{
  "_comment": "Round-2 durable anchors: proposed from model knowledge 2026-07-15 under the 'relevant and certain' bar; admitted to the ledger ONLY if results_da.json confirms them against the primary source. Grade A by construction (law/standard/theorem).",
  "anchors": [
    {
      "id": "DA-01",
      "claim": "NYC Local Law 144 of 2021, enforced from July 5, 2023, prohibits employers and employment agencies from using an automated employment decision tool unless it has undergone an independent bias audit within one year of use, with a summary of results publicly posted, and requires notice to candidates and employees.",
      "evidence": "Provenance: proposed from model knowledge during the 2026-07-15 validity adjudication (the corpus's legal floor was EU-only); admitted only after source verification against the NYC DCWP's official page/rules. Decision relevance: if AutoQA verdicts influence pay, standing, or continued engagement of NYC-based workers and function as an employment decision tool, bias-audit and notice obligations attach — the D5 workforce census must be two-sided (EU and US), not EU-only.",
      "implication": "D5 gates on a two-sided workforce census; US-side counsel review joins EU counsel review as a launch precondition where scope conditions may hold.",
      "source": {"title": "NYC Local Law 144 — Automated Employment Decision Tools (DCWP)", "url": "https://www.nyc.gov/site/dca/about/automated-employment-decision-tools.page", "date": "2023-07-05 (enforcement start)", "type": "law"},
      "used_on": [
        {"page": "decisions.html", "anchor": "d5", "label": "Decisions · first pilot locale and legal gate"}
      ]
    },
    {
      "id": "DA-02",
      "claim": "The EEOC Uniform Guidelines on Employee Selection Procedures (1978, codified at 29 CFR Part 1607) require that selection procedures with adverse impact be validated (criterion, content, or construct strategies), apply to any procedure used as a basis for an employment decision, and adopt the four-fifths rule as the practical test of adverse impact.",
      "evidence": "Provenance: proposed from model knowledge during the 2026-07-15 validity adjudication; admitted only after source verification against the eCFR text. Decision relevance: consequential AutoQA verdicts over US workers are selection-procedure-shaped; validation obligations and adverse-impact monitoring are standards-grade constraints on D1 enforcement weight and A4 authority boundaries, independent of EU law.",
      "implication": "Where US workers face consequential verdicts, validity evidence and adverse-impact monitoring are compliance requirements, not research niceties — reinforcing measurement-first sequencing.",
      "source": {"title": "29 CFR Part 1607 — Uniform Guidelines on Employee Selection Procedures", "url": "https://www.ecfr.gov/current/title-29/subtitle-B/chapter-XIV/part-1607", "date": "1978 (current codification)", "type": "law"},
      "used_on": [
        {"page": "decisions.html", "anchor": "d5", "label": "Decisions · first pilot locale and legal gate"},
        {"page": "decisions.html", "anchor": "a4", "label": "Decisions · authority boundaries (US side)"}
      ]
    },
    {
      "id": "DA-03",
      "claim": "The multitask principal–agent model (Holmström & Milgrom, 1991) shows that when some task dimensions are measurable and others are not, high-powered incentives on the measured dimensions divert agent effort away from the unmeasured ones — and can make low-powered or no incentives optimal.",
      "evidence": "Provenance: proposed from model knowledge during the 2026-07-15 validity adjudication; admitted only after source verification. Decision relevance: this is the theorem behind the feedback-monoculture and teaching-to-the-judge concerns — it bounds enforcement weight wherever quality is only partially measured (D1), and grounds the ban on stylistic guidance and rank-referenced feedback as incentive design rather than taste.",
      "implication": "Enforcement weight on partially-measured quality is bounded by theorem, not prudence; the kernel's K9/T7/T11 derivations cite this result as their economic formalization.",
      "source": {"title": "Multitask Principal–Agent Analyses: Incentive Contracts, Asset Ownership, and Job Design (JLEO 7)", "url": "https://academic.oup.com/jleo/article/7/special_issue/24/861573", "date": "1991", "type": "academic"},
      "used_on": [
        {"page": "decisions.html", "anchor": "d1", "label": "Decisions · enforcement weight (incentive theorem)"}
      ]
    },
    {
      "id": "DA-04",
      "claim": "ANSI/ASQ Z1.4 is the American national standard for acceptance sampling by attributes, providing AQL-indexed sampling plans with normal, tightened, and reduced inspection and defined switching rules between them; it descends from MIL-STD-105.",
      "evidence": "Provenance: proposed from model knowledge during the 2026-07-15 validity adjudication; admitted only after source verification against ASQ's official pages. Decision relevance: the audit-rate, escalating-inspection, and de-graduation machinery in A5/P10 is a standardized, tabled discipline with eighty years of deployment — imported rather than invented, which changes the A5 conversation from 'trust our design' to 'we are instantiating a national standard'.",
      "implication": "Audit rates and graduation/de-graduation switching adopt the acceptance-sampling template as the starting point; deviations from the standard are the thing requiring justification.",
      "source": {"title": "ANSI/ASQ Z1.4 — Sampling Procedures and Tables for Inspection by Attributes", "url": "https://asq.org/quality-resources/z14-z19", "date": "current standard", "type": "standard"},
      "used_on": [
        {"page": "decisions.html", "anchor": "a5", "label": "Decisions · permanent audit channel (sampling standard)"}
      ]
    }
  ]
}