Methodology

The catalog of People Analytics methodology, encoded and callable as APIs.

For the people analyst who actually cares about methodology. Nineteen AI-callable services. Real math. No vendor lock-in.

Subscribe to the methodology newsletterBrowse the API quickstart

The unique angle

Systems data × survey data × behavioral science.

Most People Analytics tooling falls into one of three buckets: HRIS analytics suites (strong on systems data, weak on survey science), survey platforms (strong on response collection, integration is afterthought), and BI-on-people-data (flexible viz, no methodology opinions). The toolbox occupies the union of all three with explicit microservice boundaries none of the above can match.

Systems × survey join is a first-class operation.

analytics suites are strong on systems data and weak on survey science. Survey platforms are the opposite. BI tools are flexible visualization with no methodology opinions. The toolbox occupies the union: segmentation-studio normalizes canonical fields, data-anonymizer makes the join safe under min-N gates, calculus enriches the joined records into objects that downstream consumers compose against. The same envelope carries data from a Workday extract, a survey response, or a derived rollup — consumers don't care which.

Behavioral science is in the foundation, not bolted on.

reincarnation runs adaptive item selection driven by information-gain weighting ( a-parameters, pool-based item lifecycle, reliability tracking). preference-modeler runs stated-preference methods — , , penny allocation, paired comparison — with -balanced task generation and utility estimation via Newton-Raphson. These are real psychometrics and choice-theory algorithms, not 'pick a number from 1 to 5'.

Privacy is a service, not a setting.

data-anonymizer is a cross-cutting microservice that other spokes call. Anonymity-threshold logic is built into preference-modeler (gated aggregations return { status: "blocked" } below the floor). PII detection runs against tenant-overridable rule catalogs. Privacy is a contract obligation, not a UI checkbox.

The "behavioral science into analytics" framing isn't a tagline — it's encoded in the algorithms reincarnation and preference-modeler implement and in the methodology decisions calculus enforces (Wilson CI for proportions, t-interval for small samples, anonymity gates for low-N segments).

CONFIDENCE INTERVAL — METHOD SELECTIONproportion p̂sample size nn small ORp̂ near 0/1Wilson scoreintervaln moderateStudent's tintervaln large,p̂ mid-rangeNormal approxz-intervalAlways normal approxwhat most tools shipcalculus auto-selects — Wilson / t / normal — you don't pick wrong by default.
calculus auto-selects the confidence-interval method by sample size — most tools default to the normal approximation regardless.

Composition

Spokes compose contracts over HTTP — never shared internals.

manager-effectivenesscalls three sibling spokes over their published HTTP contracts. It never imports another spoke's core/. The boundary is the API, which is what lets any spoke split into its own deploy without breaking consumers.

COMPOSITION — HTTP CONTRACTS, NOT SHARED INTERNALSperformance-validityGET /api/spokes/...performance-calibrationGET /api/spokes/...segmentation-studioGET /api/spokes/...core/import core/ ✗ forbiddenmanager-effectivenessMEI composite + archetypeMEICompositearchetyperecalibration-auditSpokes compose contracts over HTTP. The boundary is the API — never the codebase.

Application environment

Nineteen , one toolbox.

One repo, one Vercel deploy, one Supabase project. Nineteen live logical microservices isolated by Postgres schema. Each spoke owns its algorithm, its contract version, and its audit trail.

Two transports: HTTP for engineers (reads public; writes service-key gated), and (Streamable HTTP + ) for AI agents. Every route handler is wrapped with withRouteLogger. Every MCP tool call writes a row to mcp.mcp_audit. Per-spoke health probes roll up at /api/health.

The catalog framework

A catalog-of-catalogs is growing on top.

Bloomberg's moat isn't the terminal — it's the catalogs (tickers, instruments, sectors, conventions). People Analytics is in a pre-catalog state today. The toolbox is building the catalogs that make People Analytics comparable, methodology-rigorous, AI-discoverable, and cross-customer benchmarkable. Catalog 0 (Infrastructure) is already live; Catalogs 1-4 are in the methodology roadmap.

  1. Catalog 0 — Infrastructure Catalog

    live

    The 7 toolbox spokes + 51 MCP tools + their contracts. Live and AI-discoverable today via toolbox.list_services, toolbox.list_tools_by_service, and GET /api/registry.

  2. Catalog 1 — HR Metrics Catalog

    planned

    Canonical definitions of every metric in People Analytics — turnover rate, time-to-fill, internal mobility, span of control, engagement methodology variants — with formulas, units, recommended CI methods, and citations.

  3. Catalog 2 — Segmentation Catalog

    planned

    Two halves: canonical fields (already live as the 35-field priority catalog inside segmentation-studio) and canonical segments — named, cross-customer-comparable segment definitions for tenure, function, level, geography.

  4. Catalog 3 — Analysis Catalog

    planned

    Named, versioned, methodology-defined analytical units. '12-month retention curve with confidence bands.' 'Engagement driver analysis via MNL.' This is where methodology IP lands as durable, queryable, AI-callable services.

  5. Catalog 4 — Visualization Catalog

    planned

    Named viz templates — trend-with-CI-bands, small-multiples-by-segment, cohort-survival-curve. Each entry has an input shape and a registered renderer.

Together these form the methodology catalog framework. Each catalog is registry-driven, runtime-discoverable, contract-versioned, AI-introspectable. Each entry is a stable ID + canonical definition + methodology citation.

Methodology updates, monthly.

Catalog releases, methodology notes, spoke changelogs. Plain language; technical depth on demand. No vendor noise.

Back to the home page

Prefer email? mike@peopleanalyst.com