employment-decision-record-audit-stream

Employment Decision Record Audit Stream v0.1 draft. Per-hiring / promotion / performance / termination AI-tool-access events, hash-chained and signed, designed to bridge HR-system semantics (Workday + UKG + Greenhouse + Lever + iCIMS) to the Kinetic Gain Protocol Suite audit-stream spine. The Operator surface that lets an employer's HRIS + ATS + vendor AI tool emit Suite-compliant audit events covering the four primary employment-AI decision surfaces — hiring (sourcing / screening / interview-scoring / offer), promotion, performance evaluation, termination — under one schema with kind enum branching.

Part of the Kinetic Gain Protocol Suite. Opens the HR Tech / Employment AI vertical alongside the existing HealthTech, EdTech, PropTech, and InsurTech 6-packs.

Status: v0.1 draft. Schema at schema/employment-decision-event.schema.json, Node verifier at src/verify.mjs, canonical example at examples/momentumhr-hiring-2026q4/.

Why this exists

NYC Local Law 144 took effect July 5, 2023 and enforcement began. Employers using an Automated Employment Decision Tool (AEDT) that substantially assists or replaces discretionary decision-making for candidates / employees in NYC must: (1) commission an independent bias audit annually; (2) publish a summary of bias-audit results on the careers site; (3) provide candidates 10 business days' notice that an AEDT will be used + the job-qualifications-and-characteristics it assesses + how to request an alternative selection process. Illinois AI Video Interview Act (820 ILCS 42/) governs video-interview AI. Maryland HB 1202 governs facial recognition in hiring. EEOC AI Guidance (May 2023) + ADA + Section 504 + Title VII + ADEA + GINA + OFCCP-federal-contractor obligations form the federal floor.

What's missing is a uniform audit-stream that names which AI tool read which HR resource at which timestamp under which regulatory basis under which buyer-published Decision Card — provable to an EEOC investigator, a NYC Department of Consumer and Worker Protection examiner, an OFCCP compliance officer, a state civil-rights agency, or a plaintiff's-side employment attorney without forcing the employer to rebuild its HRIS or the vendor to abandon its data model.

This repo defines the schema + verifier for that stream.

Two distinct invariants

Most prior Suite audit streams enforce one human-in-loop invariant. The HR Tech audit stream enforces two, because NYC LL 144 adds an orthogonal candidate-notice obligation:

Invariant 1: human-hiring-decision required. Any event whose kind is in the adverse-action-capable set AND whose ai_recommendation.recommendation is in {decline, do-not-promote, performance-below, terminate-recommended} MUST set ai_recommendation.human_hiring_decision_required = true. The verifier exits 3 when violated.

Invariant 2: NYC LL 144 candidate-notice provided. Any event where agent.is_aedt_per_nyc_ll_144 = true AND kind is in the external-candidate-facing set MUST set candidate_notice_provided.notice_provided = true. The verifier exits 4 when violated.

The shape

Field group	Purpose
event_id, timestamp, kind	Append-only identity (14-kind taxonomy across hiring + promotion + performance + termination + accommodation-request + deletion)
source	Emitting system (HRIS, ATS, vendor AI tool)
subject_ref	Tokenized candidate / employee ID — raw SSN, name, DOB, email, social handle MUST NOT appear
subject_role_context	external-candidate / internal-candidate / current-employee / contingent-worker / departing-employee
resource	18-type resource taxonomy (resume / cover-letter / application / video-interview recording or transcript / assessment / background check / credit check / social-media scrape / reference check / performance review / compensation history / time-and-attendance / workforce-planning projection / adverse-action notice / reasonable-accommodation request) + tokenized ID + logical fields accessed
agent	AI tool card URL, Decision Card URL, is_aedt_per_nyc_ll_144 flag (the AEDT-or-not distinction is the LL 144 trigger), optional principal (recruiter / hiring manager / HRBP)
regulatory_basis	15-doctrine taxonomy spanning federal (Title VII, ADA, Section 504, ADEA, GINA, OFCCP, EEOC AI Guidance) + state (NYC LL 144, IL AI Video Interview Act, MD HB 1202, CA AB 331, CO AI Act) + consent + judicial-order
records_of_disclosure_status	NYC LL 144 bias-audit record + 29 CFR §1602 EEOC recordkeeping log
candidate_notice_provided	OPTIONAL — required by invariant 2 when AEDT + external-candidate-facing kind
ai_recommendation	OPTIONAL — recommendation, reason codes, model version, human_hiring_decision_required invariant
accommodation_pathway	OPTIONAL — present on accommodation-request-evaluated events; ADA reasonable-accommodation traceability
signature	Optional ed25519
prev_hash, hash	Hash chain (SHA-256 over canonical JSON of event minus hash)

Verifier exit codes

Code	Meaning
0	All events valid + chain intact + both invariants preserved
1	Schema validation failed
2	Chain validation failed
3	human-hiring-decision invariant violated
4	NYC LL 144 candidate-notice invariant violated
5	Usage / IO error

Canonical example

examples/momentumhr-hiring-2026q4/source.json — three events from MomentumHR Inc.'s 2026 Q4 stream:

1. employment.hiring.resume-screened — VendorE HireAssess v2.x reads four tokenized resume fields for req-2026-1843 (Senior Backend Engineer). AEDT flag = true. NYC LL 144 candidate notice provided + alternative-selection-process offered. Candidate name / email / phone / address / DOB redacted before reaching the model.
2. employment.hiring.recommendation-produced — VendorE recommends advance-with-followup with model_confidence 0.78 + human_hiring_decision_required: true. (NOT a final decision.)
3. employment.hiring.adverse-action-evaluated — MomentumHR's human hiring manager issues the decline notice under EEOC AI Guidance + Title VII disparate-impact framework, after the structured-interview round. (Demonstrates the AI's advance-with-followup recommendation correctly deferred to human judgment.)

Quick start

npm install
npm run build:examples
npm run verify

Composes with

Repo	Role
evidence-bundle-spec	Underlying audit-stream conventions
fhir-resource-access-audit	Sibling HealthTech audit-stream
student-data-access-audit-stream	Sibling EdTech audit-stream
mortgage-decision-record-audit-stream	Sibling PropTech audit-stream
insurance-decision-record-audit-stream	Sibling InsurTech audit-stream

Compliance posture

HR-Tech-readiness scaffolding for EEOC AI Guidance (May 2023) governance + recordkeeping, Title VII / ADA / Section 504 / ADEA / GINA federal-floor recordkeeping, OFCCP federal-contractor recordkeeping (where applicable), NYC Local Law 144 AEDT bias-audit + candidate-notice data sourcing, Illinois AI Video Interview Act recordkeeping, Maryland HB 1202 facial-recognition disclosure, and emerging state employment-AI laws. The schema + verifier support an employer's program toward those expectations but do not by themselves establish compliance with any of them. Per the standing public-language guardrail: readiness · evidence · posture · controls · scaffolding — never "EEOC-compliant" or "NYC-LL-144-attested" without an external attestation (NYC LL 144 attestation specifically requires an independent third-party bias auditor).

License

MIT — see LICENSE.