Skip to content

MEASURES.md

Latest commit

 

History

History
511 lines (431 loc) · 37.5 KB

MEASURES.md

File metadata and controls

511 lines (431 loc) · 37.5 KB
Raw

WorkWell Measure Catalog

WorkWell Measure Studio implements the Total Worker Health (TWH) model: OSHA occupational safety compliance and clinical quality / wellness measures managed in a single platform. The TWH instance seeds all three categories on startup.

Catalog summary

Category Count Status CQL
OSHA occupational safety — fully evaluated 4 Active Full CQL, runnable
OSHA occupational safety — catalog only 2 Draft / Deprecated Partial or no CQL
HEDIS wellness — fully evaluated 5 Active Full CQL, runnable
Permanent immunization panel — fully evaluated 3 Active Full CQL, runnable (series-completion; MMR, Varicella, Hep B)
CMS eCQM — fully evaluated 2 Active Full CQL, runnable (CMS125v14, CMS122v14)
CMS eCQM catalog (2026 performance period) 47 Draft Catalog entry only — CQL authoring pending
Total 63

Runnable (full CQL): 14 — 4 OSHA + 5 HEDIS + 3 immunization panel + 2 CMS eCQM. Hepatitis B was promoted from an Approved catalog entry to Active (E10.6).

Outcome buckets (all measures): COMPLIANT, DUE_SOON, OVERDUE, MISSING_DATA, EXCLUDED.

Category 1 — OSHA Occupational Safety (Full CQL)

These four measures have complete CQL libraries, are seeded as Active, and run against the synthetic employee dataset.

1.1 Annual Audiogram Completed

Define logic

  • Program eligibility: In Hearing Conservation Program
  • Exemption: Has Active Waiver
  • Recency: Most Recent Audiogram Date
  • Aging metric: Days Since Last Audiogram

Outcome mapping

  • EXCLUDED when Has Active Waiver = true
  • MISSING_DATA when enrolled, not waived, no exam date
  • OVERDUE when enrolled, not waived, days since exam > 365
  • DUE_SOON when enrolled, not waived, days in (336..365)
  • COMPLIANT when enrolled, not waived, days <= 335

1.2 HAZWOPER Annual Medical Surveillance

Define logic

  • Program eligibility: In HAZWOPER Program
  • Exemption: Has Medical Exemption
  • Recency: Most Recent Surveillance Exam Date
  • Aging metric: Days Since Last Exam

Outcome mapping

  • EXCLUDED when Has Medical Exemption = true
  • MISSING_DATA when in program, not exempt, no exam date
  • OVERDUE when in program, not exempt, days since exam > 365
  • DUE_SOON when in program, not exempt, days in (335..365]
  • COMPLIANT when in program, not exempt, days <= 335

1.3 Annual TB Screening

Define logic

  • Program eligibility: In TB Screening Program
  • Exemption: Has Medical Exemption
  • Recency: Most Recent TB Screen Date
  • Aging metric: Days Since Last TB Screen

Outcome mapping

  • EXCLUDED when Has Medical Exemption = true
  • MISSING_DATA when eligible, not exempt, no TB screen date
  • OVERDUE when eligible, not exempt, days since last screen > 365
  • DUE_SOON when eligible, not exempt, days in (330..365]
  • COMPLIANT when eligible, not exempt, days <= 330

1.4 Flu Vaccine This Season

Define logic

  • Program eligibility: Clinical Facing Employee
  • Exemption: Has Valid Exemption
  • Season completion: Flu Vaccine This Season

Outcome mapping

  • EXCLUDED when Has Valid Exemption = true
  • COMPLIANT when eligible, not exempt, vaccinated this season
  • DUE_SOON when eligible, not exempt, not vaccinated this season, last vaccine ≤ 365 days ago
  • OVERDUE when eligible, not exempt, not vaccinated this season, last vaccine > 365 days ago
  • MISSING_DATA when eligible, not exempt, no flu vaccine record on file

Category 2 — OSHA Occupational Safety (Catalog Only)

These two measures are seeded for catalog richness and demonstrate the full measure lifecycle (Draft → Approved → Deprecated). They have no runnable CQL evaluation. (Hepatitis B Vaccination Series was promoted to a runnable Active measure in E10.6 and is now a multi-alternative series — see Category 3c.)

Name Policy Ref Status Tags
Respirator Fit Test OSHA 29 CFR 1910.134 Draft v0.9 surveillance, respiratory, osha
Lead Medical Surveillance OSHA 29 CFR 1910.1025 Deprecated v1.1 surveillance, lead, osha

Category 3 — HEDIS Wellness (Full CQL)

Five employer wellness / HEDIS-style measures with complete CQL and active evaluation. These represent the wellness side of TWH — chronic disease management, preventive health screening, and adult immunization programs run by occupational health departments.

3.1 Hypertension BP Screening

  • Policy reference: HEDIS BPC / JPMC Wellness Rewards
  • CQL file: backend-ts/measures/hypertension.cql
  • Tags: wellness, hypertension, cardiovascular
  • Compliance window: 365 days (DueSoon 336–365)

3.2 Diabetes HbA1c Monitoring

  • Policy reference: HEDIS HBD / JPMC Wellness Rewards
  • CQL file: backend-ts/measures/diabetes_hba1c.cql
  • Tags: wellness, diabetes, hba1c
  • Compliance window: 180 days biannual (DueSoon 161–180)

3.3 BMI Screening & Counseling

  • Policy reference: HEDIS WCC / Cigna Healthcare Wellness
  • CQL file: backend-ts/measures/obesity_bmi.cql
  • Tags: wellness, bmi, obesity
  • Compliance window: 365 days annual

3.4 Cholesterol LDL Screening

  • Policy reference: HEDIS CBP / JPMC Wellness Rewards
  • CQL file: backend-ts/measures/cholesterol_ldl.cql
  • Tags: wellness, cholesterol, cardiovascular
  • Compliance window: 365 days annual

The four chronic-disease/screening measures above use the same outcome pattern:

  • EXCLUDED when Has Medical Exemption = true
  • MISSING_DATA when enrolled, not exempt, no qualifying lab/screening date
  • OVERDUE when enrolled, not exempt, days since last event > compliance window
  • DUE_SOON when enrolled, not exempt, days approaching window end
  • COMPLIANT when enrolled, not exempt, days within window

3.5 Adult Immunization Status — Td/Tdap (AIS-E)

Define logic

  • Program eligibility: In Immunization Program
  • Exemption: Has Td/Tdap Contraindication (clinical contraindication Condition)
  • Refusal: Refused Td/Tdap (documented tdap-refusal Condition — does NOT exclude; case stays open)
  • Recency: Most Recent Td/Tdap Date
  • Aging metric: Days Since Last Td/Tdap

Outcome mapping

  • EXCLUDED when Has Td/Tdap Contraindication = true
  • MISSING_DATA when enrolled, not contraindicated, no Td/Tdap record on file
  • OVERDUE when enrolled, not contraindicated, days since last dose > 3650 (>10 years)
  • DUE_SOON when enrolled, not contraindicated, days in (3590..3650]
  • COMPLIANT when enrolled, not contraindicated, days <= 3590

Refusal: a Refused define in evidence_json flags the refusal; the case stays OPEN and is routed to a case manager for intervention. Refusal does not trigger an EXCLUDED outcome.

Advisory immunization forecast: for adult_immunization cases, GET /api/cases/:id attaches an advisory immunizationForecast covering all 3 ACIP series (Td/Tdap, Influenza annual, Hepatitis B 3-dose). It is computed by the ImmunizationForecast port and is advisory only — it never affects the CQL Outcome Status (ADR-012).

The port has two implementations (ADR-029, 2026-07-13): the simulated forecaster (the default — ACIP-style windows over its own deterministic synthetic dose history) and a real adapter against a self-hosted ICE sidecar (HLN's ACIP-maintained Immunization Calculation Engine), selected by WORKWELL_IMMZ_ICE_BASE_URL alone and falling back whole to the simulated forecaster on any failure. When ICE is on, the forecast carries ICE's own recommendation and reason codes (e.g. ICE RECOMMENDED (DUE_NOW, ADMINISTER_TDAP_OR_TD)).

ICE and a WorkWell measure can legitimately disagree, and that is not a defect. ICE scores the full ACIP schedule for a vaccine group; a WorkWell measure scores its own authored rule. A subject with 2 Hep B doses reads COMPLIANT under hepatitis_b_vaccination_series if those doses complete the Heplisav-B alternative, while ICE — told the doses are a traditional adult formulation — will correctly propose dose 3. The CQL Outcome Status remains the sole compliance authority (ADR-008/ADR-012); the ICE forecast is clinical advice sitting beside it, not a second verdict.

Category 3c — Permanent Immunization Panel (series-completion CQL)

These three measures introduce the PERMANENT compliance class (E10.1 / E10.6): compliance is proven by a completed dose series, not recency — once the series is on file the employee stays COMPLIANT indefinitely ("once compliant, always compliant"). This contrasts with every other measure, which is RECURRING (windowed days-since-last with DUE_SOON/OVERDUE). The class is declared as complianceClass in each measure's YAML binding (default RECURRING); it is descriptive/routing metadata only — the CQL Outcome Status remains the sole compliance authority (ADR-008). These are the repo's first series-completion CQL measures (Count("Valid Doses") >= N, no recency filter).

Measure id Series COMPLIANT when Excludes
MMR Immunity mmr 2 doses (CVX 03/94) ≥ 2 valid MMR doses on file contraindication
Varicella Immunity varicella 2 doses (CVX 21) ≥ 2 valid varicella doses on file contraindication
Hepatitis B Vaccination Series hepatitis_b_vaccination_series multi-alternative (E11.2c): Heplisav-B 2 doses (CVX 189, ≥28d apart) OR traditional 3 doses (CVX 08/43/44/45, ACIP min intervals 28/56d) either alternative series complete contraindication

Outcome mapping (MMR / Varicella):

  • EXCLUDED when a documented contraindication Condition is present
  • COMPLIANT when enrolled, not contraindicated, and Dose Count >= 2 (regardless of dose age)
  • MISSING_DATA otherwise — including a partial series (0 < Dose Count < 2), surfaced by the roster read model as IN_PROGRESS (E10.5)
  • DUE_SOON / OVERDUE are not applicable to PERMANENT measures

Hepatitis B (multi-alternative, E11.2c / #183): COMPLIANT requires a complete alternative series — "Heplisav-B Complete" (2 doses CVX 189 ≥28 days apart) OR "Traditional Complete" (3 doses CVX 08/43/44/45 with consecutive gaps ≥28 and ≥56 days). A union "Dose Count" define (any Hep B CVX) is kept for the roster's method string only — so a mid-traditional-3 series shows the approximate "1 of 2 doses on file" (the IN_PROGRESS denominator uses the top-level series.requiredDoses 2; the canonical bucket is CQL-authoritative). EXCLUDED/MISSING_DATA/refusal behave as the other two. The codegen capability is E11.2c (ADR-015); this repoint is additive seed/app data with no schema change.

A documented refusal (declination) Condition does not change the canonical bucket; it is surfaced as DECLINED by the roster read model (E10.5) and keeps the case open (same pattern as adult_immunization). Titer-proves-immunity ("Allow positive titer") for Hep B remains deferred.

Category 3b — CMS eCQM (Full CQL)

Two CMS eCQM measures promoted from Draft catalog to Active with full CQL evaluation:

3b.1 Breast Cancer Screening (CMS125v14 / MIPS 112)

  • Policy reference: CMS125v14 (2026 eCQI; v15/2027 annual roll-forward — stay on v14 for 2026)
  • CQL file: backend-ts/measures/cms125.cql (v2.0.0 production faithful-subset)
  • Tags: ecqm, cms, cancer-screening, preventive
  • Measurement period: 12 months (periodMonths: 12)
  • Fidelity: structural report at GET /api/measures/cms125/fidelity

Official criteria (CMS125-v14.0.000-QDM): women 42-74 + visit during MP; mammogram on/between Oct 1 two years prior to MP and end of MP (VSAC Mammography); DENEX hospice / mastectomy / palliative (66+ LTC + frailty/AI = Phase 2 residual).

Outcome mapping (higher is better):

  • EXCLUDED — DENEX (mastectomy / hospice / palliative)
  • COMPLIANT — in IPP with qualifying mammogram
  • OVERDUE — in IPP without qualifying mammogram
  • MISSING_DATA — not in IPP
  • DUE_SOON — not used (cleaner eCQI proportion story)

3b.2 Diabetes: Glycemic Status Assessment Greater Than 9% (CMS122v14 / MIPS 1)

  • Policy reference: CMS122v14 (2026 eCQI; stay on v14)
  • CQL file: backend-ts/measures/cms122.cql (v2.0.0 production faithful-subset)
  • Tags: ecqm, cms, diabetes
  • Measurement period: 12 months
  • Catalog/display name matches eCQI title; library DiabetesHbA1cPoorControlCQL-2.0.0
  • Fidelity: structural + estimate + subset + literal fqm (GET /api/measures/cms122/fidelity + /diff)

Official criteria (CMS122-v14.0.000-QDM): age 18-75 + diabetes + visit; NUMER most recent HbA1c or GMI (LOINC 97506-0) in MP > 9% or missing/not performed; DENEX hospice + palliative (66+ LTC + frailty/AI = Phase 2).

Outcome mapping (lower-is-better eCQM rate; NUMER maps to OVERDUE):

  • EXCLUDED — DENEX
  • OVERDUE — in IPP and numerator (poor control or missing assessment)
  • COMPLIANT — in IPP with glycemic assessment <= 9%
  • MISSING_DATA — not in IPP

eCQM accuracy posture (vs eCQI — read this before claiming parity)

Short answer (2026-07 production-faithful promotion): the two runnable CMS measures are eCQI-aligned faithful subsets for 2026 (v14). They are still not full multi-library QICore MAT packages for MIPS submission.

What we ship Count Relationship to eCQI How we prove it
Runnable Active CMS CQL 2 — CMS122v14, CMS125v14 Faithful official-subset production CQL: 12-month MP, age/sex/visit, VSAC OIDs, GMI, Oct-1 mammogram window, hospice/palliative/mastectomy. Residual Phase 2: 66+ LTC + frailty/AI. Dual-coded synthetic data. Structural fidelity both; CMS122 also estimate + subset + literal fqm
Draft CMS catalog entries 47 Metadata only — correct CMS ID, v14 = 2026, MIPS ID, title Terminology audit 2026-07-08
OSHA / HEDIS / permanent vax 12 Not CMS eCQMs Separate authorities

Demo claim that is honest:

We evaluate CMS122v14 and CMS125v14 — the 2026 eCQI Eligible Clinician measures (MIPS 001 / 112) — with production CQL aligned to official population criteria — and, for CMS122 and CMS125 on the demo/production stack since 2026-07-30, replaced outright by CMS's published QI-Core artifacts (PR-9c / ADR-045, ADR-046). We can show structural fidelity (and for CMS122, a literal official package comparison). We do not claim full MAT multi-library submission packages or 2027 v15 until we cut over.

Remaining accuracy work: Phase 2 DENEX (LTC/frailty); optional CMS125 literal ELM; annual roll-forward to v15/2027 when product targets that year.

Category 4 — CMS eCQM Catalog (2026 Performance Period)

47 official CMS electronic Clinical Quality Measures seeded as Draft v1.0 catalog entries (CMS125v14 and CMS122v14 are now Active with full CQL — see Category 3b). The policy_ref field stores the CMS eCQM ID (e.g., CMS128v14). The spec_json stores cmsEcqmId and mipsQualityId for downstream tooling. CQL authoring for the remaining catalog entries is future work.

The measures page renders CMS IDs as blue mono badges to distinguish them from OSHA CFR citations and HEDIS references.

Two new measures added in 2026 vs 2025: CMS146v14 (Appropriate Testing for Pharyngitis) and CMS154v14 (Appropriate Treatment for URI) in the new Respiratory / Antimicrobial Stewardship domain; CMS1173v1 (Diagnostic Delay of VTE) added to Cardiovascular; CMS1154v1 (Screening for Abnormal Glucose Metabolism) added to Diabetes. CMS249v7 (DXA Scans) retired from 2026 eligible clinician list.

Domain breakdown

Domain Measures
Mental Health / Behavioral CMS2v15, CMS128v14, CMS136v15, CMS137v14, CMS149v14, CMS159v14, CMS177v14
Cardiovascular CMS22v14, CMS90v15, CMS135v14, CMS144v14, CMS145v14, CMS165v14, CMS347v9, CMS1173v1
Diabetes CMS122v14, CMS131v14, CMS142v14, CMS951v4, CMS1154v1
Cancer Screening / Preventive CMS69v14, CMS124v14, CMS125v14, CMS130v14, CMS138v14, CMS139v14, CMS153v14, CMS155v14
Respiratory / Antimicrobial Stewardship CMS146v14, CMS154v14
Pediatric / Immunization CMS74v15, CMS75v14, CMS117v14
HIV / Infectious Disease CMS314v3, CMS349v8, CMS1157v2, CMS1188v3
Oncology CMS129v15, CMS157v14, CMS645v9, CMS646v6
Ophthalmology CMS133v14, CMS143v14
Functional Status / Orthopedic CMS56v14
Medication Safety CMS68v15, CMS156v14
Care Coordination CMS50v14
Urology CMS771v7
Radiology / Patient Safety CMS1056v3

Full list with MIPS Quality IDs is embedded in MeasureService.CMS_ECQM_CATALOG and visible in the measures catalog at /measures.

Canonical Status Source

For all runnable measures, the canonical stored status is the value of CQL define Outcome Status.

Persistence path:

  1. CQL engine evaluates all defines.
  2. Outcome Status string is read from expression results.
  3. Status is persisted in outcomes.status.
  4. Full define-level results are persisted in outcomes.evidence_json.expressionResults.

Evidence Traceability

Each outcome evidence payload includes:

  • Key eligibility/exemption/recency defines for the measure.
  • Computed day-difference define where applicable.
  • Outcome Status define result.
  • why_flagged derived fields for the UI (last exam date, compliance window, days overdue, waiver status).

Implementation Notes

  • Incremental/delta evaluation (#263 / ADR-035, 2026-07-24 — descriptive only, ADR-008). A recurring population run may reuse a subject's prior Outcome Status (copy-forward) instead of re-running the CQL, when the subject's data + measure logic are unchanged and the status cannot have moved. This never changes any measure's semantics or authored CQL — it decides only whether to re-ask the engine, and a reused subject's outcome is byte-identical to a fresh run (proven by the parity suite). Only windowed-recency + PERMANENT measures get across-day reuse; flu_vaccine (seasonal) and cms122/cms125 (measurement-period) are same-day-only, since their status is not a monotone function of days-since-event. Inert unless WORKWELL_INCREMENTAL_EVAL=true. See DATA_MODEL §3.27, ARCHITECTURE.

  • CQL→SQL (WCDB) demo translations (#292 / ADR-034, 2026-07-20 — descriptive only, ADR-008). Four observation-backed windowed-recency measures (hypertension, cholesterol_ldl, obesity_bmi, diabetes_hba1c) also have generated MariaDB SQL against the WebChart dev-DB schema: pnpm generate:sql (engine/cql/codegen/generate-sql-cli.ts) templates the same rule params that compile to CQL (plus the crosswalk's LOINC sets, loincCodesForMeasure) into parameterized per-patient / single-patient / cohort statements, committed under wcdb-fhir-shim/sql/ (freshness-tested) and executed only by the standalone shim's compliance API. CQL remains the sole Outcome Status authority — the SQL is parity-gated per ADR-025 (the CQL engine over the shim's FHIR output is the oracle; cohort counts verified equal 2026-07-20) and serves nothing in the product. Series-completion SQL is deliberately absent (WCDB has no immunization table to prove parity against).

  • Live WebChart enrollment and groups (ADR-033). When the live tenant seam is configured, WORKWELL_WEBCHART_ENROLLMENT_JSON may map raw Patient ids to explicitly enrolled measure ids. Otherwise the safe demo default enrolls every live subject in every ROSTER_ELIGIBLE_MEASURES member. Enrollment only supplies occupational-health context: each measure's CQL age, sex, diagnosis, visit, exclusion, and clinical-data gates remain authoritative, and CQL alone sets Outcome Status. The demo-segment baseline (All Employees) now covers the fixed live site WebChart out of the box (WEBCHART_LIVE_SITE, folded into the seed's site list), so on any fresh DB (a local demo, a new instance) WebChart subjects are applicable and the /compliance roster shows their real per-measure chips immediately — no manual admin step. (An already-seeded DB predating this change keeps its old baseline row, since seeding is name-idempotent and never auto-mutates an operator's segment; the owner-gated repair — edit All Employees at /admin → Groups, add site WebChart, save, recording an audited SEGMENT_UPDATED — still applies there, and the live Neon stack leaves the seam off regardless.) Cases are still not created for wc| subjects by default and rerun-to-verify on any resulting wc| case returns a non-mutating 409 until fetch-one-patient lands.

  • Synthetic-corpus fidelity to the OFFICIAL artifacts (2026-07-27, ADR-038). Every eCQM code the synthetic corpus stamps is now a verified member of the official artifact's own expansion of the value set it is registered under (CANONICAL_CODE_VALUE_SETS in engine/cql/bundled-ecqm-expansions.ts, enforced by wiring/corpus-membership.test.ts). An audit against the vendored CMS122/CMS125 terminology found 12 of 24 codes belonged to a different set than the one they were registered under — e.g. SNOMED 103735009 is in "Palliative Care Intervention" but not "Palliative Care Diagnosis". No measure test could catch it: that file supplies both the stamped code AND the offline expansion the authored CQL resolves, so a wrong code is wrong in both places and every authored outcome stays exactly as seeded. Also fixed: CMS125's official IPP reads the us-core-sex extension (not Patient.gender) and its numerator retrieves [Observation: "Mammography"] (all 92 members LOINC) where the corpus emitted only a CPT Procedure — so the corpus now emits both, and Conditions carry an onsetDateTime. Authored outcomes are byte-identical (an Observation is not a Procedure; no authored CQL reads onset or the sex extension), and the official artifacts now score all five synthetic targets per measure as the corpus intends — up from cms122 4/5 and cms125 0/5. Scope: the static synthetic corpus. Real WebChart data gets neither CMS125 fix (normalizeWebChartBundle synthesizes no us-core-sex; the crosswalk maps mammography to a CPT Procedure with no LOINC), so official CMS125 over live data would still read out-of-population — tracked as a PR-9 blocker.

  • CMS125 over real WebChart data, updated 2026-07-30 (ADR-042). The us-core-sex half is now closed for the shim / dev-DB export path: both mapping sites emit the extension from patients.sex, and official CMS125 agrees with the authored measure on all 56 dev-DB subjects (52 MISSING_DATA, 4 OVERDUE). It is not closed for a live third-party WebChart FHIR server (teatea), which supplies no us-core-sex and whose roster therefore reads out-of-population — deliberate, fail-closed, and a limit to state before a tenant is onboarded. The mammography half is still open and fails in the dangerous direction: the crosswalk emits CPT 77067 / HCPCS G0202 on a Procedure, the official numerator retrieves [Observation: "Mammography"] from a value set of 92 LOINC codes only, and isDiagnosticStudyPerformed additionally requires category ~ imaging. Measured: with one crosswalk-shaped mammogram, authored says COMPLIANT and official says OVERDUE — a false non-compliance on a screened patient. A LOINC Observation alone does not fix it (no category), and with category it flips the error to authored. The remedy is dual-stamping both representations, as the synthetic corpus does. All four states are pinned in devdb-official-eval.test.ts.

  • Terminology & standards currency (2026 audit, 2026-07-08 — docs/archive/TERMINOLOGY_AUDIT_2026-07-08.md). A three-way verification (our implementation vs MIE's WebChart dev DB vs the current 2026 authorities — CMS eCQI, CDC CVX, LOINC, VSAC, AMA CPT, eCFR) confirmed everything load-bearing is correct and current: all 49 CMS catalog versions/MIPS IDs (v14 = 2026 confirmed), all OSHA CFR citations, and all runnable LOINC/CPT codes. The one defect class — vaccine CVX currency on the WebChart crosswalk — was fixed: influenza matching expanded from 141/140-only to the full active seasonal CVX set (VSAC "Influenza Vaccine" OID 2.16.840.1.113883.3.526.3.1254); the inactive Td code 139 supplemented with active 09/113/196; MMRV 94 now counts toward varicella; deleted HCPCS G0202 marked read-only. All fixes are additive to the WebChart read path (engine/ingress/webchart/terminology.ts) — synthetic outcomes are unchanged (the synthetic CQL matches urn:workwell:* codes, not CVX numbers). Inactive codes are matched on read for legacy records, never emitted. Durable follow-up: resolve flu membership from the VSAC value set via the ADR-023 resolver rather than the hardcoded active list.

  • All active CQL measures now use inline code-filter expressions on both the qualifying event (Procedure or Immunization) and the enrollment/exemption Conditions, matching the system/code stamped by SyntheticFhirBundleBuilder. This replaces the earlier exists([Condition]) / Count([Condition]) > 1 pattern that was semantically correct but not code-scoped. HAZWOPER (hazwoper.cql) and TB Screening (tb_surveillance.cql) were the last two still on the un-scoped pattern; the Fable H3 hardening fix (2026-07-03) brought them into line — a patient with unrelated Conditions no longer false-positives as enrolled/exempt on the arbitrary-bundle path (evaluateBundle/pnpm evaluate), where the synthetic per-measure bundles had masked it. A foreign-condition-scoping.test.ts golden regression guards it; the ELM was recompiled (pnpm compile-measures); synthetic outcomes are unchanged. True ValueSet token expansion (resolving urn:workwell:vs:* OIDs via the VSAC or a local expansion service) is a known evaluator limitation of the in-memory CQF path; the inline-code pattern is the stable workaround until a resolver is wired. (E3.2 / #90 update) A ValueSetResolver seam now supports real value-set expansion: the engine can run in an expansion mode (an optional resolver → a populated cql.CodeService) where a CQL value-set retrieve ([Procedure: "Audiogram Procedures"]) filters by real membership. Audiogram ships a value-set-retrieve ELM variant proven byte-equal to the inline path (cross-mode golden parity); the inline path remains the default. (ADR-023 update, 2026-07-05) The live VSAC (NLM UMLS) adapter is now real behind the port, superseding the "future drop-in" language above: a CompositeValueSetResolver routes dotted-numeric VSAC OIDs → a live VsacValueSetResolver (GET {base}/ValueSet/{oid}/$expand) and urn:workwell:* references → the local store, selected only when WORKWELL_VSAC_API_KEY is set (inert-unless-configured; engineForEnv is key-gated so the unkeyed path is byte-identical to today). The owner-run pnpm resolve-valuesets CLI imports official VSAC expansions into value_sets (source="VSAC", no DDL; DEPLOY.md). Descriptive only (ADR-008) — expansion feeds the CodeService, never Outcome Status (guarded by the audiogram cross-mode VSAC parity test). (E14 PR-3, 2026-07-05 — SHIPPED) GET /api/measures/cms122/fidelity/diff now runs a real, subject-by-subject execution outcome diff for CMS122: for each subject in the latest cms122 population run it builds the synthetic bundle, additively enriches it with real VSAC-member codes, evaluates both WorkWell's authored cms122 and an official-subset CMS122 measure fresh, and diffs — attributing each divergence to the first differing official gate (age 18–75 / qualifying visit / diabetes diagnosis / hospice / palliative / HbA1c-missing / WorkWell-side exclusion). It resolves the imported VSAC value_sets (source="VSAC") rows from the store, so no runtime VSAC key is needed (the key was only for the one-time pnpm resolve-valuesets import); when those rows are absent (e.g. local/dev), the route degrades to the unchanged PR-2 criteria-impact estimate. The official measure is a faithful official-SUBSET — measures/cms122_official.cql, using FHIR '4.0.1' in the proven value-set-retrieve style, driven by the VSAC OID value sets and compiled to committed ELM (DiabetesHbA1cPoorControlOfficialCQL-1.0.0) — not the literal multi-library QICore artifact. A compile-feasibility spike (2026-07-05) proved the literal CMS122v14 QICore CQL is un-compilable under the pinned JVM-free translator @cqframework/cql 4.0.0-beta.1 (its modelinfo loader can't resolve the cross-model FHIR.*/USCore.* type refs, so the whole QICore model fails to load) and that the runtime engine links no multi-library include graph; the literal path is to be revisited when the translator ships a stable multi-model release. The diff is descriptive only (ADR-008) — it writes nothing and never sets an Outcome Status; WorkWell's cms122 outcomes stay byte-identical (the enrichment is harness-local — it appends codings to a copy for the diff harness, it is not a change to the live fhir-bundle-builder). The GMI numerator alternative is now modeled in the official-subset CQL (2026-07-05): the numerator takes the most recent of an HbA1c (VSAC HbA1c Laboratory Test set) OR a Glucose Management Indicator (GMI, LOINC 97506-0) within the period, so a newer GMI supersedes an older HbA1c. GMI uses a direct LOINC-97506-0 code filter (the synthetic corpus ships no GMI value set, so no VSAC OID is invented) — the remaining simplification for that criterion is terminology-only (this code filter vs the official combined "Glycemic Status Assessment" VSAC set), and WorkWell's own authored cms122 still models neither GMI nor a recency window (see Fable L15 above). The SUBSET diff is CMS122-only (it executes a hand-authored cms122_official.cql), and the route enforces that: any other measure degrades literal → estimate, never literal → subset. (#258, 2026-07-09 — LITERAL tier SHIPPED; supersedes ADR-024's "revisit on a stable translator" clause — ADR-026. Generalized 2026-07-27, ADR-039.) The fidelity diff has a three-tier ladder — literal → subset → estimate — surfaced by an additive mode field in the response. The literal tier executes the actual official multi-library QICore artifact of any vendored measure (MADiE FHIR export, using QICore '6.0.0' — the exact CQL ADR-024 proved un-compilable under the pinned JS translator) via MITRE's fqm-execution over the pre-compiled ELM shipped inside the bundle's Library.content (application/elm+json) — no translation happens. Bundles are vendored with provenance under backend-ts/measures/official/<catalogId>/ (cms122 + cms125 today); value sets come from the artifact's own terminology sidecar (ADR-036) with no fallback to the resolve-valuesets VSAC import. fqm-execution is a diagnostic-only dependency, declared by packages/official-executor alone and reached through a lazy await import — never the run pipeline / ingress / worker (arch-tested by fqm-isolation.test.ts).

    Per-subject population membership (IPP/DENEX/DENEXCEP/NUMER) maps to the outcome vocabulary with population-level gate attribution, through the fail-closed officialMeasureSemantics table — a measure with no recorded reading of its numerator is unavailable for the tier rather than scored under another measure's. Since ADR-039 the tier is a genuine SHADOW of the runtime: the same measurement period (officialMeasurementPeriod), the same preparedForQiCore copy with the authored engine still seeing the original bundle, the same outcomeFromPopulations mapping, and no harness enrichment — that stays in the SUBSET tier, where manufactured divergence is the point. Memoized per measure + run-id. WorkWell's own outcomes stay byte-identical (ADR-008 guard). The subset tier is the fallback when cms122's vendored bundle is absent or its literal execution fails at runtime; the estimate remains the floor.

  • All five HEDIS wellness measures (including adult_immunization) are seeded via ensureInstanceSeeds() when WORKWELL_INSTANCE=ecqm or twh.

  • The synthetic FHIR bundles declare QI-Core conformance: each resource carries a QI-Core meta.profile canonical + the required structural elements (#92 / E3.4). Structural alignment (JVM-free), not IG/validator-validated — meta.profile is metadata, so evaluation outcomes are unchanged. See docs/STANDARDS_CONFORMANCE.md.

  • All 49 CMS eCQM catalog entries (2026 performance period) are seeded via ensureCmsEcqmCatalogSeed() for the same instance values. They are Draft-only and do not participate in CQL evaluation runs until CQL is authored and compiled. On re-seed, existing measures are looked up by CMS ID prefix (LIKE 'CMSNNNv%') and updated in-place so version bumps (e.g., v13→v14) do not create duplicate DB rows.

  • A headless CLI (pnpm evaluate --patient <bundle.json> --measure <id>, backend-ts/src/engine/cli/) evaluates one FHIR R4 patient bundle against a measure with no server or DB — the same CqlExecutionEngine the run pipeline uses. Golden regression over backend-ts/spike/synthetic asserts outcomes for all 11 measures × 4 scenarios (#72 / E2, updated for E6).

    • E12 PR-1 (#184): the same DB-less evaluation is also a library entry — evaluateBundle(bundle, measureId) (single JSON/FHIR object) + evaluateBatch(bundles, measureId) (a "bucket", per-item error isolation), from backend-ts/src/engine/ingress. A PatientDataSource port + resolveDataSource(env) make the ingress pluggable (JSON default; WebChart adapter is an inert stub until E12 PR-2). FHIR-native-first — adapters feed the unchanged engine (ADR-017).

  • Immunization forecasting (E6 / #76): GET /api/immunization/forecast?subjectId=&asOf= returns an advisory ImmunizationForecast (Td/Tdap, Influenza, Hepatitis B next-dose-due) computed by the ImmunizationForecast port (backend-ts/src/engine/immunization/immunization-forecast.ts). The simulated forecaster is the default; an ICE adapter can be activated by setting WORKWELL_IMMZ_ICE_API_KEY + WORKWELL_IMMZ_ICE_BASE_URL (inert stub until configured). The forecast is advisory only — the CQL Outcome Status remains the sole compliance authority (ADR-012).

  • A completed single-measure run can be exported as a FHIR R4 MeasureReport (summary + per-subject individual + a collection Bundle) via GET /api/runs/{runId}/measure-report — built from persisted outcomes with a proportion population model whose individual membership labels reconcile 1:1 with the summary (#89 / E3.1; ADR-031). Reported DENOM includes DENEX membership; the score is NUMER / (DENOM - DENEX). For cms122/cms125 only, binding metadata maps MISSING_DATA to out-of-population because their authored CQL uses that status for not Initial Population; OSHA and HEDIS-style measures keep it in IPP/DENOM. Value-set expansion + QRDA are separate E3 items (#90/#91).

  • A completed single-measure run can be exported as an HL7 QRDA Category III aggregate stub via GET /api/runs/{runId}/qrda?format=xml — well-formed CDA carrying the aggregate population counts + performance rate (reuses the MeasureReport countPopulations); a stub, not IG-validated (#91 / E3.3). See docs/STANDARDS_CONFORMANCE.md.

  • E7 action-evaluator order map (#77): each runnable measure has a corresponding proposed order code in backend-ts/src/order/order-catalog.ts. Codes reuse the terminology_mappings seed standards where present: audiogram → CPT 92557; tb_surveillance → CPT 86580; flu_vaccine → CVX 141; hazwoper → hazwoper-exam in urn:workwell:vs:hazwoper-exams. Measures without a seed mapping (e.g., BMI screening, hypertension, cholesterol, CMS eCQMs) use LOCAL urn:workwell:orders codes pending standard terminology alignment. GET /api/orders/proposals returns ProposedOrder records (or FHIR R4 ServiceRequest bundles) for at-risk subjects; proposals are advisory only.