Portfolio member · home repo

Lives in its own CI and deploy — not listed in toolbox /api/registry or aggregate /api/health.

Meta-Factory

Turns unstructured source materials — books, filings, spreadsheets, corpuses — into canonical, auditable reads consumable downstream.

Character

Researchers, compensation strategists, and job-taxonomy stewards ingest PDF towers, heterogeneous HR artefacts, and long-tail datasets; they need repeatable extraction pipelines with registry bookkeeping, not one-off parsers.

Problem

External. High-value institutional knowledge hides in annexes nobody hydrates into analytics warehouses; spreadsheets become the integration layer.

Internal. Analysts beg for bespoke scrapers whenever leadership funds a refresh because prior runs never hardened into contracts.

Philosophical. Signals locked in unstructured mediums stay ethically worthless if they remain non-joinable — disciplined factories unblock honest measurement downstream.

Guide

Meta-Factory divides engineering cleanly: ingestion plus Universal Asset Registry logic ships from the engine repository while the hardened REST/MCP-read host bundles console surfaces under people-analyst/meta-factory-prod. Marketing treats them as one consumer promise — unstructured in, deterministic registry-backed reads out — with repo nuance surfaced as footnotes, not fractured SKUs.

Abstract

Background. Workforce analytics depends on heterogeneous corpora spanning textbooks, refereed literature, jurisdictional filings, bespoke client comp tables, census-class public tables, and internal HRIS artefacts; leaving them unstructured walls them off from People Analytics Toolbox spokes that expect typed payloads.

Methodology. Meta-Factory encodes ingestion, extraction, normalization, Universal Asset Registry updates, plus read-boundary contracts surfaced through REST v1 payloads and MCP tools on Vercel — integration with toolbox consumers stays HTTP/MCP disciplined (no inlined vendor internals here).

Scope. Evidence factory plus asset registry servicing research (Principia) and analytical spokes downstream; toolbox responsibility ends at respecting those outbound contracts when composing.

Contribution. Makes previously inaccessible source materials observable to analytics stacks through stable IDs auditors can trace alongside MCP audits on the toolbox side.

Evidence / Provenance. The production host resolves /state with HTTP 200 for canned smoke curls; previews below omit redirect flags unlike the marketing root path that 307-hop into the shell — still unmistakably outside toolbox /api/*.

Plan

  1. 01

    Route analysts to authoritative repos

    Separate engine repos from deployed API hosts explicitly in onboarding — consumers vendor whichever manifest remains semver-frozen per release cadence there.

  2. 02

    Call read surfaces deliberately

    Respect REST/MCP gateways published by meta-factory-prod exactly like external vendors; toolbox POST discipline (PAT-11) mirrors across portfolio boundaries where writes exist.

  3. 03

    Pair with Principia and toolbox spokes truthfully

    Principia ingestion can subscribe to curated outputs once contracts land; toolbox spokes hydrate metrics only when registry rows exist — honesty chip above encodes separation from this CI.

Call to Action

Direct. Replay the packaged GET snippet against /state — HTML renders without toolbox headers.

Transitional. Inspect OpenAPI snapshots and MCP notes inside meta-factory-prod docs folders before budgeting integration resourcing.

Spoke I/O (visual language v1)

Portfolio members reuse the toolbox’s Spoke I/O visual language — composition view vs. toolbox microservice boundaries documented in PAT-PROTOCOL-A §15. Source package: @people-analytics-toolbox/spoke-illustrations.

Meta-Factory — heterogeneous raw → joinable substrateINPUTSMAIN ACTIONSOUTPUTSBooks · research corpusesunstructured ingestHRIS · comp artefactsexports · PDFsPublic datasetsBLS · O*NET · vendorsIngestExtract signalsCanonicalize + registerCanonical entities · assetsUniversal Asset RegistryRead API payloadsREST v1 surfacesMCP-read toolsmeta-factory-prod host↳ factory lanes↳ queryable downstreamCOMPOSES WITHprincipia-connectorjob-family-agentcalculustoolbox MCP

Try it now

Copy this curl — it targets https://meta-factory-prod.vercel.app on its own deploy, not People Analytics Toolbox production URLs.

meta-factory.host-state

GET

Loads the routed application state shell (HTTP 200) on the independently deployed Meta-Factory prod host.

curl -sS "https://meta-factory-prod.vercel.app/state"

Source repository · separate build

There is nothing to vendor from src/spokes/*/contracts inside this toolbox deploy — the portfolio member publishes its contracts from its home repository and ships on its own CI.

Try in Dev ConsoleOpen meta-factory engine repoOpen meta-factory-prod host repoBack to all spokes

Failure

Analysts rerun scraper science fair projects quarterly; MCP agents hallucinate inventories without registry grounding.

Success

Canonical IDs unify research ingestion, Principia bridges, SOC-aligned classifiers, and economics-ready metrics without smuggling unstructured PDFs straight into calculators.