Perform a SEPARATE FROZEN NO-TOOLS semantic audit. Do not browse, execute code, call tools or add new source facts. File upload was unavailable, so audit only the complete text packet below. Do not claim you read or verified the untransmitted full JSON, local tests, file bytes or hash. Treat reported test results as author-reported. Return compact JSON with verdict, critical_conflicts, required_changes_for_draft, nonblocking_holds, accepted_decisions and audit_scope. Evaluate suitability for a PUBLIC REVIEWABLE DRAFT, not canonical assurance.

FROZEN CANDIDATE: WM-ORG-017 Performance Objective / Review; aggregate root = one purpose-qualified case for one subject and one governing work context. Codex authored the structured JSON after comparing your independent browser memo and Grok independent memo; neither provider is credited with emitting that JSON. Original Claude CLI transport timed out, recorded separately. Reported validator result: schema-valid, 8 primary-source precedents, 7 bundles, 14 layers, 24 findings, 72 distinct questions, 24 artifacts and 8 proposed operations. The full candidate packet SHA-256 is ae9d5dfbcfd9b509cf8631440d62bf3cc3c48098b61bc434584d643efdb55032; this is a reference identifier only, not something you can independently attest.

BOUNDARY DECISIONS: Person, Organization, Employment, Membership, Assignment, Contract and original evidence keep their own masters. Team/shared objectives are referenced; case-local objective commitments are owned here. Shared Calibration Session is a peer; 0..n case-local before/after outcomes are owned here. Independently issued awards/payments remain external; local acknowledgement/recommendation cannot write a grade or payroll. Scale versions have separate identities; numeric/ordinal assessment requires one exact version, narrative-only requires none. Case IDs are stable; revisions and assessment assertion IDs are different. Subject/context/program/period is a duplicate-detection signal, not a universal unique key, because purpose-specific review lanes may coexist. One governing context per case, with advisory related contexts. Employee profile uses Employment; contractor uses Contract; founder uses an explicit mandate. Membership-only and team-subject profiles remain follow-up scope. No universal annual cadence, minimum tenure, numeric scale, monitoring ceremony or appeal-depth limit.

STRUCTURE: B1 identity/context: case identity and master; revision lineage; governing context; neighbor bindings. B2 objectives/period: review period/program; purpose declaration; objective success criteria and opportunity constraints; objective change/shared alignment. B3 evidence/attribution: source/method/observed and recorded time/unit of analysis; quality/admitted-contested-excluded and proxy exclusion; attributable contribution claims with optional unknown/asserted/disputed shares and method; conflicting evidence and outcome qualification. B4 assessment/scales: rater authority and recusal; assisted drafting versus accountable human issuance; contextual scale version and loss-aware comparison; 0..n assessment assertions with at most one current issued outcome in this reference profile. B5 calibration/recognition: repeatable before/after adjustments linked to external sessions; review-only outcome and separate downstream decisions; acknowledgement/reward recommendation. B6 contests/disclosure: fact correction, attribution correction, opinion contest, process contest, proxy exclusion; correction/notification hooks; purpose-limited one-object Projection with source revision, recipient, field set, expiry and authority. B7 governance/interop: startup and enterprise profiles, restricted retention/erasure/legal-hold policy; source crosswalk, bounded validation and explicit holds.

OPERATIONS: open scoped case; agree/revise objective; cite/admit evidence; draft assessment; human issue assessment; record calibration; resolve contest; produce permitted case projection. These are proposed guarded operations, NOT implemented HRIS actions. No operation can hire, dismiss, promote, pay, create a portable person score or turn activity counts into an automatic decision. Agents can organize evidence and draft questions under delegation; unknown stays unknown.

KEY INVARIANTS: one case/master/governing context; no duplicate person or employment; unknown evidence is not zero contribution; shares are claims not IP ownership; source-qualified conflicts retained; issued outcome cites admitted non-proxy evidence, accountable human and exact optional scale; repeated calibration and appeal preserve before/after and authority; correction supersedes a prior assertion and preserves valid versus recorded time; all history subject to retention/erasure policy, not retain-forever; each Projection describes one case, multiple objects form a context pack; no automatic employer dossier transfer; disclose applicable disagreement and qualifications with assessment, with separately authorized exceptions where necessary; leave/opportunity context can qualify objectives without copying medical details; preserve original language/translation provenance.

COMPANION IMPLEMENTATION (bounded reference contract, NOT native Dimension binding or production IAM): JSON Schema 2020-12, strict fields. Required root fields: format=vercy-performance-case-example, schema_version=0.1.0, case URI, revision integer, program URI, profile(employee|contractor|founder), subject URI, governing_context(kind, reference, owner_org_ref, authority_ref), start/end RFC3339 interval, state(draft|agreed|evidence-open|assessment-open|issued|disputed|corrected|closed|retired), purpose, arrays of scale_versions/objectives/evidence/assessments/calibrations/appeals/recognition/disclosures/history, policy refs. Evidence has source/method/time/unit/kind/purpose/context/use_state/reason/attribution claims. Assessment has author, actor_kind, narrative, evidence_refs, nullable scale_ref/scale_version/rating, issue time, status(draft|issued|superseded), supersedes, decision_effect=review-only, known_conflict. Calibration has distinct before/after assessment refs, panel, reason, time. Appeal has assessment ref, typed ground, statement, state, authority and optional correction ref. Recognition has basis/kind and changes_rating=false. Disclosure has one canonical subject, purpose, recipient, explicit JSON-pointer field paths, authority URI, expiry, source owner and source revision. Policy flags automatic_employment_decisions=false, portable_person_score=false. The validator validates referential/temporal/selected semantic constraints; it does NOT authenticate authority URIs, enforce real access, evaluate law or export data. Presence of a disclosure record is NOT permission to send it. A production policy engine must evaluate each field, dispute qualification, audience and purpose.

REPORTED TESTS: 4 positive cases (employee with two employments but one reviewing context, contested shared credit, corrected assessment and isolated source owner; founder draft without Employment/scale/assessment; narrative contractor with contract context; repeated calibration plus repeated appeals). 23 negative mutations expected rejected: proxy-only rating, unpinned scale, automated issuance, hidden conflict, erased correction predecessor, master transfer to new employer, wrong Projection subject, evidence purpose change, wrong evidence employment, absent attribution, reversed period, duplicate child ID, identical calibration before/after, wildcard disclosure, recognition grade mutation, automatic employment flag, portable score flag, missing case authority, stale projection revision, unknown JSON-pointer field, one-sided assessment export omitting dispute context, contractor with Employment kind, absent disclosure authority. Reported 27 pass, 0 fail. These do not prove full lifecycle transition enforcement, jurisdiction validity, anonymization, production permissions or universal completeness.

SOURCE DISCIPLINE: adopted official HR Open public EPM overview, W3C ORG/PROV final Recommendations, SFIA 9 LEDA distinction, SPACE primary publication, OPM public cycle overview, Vercy whole-object profile, ISO 30414:2025 public abstract only. No full HR-XML field crosswalk, ISO formula, licensed taxonomy copying, legal compliance or AI Act timing claim. Historical editor commits, legal mirrors, vendor timing assertions and incomplete ellipsis URLs were not promoted as primary authority. Source refs in findings ground precedents; exact new schema fields/invariants are design proposals, not quoted standard clauses.

VISIBLE HOLDS: full source/licence and field-level crosswalk verification; specialist legal/regional/employee-representation and retention review; production IAM/adapters/storage; native Dimension installation fixture; group-subject, leave fairness, multilingual and membership-only profiles; ratification of shared Program/Scale/CalibrationSession/Recognition objects; complete broad EM-PEO-06 research acceptance. This publication is usable as a versioned REVIEWABLE-DRAFT research specification with companion examples; the broader contour stays partial. Is there a critical contradiction in this frozen design that prevents even that honestly scoped draft? Separate mandatory draft fixes from nonblocking further research.