 8.00 / 8.00   [new domain, first-shot pass]
  casual     9.00 / 8.00 / 8.67
  legal      8.67 / 9.00 / 8.33
  marketing  9.00 / 8.00 / 9.00
  technical  8.33 / 9.00 / 9.00

Meaning baseline 7.733 to 8.389 (+0.656).

Changes (SKILL.md + patterns/en.md + domains/en_overrides.md):

- SKILL.md step 9: "Rewrite, don't delete" paragraph-count rule (ported from
  blader/humanizer v2.6.0 PR blader#84). Cover everything the original covers.
- SKILL.md step 10: elevated to hard-constraint claim inventory. List claims
  in source, verify each survives in output, restore any missing.
- SKILL.md final audit (step 11): two new universal checks:
    + Fabrication check (no claims/details/specs introduced not in source)
    + New authorial positions check (no added hedges/skepticism/endorsements)
- domains/en_overrides.md technical: preserve functional standing claims
  ("industry best practices", "battle-tested", "production-grade") + never
  introduce implementation specifics not in source (no inventing "exponential
  backoff" when source said "backoff strategy").
- domains/en_overrides.md marketing: buzzword-in-phrase rule + preserve
  positioning angles + 5-point preserve-everything checklist (feature claims,
  tonal attributes, brand-tier descriptors, emotional-fit phrases,
  aspirational tone) + brand-tier audit step.
- domains/en_overrides.md casual: concept-noun preservation rule with named
  anti-examples (creativity at scale, team alignment, team agility,
  collaboration, software-development-evolution, integration, user impact).
- patterns/en.md PERSONALITY AND SOUL:
    + "Do not invent the author's opinions" rule
    + Domain note strengthened with upstream wording ("neutral and plain IS
      the correct human voice for encyclopedic/technical/legal/reference")

New `career` domain (cover letter / CV / LinkedIn / Anschreiben):
- SKILL.md domain table row added
- domains/en_overrides.md override matrix column + 3-paragraph guidance:
  career-specific AI tells (results-driven, uniquely positioned, passionate
  about, leverage-as-verb, etc.), preserve rules (metrics sacred + proper
  nouns + tech stack + JD keywords + concrete achievements verbatim), and
  cultural-register note (US/UK confident; DE/AT/CH factual-modest deferred
  to Phase 2 DE pack)
- Career Writer persona at vault path
  __obsidian_vault/AI/AGENT PERSONAS/Agents/writer-career-writer-agent.md

New eval corpus case `evals/corpus/en/e2e/ai_career_01.json`:
- Prototypical AI-generated cover letter with classic tells
- Sacred preserve targets: 18-month migration, 40% p99 latency improvement,
  3x scale, Kubernetes + Go + PostgreSQL stack, Senior Software Engineer
  role at TechCorp
- First-shot E2E baseline = hn 9.00 / mean 8.00 / len 8.00

Tests: 78/78 pytest pass. Pattern eval rate 0.619 unchanged.
FP edit ratio 0.2039 + density preflight 1.00 unchanged.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>