Workforce Wonkery / Data method

Workforce Access & Equity Method v1.0

The production operating standard for identifying community need, workforce-system reach, service differences, outcome differences, hidden populations, and practical responses without collapsing equity into a single score. Version 1.0 was frozen after calibration in Santa Cruz and San Diego.

Decision question

Who could benefit more from workforce services, where does the evidence suggest a reach, service, or outcome gap, what local evidence may explain it, what have other communities done about the same problem, and what can a WDB reasonably test or change?

Zero manual input rule

Every Workforce Access & Equity field must be maintainable from evidence ChatGPT can retrieve from public sources, authorized connected sources, or accessible administrative/aggregate data.
WDB staff, providers, partners, and participants are not expected to enter, confirm, or maintain data for Workforce Wonkery.
Existing staff-produced plans, contracts, evaluations, emails, and reports may be used when they are already available through an authorized connected source.
Do not create workflows that require surveys, partner outreach, manual case updates, referral status entry, or one-off staff spreadsheets to keep the product current.
When a needed source is unavailable, retain U / Insufficient evidence / Unknown and place the evidence need into an automated acquisition or research queue.

Unit of analysis

Primary: Local Workforce Development Area.
Community drill-down: counties, cities/CDPs, and census tracts when estimates are reliable.
Labor-market context: linked MSA/CSA profile for occupations, wages, industries, and career opportunity.
Population case: a defined resident or participant population with a defensible denominator and evidence trail.

Six evidence lenses

Lens	Measures	Interpretation
Need	Employment, labor-force participation, poverty/low income, earnings, education, disconnection, and access barriers.	Shows who may benefit. It does not prove a workforce-system gap.
Reach	Participants by population, participants per relevant population denominator, local/state comparison, and trend.	Shows whether a population appears to be reaching the system at a lower or higher rate.
Service	Training, individualized services, supportive services, work experience, OJT/apprenticeship, and other substantive services.	Tests what happens after entry.
Outcome	Employment, median earnings, credential attainment, measurable skill gains, completion, and retention when available.	Tests whether results differ after service.
Barrier	Local plans, evaluations, community evidence, journalism, partner evidence, participant evidence, and other retrievable qualitative signals.	Helps explain the pattern while keeping causal claims appropriately limited.
Ecosystem response	Partner capabilities, observable coordination evidence, coverage gaps, comparable WDB practices, and opportunity assets.	Identifies who can contribute, what the available evidence says about coordination, and what the WDB could test or change.

Diagnostic statuses

Need signal: Strong / Moderate / Context only / Insufficient evidence.
Reach: Potential gap / Broadly comparable / Higher reach / Insufficient evidence.
Service: Potential gap / Broadly comparable / Higher access / Insufficient evidence.
Outcome: Potential gap / Broadly comparable / Stronger outcome / Insufficient evidence.
Overall finding: plain-language synthesis only. No composite equity score and no ranking of populations.

Production rule: v1.0 does not use one universal percentage threshold across all LWDAs. Santa Cruz and San Diego calibration showed that denominator fit, source availability, sample size, program eligibility, statistical uncertainty, and local context vary too much for a single representation cutoff to be defensible. Every Reach finding must state its denominator, comparison logic, evidence fit, and uncertainty.

Decision portfolio evidence-sufficiency gate

The Decision Portfolio selects cases automatically without creating a composite equity score or ranking populations. Canonical gate records live in Google Sheets WAE Evidence Gate; current Santa Cruz case states live in WAE Decision Portfolio.

Featured / Publishable: at least one defensible quantitative signal plus an independent evidence family such as Barrier, Ecosystem Response, or a credible Comparable Practice at B confidence or better, or unusually strong direct administrative system-gap evidence. A Featured Need/design case must explicitly say that a workforce-system gap is not established when Reach, Service, and Outcome remain unknown.
Monitor: a meaningful signal and actionable question exist, but evidence is not sufficient for a stronger case.
Researching: the population or barrier is decision-relevant but a key denominator, system measure, or evidence family remains unavailable.
Suppressed: privacy, small-cell, reliability, or identification risk prevents responsible publication.
System-attribution rule: “underserved,” Reach gap, Service gap, or Outcome gap may be used only when the corresponding system evidence supports it. Strong community Need alone never qualifies.
Selection rule: include up to eight cases. Include all Publishable cases first, then Monitor cases with the strongest evidence completeness and actionability, then Researching cases only where decision relevance is high and automated acquisition is active. Show important nonselected candidates and why they are outside the portfolio.
No ranking: portfolio state is an evidence/readiness state, not a ranking of which population matters most.
Zero-manual-input: every gate must be maintainable from evidence ChatGPT can retrieve. An evidence gap becomes an automated acquisition/research item, not a staff task.

Two-pass population discovery

The first pass uses workforce plans, ACS, EDD, WIOA sources, and other standard workforce evidence. Before the portfolio is frozen, a required second pass searches for populations and system signals that those sources can miss.

Hidden Within Aggregate: test whether broad categories such as limited English, low income, employed, unemployed, youth, farmworker, race/ethnicity, or geography conceal a locally evidenced subgroup with materially different barriers or response needs.
Blind-spot source families: check all locally applicable sources covering WIOA/administrative subgroup performance, homelessness and housing, tribal/sovereign workforce systems, migrant/seasonal and labor-specific systems, survivor/safety systems, public benefits, justice/child welfare/reentry, disability/behavioral health, military-connected populations, identity-specific equity evidence, youth/education special populations, immigrant/refugee/language-access and community-based evidence, and job-quality/employed-but-precarious evidence.
Required resolution: every applicable source family must resolve to a candidate, nested case, watchlist, explicit no-signal finding, or automated evidence gap. A source family may not be silently skipped.
No automatic attribution: discovery of a hidden population establishes neither WIOA under-service nor causation. Reach, Service, and Outcome remain separate tests.

Direct system-performance disparity escalation

Before final portfolio selection, search accessible administrative and workforce-performance evidence by subgroup. A material difference in Reach, Service, or Outcome must be surfaced even when the population did not first appear to have elevated community Need.

Verify subgroup counts, exact rates, suppression/small-number treatment, time period, program definition, and statistical method when available.
If reliability is incomplete, keep the finding as Monitor or Researching rather than discarding it.
Do not label a confirmed gap or causal effect until the evidence supports that language.
Direct system-performance evidence receives diagnostic attention before weaker signals based only on population size or the existence of a program.

Production calibration status

v1.0 is the production baseline. Santa Cruz and San Diego produced materially different portfolios under the same evidence rules. San Diego exposed the need for the blind-spot second pass and direct-performance escalation; the subsequent Santa Cruz rerun improved the portfolio without requiring another core rule. Future method changes should require a repeated, material exception across production LWDAs rather than a one-off difficult case.

Reach denominator and representation standard

Reach compares participant representation with the most conceptually appropriate population evidence. It does not use the total adult population as a universal denominator.
A direct penetration or representation rate is allowed only when the participant definition, age range, geography, and comparison population align closely enough to support it.
When definitions do not align, use a triangulated test: participant share, local trend, California or peer-LWDA comparison, and a clearly labeled community proxy.
Low-income WIOA status is broader than an ACS poverty measure; basic-skills status is broader than educational attainment; those community indicators are proxies rather than exact denominators.
Never reverse-engineer suppressed federal cells. Small or suppressed cases remain U / Insufficient evidence.
PIRL 808 is tied to National Farmworker Jobs Program reporting and is not treated as a complete count of farmworkers reaching the local WDB. Farmworker Reach requires alternate administrative/community evidence.

Participant-data acquisition fallback

Use the latest local WIOA participant records when they are available through an approved route. Filter by the verified WDB reporting code and aggregate before analysis.
If the public participant file cannot be retrieved through the automated route, search for a current local CalJOBS/PIRL export or other definition-compatible administrative extract.
Current DOL or EDD annual aggregate reports may validate program scale and outcomes, but they do not substitute for subgroup Reach or Service.
If neither current participant source is available, keep Reach, Service, and population-specific Outcome as U. Do not fill the gap with an old demographic report, a program-wide average, or a modeled estimate.
Internal plans, provider observations, and grant proposals may inform barrier hypotheses or action design, but they cannot substitute for participant counts, service rates, or outcomes.

Population universe

Core cases: low-income adults and working-poor households; unemployed adults; adults outside the labor force; adults without postsecondary credentials; English learners/limited-English populations; people with disabilities; opportunity youth; single parents; older workers; veterans; racial and ethnic groups; women and occupational segregation; migrant and seasonal farmworkers; justice-involved residents; people experiencing homelessness; foster youth; long-term unemployed residents; high-unemployment communities; households without vehicles; and households with limited digital access.

A case is featured only when evidence is material enough to support a useful local decision. Small cells and unstable estimates are suppressed or combined.

Core quantitative sources

U.S. Census Bureau, ACS 5-year: community conditions and small-area denominators.
California EDD: LWDA demographic profiles, local labor-force estimates, city/CDP labor-market context, and boundaries.
U.S. Department of Labor WIOA Performance Records Public Use Files: participant characteristics, services, and outcomes using WDB reporting codes.
Supplemental sources: California and local administrative/open data when definitions, geography, timeliness, and reproducibility are documented.

Qualitative evidence standard

Each qualitative record receives: geography, population, barrier/action tag, source type, publication date, event date when different, source link, direct finding, implication, and evidence confidence. One anecdote does not establish prevalence. Repeated independent evidence can strengthen an interpretation, while contradictory evidence must remain visible.

Ecosystem response standard

After Need, Reach, Service, Outcome, and Barrier evidence are assembled, the model maps the local capabilities that could address the barrier. This is a response layer, not another score.

Capability fields: reach/trust; navigation/case management; stabilization/basic needs; education/training; employer/job connection; retention/advancement; and data/system feedback.
Partner-fit fields: geography, populations/eligibility, language/accessibility, referral route, WDB relationship, evidence confidence, limitations, and the next automated source check.
No ranking: do not calculate an overall partner score or rank organizations. A strong partner for one barrier may have no role in another.
Separate direct service from system capacity: distinguish organizations that serve individuals from intermediaries, conveners, funders, or data partners.
A route is not access: listing a partner or documenting a referral route does not establish that people connected, were eligible, received the service, or benefited.
Handoff types: co-service/co-enrollment; warm handoff; specialist referral; trusted-partner referral; and system coordination.
WDB action types: deliver directly; co-enroll; warm-handoff; fund missing capacity; redesign a process; or coordinate a system-level response.
Case linkage: each featured population/barrier case should link only to partners with a defensible role in that specific response.

Canonical registry: Google Sheets tabs Workforce Ecosystem Registry, Ecosystem Case Map, and Ecosystem Coverage Gaps. Private method companion: page 4941.

Observable coordination evidence

Workforce Wonkery is not a referral platform and does not depend on staff-entered confirmation. The model uses only public information, connected internal sources, administrative data accessible to ChatGPT, and other retrievable evidence to describe how much is known about coordination among local partners.

No manual inputs: WDB staff, providers, partners, and participants do not update this evidence layer.
No referral operations: Workforce Wonkery does not receive, transmit, acknowledge, or close referrals.
Evidence stages: No observable evidence; Capability only; Route documented; Coordination process documented; Aggregate utilization available; Aggregate service/outcome evidence available.
Unknown stays unknown: absence of accessible evidence is not proof that coordination does not exist.
Source rule: every advancement must be supported by a public or authorized connected source ChatGPT can retrieve and cite or register.
Privacy: no participant-level PII, health, disability, legal, benefit, HMIS, school, or referral transaction data are stored.

Canonical protocol: Workforce Access & Equity Observable Coordination Evidence Protocol v0.2, page 4983. Canonical data: Google Sheets tabs Coordination Evidence Registry and Coordination Evidence Queue.

Coverage-gap diagnostic

Do not equate partner existence with ecosystem sufficiency. Record handoff, eligibility, capacity, coverage, geography/schedule, language/accessibility, and data gaps separately. The coverage diagnostic has no composite score and does not rank communities or organizations.

Comparable practice evidence standard

Each featured local gap triggers a targeted search for practices from other WDBs and communities that address the same population, barrier, or service problem. The goal is not to collect interesting examples. It is to identify operating features that are supported by evidence and are plausible to adapt locally.

Best practice: credible implementation and outcome evidence, a clear intervention, and a reasonably comparable problem or population.
Promising practice: relevant implementation with positive evidence, but incomplete evaluation, scale, duration, or transferability evidence.
Emerging model: a relevant operating model with real implementation but limited evidence of sustained outcomes.
Example only: useful design information without enough evidence to characterize effectiveness.

Each comparable-practice record stores: community and geography; population served; diagnosed problem; operator/host; intervention design; implementation date; partners; funding and staffing when known; quantitative outcomes; qualitative results; evaluation source; evidence classification; limitations; transferability assessment; local feature worth borrowing; and source links.

Matching rule: prefer similarity of problem, population, delivery context, and operating conditions over superficial similarity of program name. A practice is not recommended for local replication merely because it worked elsewhere.

Evidence confidence

A: strong quantitative pattern plus direct local/system evidence.
B: good quantitative evidence plus partial local explanation.
C: suggestive pattern, imperfect denominator, or limited local evidence.
D: weak, stale, unstable, or conflicting evidence.
U: evidence unavailable.

Publication gates

Denominator is conceptually appropriate and documented.
Reliability, margins of error, small cells, and suppression are handled.
Community need is not mislabeled as workforce-system failure.
Qualitative evidence matches the population, geography, and barrier.
The action is within a WDB/partner sphere of influence.
Comparable practices, when presented, address the same diagnosed problem and are labeled according to the strength of evidence and transferability.
Unresolved questions and evidence that could change the finding are disclosed.
No causal or discrimination claim is made without direct supporting evidence.

Refresh cadence

Quarterly: WIOA public-use files, EDD labor-force context, qualitative evidence, and action examples when new releases are available.
Annual: ACS denominators, geography validation, population-case recalibration, and publication-gate audit.
Event-driven: major service redesign, new AJCC/community access point, major local layoff or closure, significant grant/program launch, or credible new evidence of a barrier.

Status: Production standard. Method v1.0 has now been applied across all 45 California Local Workforce Development Areas. Future core method changes should require a repeated, material exception across production profiles rather than a one-off difficult case.