Many designs. One conviction. Feedback makes people — and A.I. — better. Data workflow is the spine.

Two convergent problems. AI coding tools' productivity claims rest on agent-side measurements — lines produced, tasks completed, time-to-PR. If the Ironies of Automation are operative (operator vigilance falling as agent reliability rises), those measurements systematically overstate net effect. There is no operator-side cockpit catching the loss. And on the portfolio side, every repo accumulates its own patterns, architecture maps, handoffs, and decisions — most of which are unreachable from any other repo. The result is rebuilding the same primitive five times across five products and not noticing.

A local dashboard + MCP server + CLI + Chrome extension that orchestrates day-to-day work inside one target project (kanban, actions, dispatch, session reports). A multi-agent kanban with a completion-block protocol that tracks per-card execution across heterogeneous AI tools. Two-phase actor handoff (builder → reviewer) where the second transition requires an artifact only the reviewer can produce. Cross-tool sync via a hub SDK so an operator coordinates Cursor, Claude Code, Replit, and other agents through one board. The continuous production telemetry that runs the C1 risk-compensation field study — a pre-registered test of Bainbridge 1983 in real coding work, with hypotheses, analysis plan, and falsification criteria specified before data accumulates. The cross-project documentation hub: pattern library (~20 production-validated patterns), architecture maps per repo, session-handoff archive, assignment registry across repos, API registry, decision log, capabilities catalog, and the Portfolio Capability Platform Playbook that governs how every product folder is structured (src/capabilities/<name>/{contracts,core,adapters,ui,tests} with extraction maturity levels 0–3). Performix is the first reference implementation of the doctrine.

• 01Two-phase actor handoff (builder → reviewer) where the second transition requires an artifact only the reviewer can produce — enforces review without trusting it
• 02Coordination-event log as a research instrument, not just an audit trail — the apparatus for the C1 risk-compensation field study
• 03Completion blocks as a protocol — every assignment ends with structured machine-readable completion, not free-text close-out
• 04Hub-and-spoke sync between heterogeneous AI tools so an operator coordinates Cursor + Claude Code + Replit + custom agents through one board
• 05Cross-project documentation hub — patterns, maps, handoffs, assignments, APIs, decisions, and capabilities legible across every repo in the portfolio. Reduces the rebuild-it-five-times tax that solo-cadence multi-product work would otherwise pay.
• 06Capability Architecture Doctrine — the portfolio-wide playbook for structuring applications as compositions of extractable capabilities with contracts, adapters, and props-driven UI. Performix is the first reference implementation; subsequent products inherit the structure.

Private. The operator-side coordination spine for the multi-app portfolio. Live locally, measuring, instrumented for the C1 field study. Productising as a multi-tenant SaaS per docs/DEVPLANE-ROADMAP.md. Portfolio consolidation plan in flight: 5 clusters + DevPlane (Fantasy Football absorbs mfl-command-center; namesake/ monorepo absorbs baby-namer family; etc.). Pattern library ships ~20 reusable patterns; capability doctrine is the architectural framework for everything built going forward.

Fantasy products are either marketplaces with shallow projections or hardcore stat tools that do not make a case. The middle — readable intelligence with a point of view — is empty. And the strategy-game adjacency is even more under-served: there is no serious head-to-head coaching game where the simulation engine respects fantasy-league context and the analytics that drove the season inform the play.

A monorepo with four converging efforts: (1) **MFL Command Center** — GM workflow app at `apps/web` (lineup, waivers, trades, draft mode, rankings) with the MFL adapter live; (2) **Gridiron Analytics API** — Python / FastAPI service with PRISM (analytics-grade ranking and decision support) and CAMS (Capability / Alignment / Motivation / Support — the portfolio framework adapted to football roles and matchups); (3) **Insight Card system** — rich composable card components (PRISM trend, CAMS alignment, market signal, more) surfaced at `/insights` (browse) and in-context across Command Center / Strategy League (drill-down); (4) **Strategy League / Football IQ** — coaching-strategy game layered on fantasy leagues with a simulation engine. Migrations through 0023 (LLM enrichment tables); insight-card system has its own vision doc + testing plan + outstanding-work inventory.

• 01Four converging efforts on one substrate — Command Center / Analytics API / Insight Cards / Strategy League. Each ships standalone value; together they compose into a cohesive product narrative.
• 02CAMS ported from people analytics to football — the same capability-alignment-motivation-support framework that drives Performix's diagnostic instrument, adapted to roles, matchups, and weekly performance. Cross-portfolio framework reuse without re-implementation.
• 03PRISM — analytics-grade ranking + decision support built into the Python/FastAPI service; not a generic projection engine. Same statistical-rigor posture the rest of the portfolio inherits.
• 04Insight Card system as the legibility surface — composable cards (PRISM trend, CAMS alignment, market signal) browse-able at `/insights` and embeddable in context. Cards are first-class products; meta-tagging + context filtering govern when each one appears.
• 05Strategy League / Football IQ — coaching-strategy game with a simulation engine layered on fantasy leagues. Unusual adjacency for an analytics product; the strategy-game audience and the analytics audience are different shapes.
• 06MFL adapter pattern — formal adapter contract for fantasy provider swaps; the platform is provider-agnostic by design even though MFL is the first integration.

Private. Four efforts in active development; Insight Cards have their own vision doc + testing plan + outstanding-work inventory tracked separately. Database migrations through 0023 (LLM enrichment tables shipped). Card surfacing decision locked per ADR 0004 (cards ship in `apps/web` at `/insights` browse + in-context drill-down; `apps/analytics` is not the canonical card surface until a new route-to-market). Games shelf + Clerk auth + Strategy League game-plan persistence shipped in April 2026.

Most baby-naming products are SEO listicles dressed up as products. They cannot tell you whether a name is currently rising or fading; they cannot attribute the rise (which film, which royal birth, which character); they cannot help two parents converge on a shared decision; and they cannot tell you why one name carries narrative weight and another does not. The decision warrants more than a top-100 list. Nothing on the market treats it that way.

Live product at namesake.baby. $24 one-time tournament unlock — no subscription — that gates the builder, single-elimination bracket UX, village voting, shower mode, and birth reveal. A wizard flow walks parents through their story, taste, and constraints; the system produces a scored, swipable list (the *Namesake score* is a deterministic composite — uniqueness × 0.30 + trend × 0.25 + meaning × 0.20 + feasibility × 0.13 + search × 0.12, percentile-normalized into a 5–95 display range against the dataset). A cultural-intelligence content layer maps name spikes to specific films, shows, and events with attribution drawn from OMDb, TMDb, Wikipedia, and AI synthesis — the definitive record of how popular culture has shaped American baby naming. Visual identity is light-mode taste-grade (Libre Baskerville + Source Sans 3, terracotta and cream palette); profile illustrations are generated through a FLUX pipeline via FAL.

• 01Decision-discipline product framing — the locked positioning is *a name chosen, not stumbled upon*. Every surface is built around helping parents converge on a single name, not browse infinitely.
• 02Tournament-bracket UX — single-elimination naming bracket, village voting, shower mode, and birth reveal as bundled features under a $24 one-time unlock. The bracket forces the choosing-not-browsing motion the positioning names.
• 03Twenty years of weekly search data plus systematic cultural-event attribution. Most naming products show annual SSA rankings; Namesake shows weekly trend velocity with named causal events (the film that landed, the character who became iconic, the royal birth) attached. No competitor has the dataset.
• 04Cultural-diffusion research apparatus — Bass diffusion, Hawkes self-exciting processes, Moran's I, Granger causality, Lieberson null model, phonetic-spillover graphs, variance decomposition. The research surface is mirrored to `/research/namesake/` and is structural substrate for the network-mediated-adoption book project: same mechanism on a cultural-name corpus that the book argues for on organizational adoption.
• 05Deterministic, percentile-normalized scoring — no LLM-on-render. The Namesake score is computed in a batch rescore script (`npm run recalculate-scores`); the LLM enrichment runs once per name, not per request. Every name's display score is reproducible and bounded; the substrate stays interpretable.
• 06Light-mode executive-tasteful aesthetic — Libre Baskerville serif, terracotta palette, FLUX-pipeline illustrations on the cooler illustration-surface token. The visual register is the audience's; the editorial discipline does not leak.

Live at https://namesake.baby. Paid product at $24 per tournament unlock — Stripe live. Research surface mirrored to peopleanalyst.com/research/namesake (cultural-diffusion methodology, Bass / Hawkes / Moran reports, predictability-ceiling finding, null-model + phonetic-spillover phases). Cultural-intelligence content strategy laid out for `/culture/{films,shows,events}` SEO surface; route currently redirects to `/names` pending hub build.

Image platforms either flatten taste into engagement metrics or hide behind gatekeepers. Editorial platforms publish on calendars rather than to readers. Neither produces a reading rhythm. Neither learns from you. Vela is built on the bet that a single substrate can do both — taste-driven figurative discovery and longform editorial work — when adaptive measurement runs underneath.

385 active works pulled from museum APIs (ARTIC, Met, BnF, Smithsonian, Europeana) under full attribution and license discipline. The Reincarnation engine learns per-reader desire and pool composition across visual rhyme and emotional register; first production batch shipped. A five-layer Written System — analytical essays, Mosaic pieces (1,353-passage testimony corpus), fiction, audio readings, and submissions — paced per-user (each reader's magazine begins when they arrive, not on a calendar). The Narrative Intelligence Platform generalizes the Mosaic engine into a three-domain spine (lived / fiction / research) with five admin surfaces deployed (Dashboard · Sources · Excerpt Review · Quote Explorer · Insight Workbench · Guide Builder). Artist Studies as the flagship editorial arc pattern — Warhol shipped (10 ASNs, Virtual Docent extension, Pittsburgh walking tour); Schiele, Klimt, and Sargent staged. Penwright lives at /labs/penwright (F-03 Authorship Packet UI MVP shipped; F-19 Adaptive Authorship Control Kernel is the spine; *They Say / I Say* method registered as the third primitive 2026-05-13). Editorial Office: Writer's Desk for 1:1 with each writer; The Office for multi-writer convening with round-2 react. Derivative Pipeline (unlicensed → Vela-licensed, with transformation-distance scoring) live as a new push. Stripe membership in live mode. Preregistered research program at /study with consent-gated session, instrument-validation cohort, and dissertation-track scaffolds for RQ1–RQ4.

• 01Reincarnation engine: per-user desire scoring with RID/SID adaptive measurement and visual-rhyme sequencing — the secret sauce, treated as scientific instrument under formal scrutiny (math spec, segmentation explorer, replay pipeline, instrument-validation cohort)
• 02Per-user magazine pacing — each reader's editorial schedule begins when they arrive; positioning wedge is "your magazine begins when you do"
• 03Five-layer Written System on a single substrate — analytical essays, Mosaic testimony, fiction, audio readings, and submissions — coexisting around three editorial axes (figurative response, emotion architecture, developmental theology); emotion is the core axis post the 2026-04-30 pivot
• 04Narrative Intelligence Platform — generalizes the Mosaic editorial engine into a three-domain spine (lived / fiction / research); five admin surfaces deployed (Dashboard, Sources, Excerpt Review, Quote Explorer, Insight Workbench, Guide Builder) for processing the 1,353-passage testimony corpus
• 05Artist Studies arcs as the flagship editorial pattern — research → essay → gallery → AI method studies → walking tour → retrospective; Warhol shipped (Virtual Docent + Pittsburgh tour); Schiele, Klimt, Sargent staged in the queue
• 06Editorial Office — Writer's Desk + multi-writer convening turns the writer roster from production tool into colleagues
• 07Adaptive Authorship substrate underneath (lib/platform/, CI-guarded against Vela-specific imports) — Vela is property #1; siblings reuse the substrate
• 08Preregistered research program at /study — consent-gated session, arm-assignment log, instrument validation, dissertation scaffolds for RQ1–RQ4. The site is the exhibit; the engine is the science.
• 09Museum-grade attribution and license discipline as a first-class feature, not a footnote

Live at vela.study. Stripe membership in live mode. Magazine publishing on per-reader pacing. Reincarnation first production batch shipped. Narrative Intelligence Platform: five admin surfaces deployed, 1,353-passage testimony corpus live, dual-write from Mosaic → Excerpts running. Artist Studies: Warhol shipped (ASN-306..315 + Virtual Docent extension); Schiele, Klimt, Sargent staged. Labs migration: seven experimental surfaces (Coincidences, Reveal, Three Rooms, Tournament/Flow Preview, Personal Composites, Constellation, Reveal Grid, Penwright) at /labs/* with explicit forward-looking framing. /study research program staged behind first-pass instrument validation. Penwright F-03 Authorship Packet UI MVP shipped; F-19 Adaptive Authorship Control Kernel is the active spine.

Most AI writing tools optimize for output fluency. They make it easier to produce something faster — and that something is often shaped by the model rather than the writer. The longer-term cost (capability erosion, voice flattening, sycophancy spirals, source attribution buried) is barely measured because the field measures what is easy to measure. The result is a generation of tools that look like assistants and act like substitutes.

An authorship environment that inverts the prompt-then-edit pattern. Writers assemble Authorship Packets — intent · structure · key ideas · relevant passages · counterpositions — before the AI is invoked. Corpus selection is explicit: writers choose which sources influence the work rather than inheriting the model's training distribution. The Adaptive Authorship Control Kernel (F-19) is the spine — central registry of skill measurement, intervention, and genre-aware behavior (memoir / nonfiction / fiction never collapsed). The Penwright Measurement Framework — six skill dimensions, six derived indices, three measurement layers, five-step learning loop, and four non-negotiable failure modes — determines whether a session made the writer better. Lives inside Vela's repo (app/labs/penwright/) for now; graduates when the design stabilizes.

• 01Authorship Packet Model — replaces freeform prompting with structured input units; the structure itself is data
• 02Corpus Control Layer — writer selects sources rather than inheriting the LLM's training distribution
• 03Adaptive Authorship Control Kernel (F-19) — central registry of measurement and intervention; genre-aware behavior forks copy + schema enums + prompts + metrics rather than collapsing them
• 04Penwright Measurement Framework — first multi-dimensional measurement system for AI-augmented writing skill development; four non-negotiable failure modes (output-only optimization · over-automation · weak measurement · ignoring genre differences) act as veto
• 05Accumulating named writing-method primitives — Borrowed Architecture (v1.0, 2026-05-12), They Say / I Say (third primitive, v1.1, 2026-05-13). Each is a registered method with structural anatomy + worked examples + anti-patterns, available to the writer as a callable move inside the kernel rather than as advice in a style guide.
• 06Anti-invention constraint — when a structural rhetorical move requires biographical material the user has not supplied, the tool refuses to render rather than confabulating
• 07Writing-craft corpus as the method-reference substrate — Layer-1 writing-craft books (Carson, Bluets, the recent Cluster A ingestion) tagged for retrieval against method-development sessions, so the kernel cites primary sources when proposing a move, not just trained patterns
• 08Has its own published research program at peopleanalyst.com/research/ai-human-interaction (12-paper Penwright Research Program across three tiers)

Early build inside Vela's repo (app/labs/penwright/). F-03 (Authorship Packet UI MVP) shipped. F-19 (Adaptive Authorship Control Kernel) is the architectural spine; it ships first or in parallel with the first feature. 19 features (F-01..F-19) sequenced across 6 implementation waves. Three named writing-method primitives registered so far (Authorship Packet structure, Borrowed Architecture, They Say / I Say); writing-craft corpus Layer-1 (Carson + Bluets + cohort) ingested as method-reference substrate. Research program at peopleanalyst.com/research/ai-human-interaction is the public-facing trajectory.

Organizations cannot improve what employees cannot safely say. Performance is not produced by capability alone; capability becomes realized performance only when alignment, motivation, and support are present. Most products miss this — they treat performance data as a transaction record (HRIS posture) or a dashboard (BI posture) or a sentiment score (engagement-survey posture). None of those instruments the question that actually matters: what is blocking team performance right now, and what is one accountable action that would unblock it? The richer signal — protected, comparative, longitudinal, distribution-aware, small-N-honest, mechanism-grounded — gets averaged into uselessness or lost in privacy theater.

MVP 1 — Protected Team Performance Diagnostic — in active build. Manager picks a team, answers ~12 protected survey statements (four CAMS dimensions × three items), and the system renders one binding-constraint card showing which dimension is starving plus safe comment themes plus recommended action plus follow-up pulse plan. Seven capabilities staged: protected-feedback (min-N + redaction primitive every other capability passes through), survey-collector-adapter, segmentation-adapter, cams-diagnostic, performance-science-library (cited findings backing every evidence pill), insight-player (the Player-Paradigm-inherited UI; the user's home), action-loop. CAMS canonical spec, subconstruct boundaries, binding-constraint rule, and code contract all live in `docs/CAMS.md`. Architecture is precompute-and-playback — metrics are calculated upstream by segment and stored as Insights / InsightCharts / Smart Lists; the player retrieves, never re-computes at render. Safe AI Insight Interpreter sits at the interpretation layer, never the headline — summarizes patterns, drafts leadership communications, flags weak evidence, respects min-N and role-based visibility. MCP consumer of the People Analytics Toolbox via the Reincarnation adapter as of 2026-05-11 (PFX-30) — typed-contract dependency, no shared substrate coupling.

• 01Protected feedback as a substrate primitive, not a privacy setting. Min-N enforcement, small-cell suppression, comment redaction, identity-risk scoring, role-based visibility, and safe aggregation all live in the gate every Insight passes through before it reaches storage. The player never has access to anything that has not been suppression-checked.
• 02CAMS as the binding-constraint diagnostic — not a dashboard of metrics. The output is one card per team naming the dimension that is currently starving the system. Capability / Alignment / Motivation / Support; whichever is lowest is the constraint; that constraint is what to act on. Capability alone never produces realized performance.
• 03Precompute-and-playback architecture, not BI. Insights are first-class records calculated upstream and stored; the player and the library retrieve. The framing is musical: lists, collections, and the now-playing experience. Where any surface starts computing metrics in response to a user request, that is a bug.
• 04Player Paradigm — the consumption discipline inherited from Vela, adapted to executive enterprise consumption (modern light mode, taste-grade typography, no dark-mode museum aesthetic). The player is the user's home; the library is the secondary discovery surface. Same paradigm as Vela; different aesthetic for the enterprise audience.
• 05Safe AI Insight Interpreter — AI is not the headline; AI is the interpretation layer that summarizes team patterns, drafts leadership communications, flags weak evidence, recommends follow-up. Hard constraints: never expose raw responses, never infer who said what, never generate individual employee scores from protected feedback, never support retaliation or unsafe targeting.
• 06MCP consumer of the People Analytics Toolbox. Performix vendors typed Zod contracts from the toolbox (Reincarnation for adaptive measurement; data-anonymizer for the suppression gate; segmentation-studio for cohort resolution; calculus for confidence intervals on small-N segments) instead of re-implementing them in-product. First-of-fleet test of the substrate-not-product positioning the toolbox is built around.
• 07Same product, three beachheads, three front-door messages. Sales-performance variance for CROs / VP Sales / RevOps; AI-transformation readiness for Chief AI Officers / COOs / CHROs; post-acquisition integration for Chief Integration Officers / PE op partners. The underlying instrument is identical; only the buying motion changes.

Early build. MVP 1 scope locked at docs/VISION.md (canonical 2026-05-10; tiebreak source: PRD V2). MCP consumer of the toolbox via PFX-30 (2026-05-11). Three beachheads on the GTM roadmap (sales-performance variance first). Solo build with Alvan on system architecture and platform engineering; Mike owns performance science, CAMS, measurement constructs, and customer-use-case validation.

Cross-cutting infrastructure — book ingestion at chapter-respecting fidelity, schema-conformant extraction of behavioral constructs, job / competency / persona generation, survey factories, the canonical-vocabulary substrate every consumer needs to compare measurements across products — gets re-implemented per consumer when there's no production-factory substrate. The cost isn't only engineering: each consumer ends up with its own slightly-drifted definition of competency, persona, engagement, effect-size, and behavioral constructs that should be comparable across products become incomparable. The earlier 'Universal Information Factory' framing tried to do too much; the resulting analytics-vs-production drift left the repo half-renovated for a stretch. The fix was a deliberate split: production-factory engine in one shell, consumer-facing API in another, with a clean seam between.

Two-repo architecture. **OLD (`meta-factory`)** — AI-controlled engine, local-only on Mike's Mac, no human UI. Owns ingestion pipelines (collector → organizer → referee → classifier 14-stage book flow; research_agent + deep_research_agent for articles), the named factory roster (Persona Factory, Survey Factory, Competency Factory, Models Factory, Requirements Factory, Prompt Factory, Publishing Factory, Business Ideas Factory, Application Designs Factory, plus checkpoint-charlie and orchestrator infrastructure), the asset registry (5,013 entries as of 2026-05-11), and canonical_outputs production. Cryptographic provenance with SHA-256 tracking on every source file; safe-delete invariants require hash verification before any local delete. Dual-grade corpus ingestion migrated in from Vela — same database holds editorially-selected curator passages and bulk research chunks, distinguished by tag. **PROD (`meta-factory-prod`)** — pure API host on Vercel. Owns the v1.2.0 REST + MCP contract (`docs/API-CONTRACT.md` + `CONTRACT-CHANGELOG.md`), the cross-portfolio library layer (Stream 7, library snapshot of ~944 records consumed by every portfolio product), consumer-onboarding doc (`CONSUMERS.md`), auth boundary (`META_FACTORY_API_SECRET` shared-secret), and cloud content access (bundled snapshot + Supabase Storage). Same engine; different shells.

• 01Split engine + API host architecture — OLD does the work, PROD makes it accessible. The seam is a snapshot + cloud-storage refresh, run manually on a manual cadence. The two repos can evolve independently without coupling consumer integration to engine internals.
• 02Named factories as units of production — Persona Factory, Survey Factory, Competency Factory, Models Factory, Requirements Factory, Prompt Factory, Publishing Factory, Business Ideas Factory, Application Designs Factory. Each is a package that outputs a structured behavioral artifact; analytics offerings explicitly excluded. The roster grows; the substrate doesn't.
• 03Cross-portfolio library layer (Stream 7) — a single read surface for ~944 corpus records that every product consumes, with a shared lifecycle and spec at `peopleanalyst-site/docs/library/SPEC.md`. The portfolio has one library, not seven slightly-different ones.
• 04Dual-grade corpus ingestion — same database holds editorially-selected curator passages and bulk research chunks, distinguished by tag. Vela's pipeline lifted into the substrate.
• 05Cryptographic provenance contract — SHA-256 tracked for every source file; safe-delete invariants require hash verification on backup and durable storage before any local delete. The substrate cannot lose source material to a careless deletion.
• 06~$0.13 per research-run synthesis at 30K+ passage scale — most 'AI research' tools run 10–100× more expensive because they retrieve without pattern extraction. The discipline is statistical: extract patterns once, cite the evidence, don't re-retrieve.
• 07Schema-extracted measurement vocabulary (@measurement/core) shared across consumers — constructs, items, instruments, effect-sizes defined once in canonical form, so behavioral measurement compares cleanly across Performix / Principia / PA Platform / the toolbox. Cross-product comparison becomes structurally possible rather than aspirational.
• 08v1.2.0 REST + MCP contract — PROD exposes the same engine over HTTP for engineers and MCP for AI agents. Consumer integration is contract-versioned with a changelog; consumers pin to the version they were built against and migrate explicitly on major bumps.

Private. OLD/PROD split shipped; v1.2.0 API contract live on Vercel at `meta-factory-prod.vercel.app`. Asset registry at 5,013 entries; cross-portfolio library snapshot at ~944 records (Stream 7, 2026-05-11). All major portfolio consumers integrated (PA-site, vela, principia, Fourth & Two, Performix, DevPlane). Operator commands (`registry:build`, `registry:verify`, `snapshot:registry`, `upload:content`) are the engine-health surface for Mike or an AI agent. The half-renovated state is behind us; the substrate is legible in thirty seconds.

HR analytics products treat behavioral data as if it were transactional. Engagement gets reduced to a survey score; performance to a rating average; retention to a churn rate; compensation to a band. The richer questions — what actually drives engagement in this organization, what signal in performance distributions matters for decisions, what statistical posture handles small-N segmentations honestly, what value would more information have before another study runs — get lost. Cross-cutting concerns (anonymization, metric calculation, segmentation, survey delivery, decision support) also get re-implemented per product, brittle and fragmented. The combination — wrong analytical posture and fragmented infrastructure — is the field's default state.

Live spokes — reincarnation (adaptive psychometric diagnostic engine; IRT-weighted item selection; pool-based item lifecycle), preference-modeler (Likert/multi-choice/free-text plus MaxDiff, conjoint, penny allocation, paired comparison; BIBD-balanced task generation; MNL utility estimation via Newton-Raphson), data-anonymizer (PII detection, deterministic HMAC tokenization, k-anonymity min-N gate, substitution-strategy registry), segmentation-studio (HRIS canonical-field normalization with 35-field priority catalog; multi-membership cohort resolution; OneModel adapter; recipes; pack publishing), calculus (statistical enrichment, anomaly detection, time-series imputation, metric × segment × period combinatorial factory; auto-selects Wilson / t-interval / normal CI), anycomp (comp models, market band math, stateless evaluation, auditable cycle runs), forecasting (Monte Carlo simulation, EVPI, discrete EVSI on aligned-chance decision trees), and Conductor (metadata-grounded SQL/Python codegen — the model sees schema, field semantics, and canonical metric definitions, so the queries it produces are construct-honest by construction) — plus a reserved namespace job-family-agent whose canonical home is meta-factory-prod. Each spoke owns its own schema, contract, and audit trail. Per-route structured logs plus per-tool audit rows in mcp.mcp_audit. Consumer apps vendor the typed contracts; the algorithms live here.

• 01Substrate, not product. Consumer apps (Performix, vela, future analytical products) vendor only the typed Zod contracts; the algorithms stay in the toolbox. Adoption is spoke-by-spoke; no all-or-nothing migration.
• 02AI-native by construction. Every spoke is callable from AI agents over MCP (Model Context Protocol) without bespoke integration. Per-consumer auth, scope-restricted keys, fire-and-forget audit log. Performix migrated to MCP transport 2026-05-11 as the first external consumer; devplane operates the wildcard key.
• 03Behavioral science in the algorithms, not bolted on. Reincarnation runs IRT a-parameter-weighted adaptive item selection with Cronbach α tracking. Preference-modeler runs MNL utility estimation on real MaxDiff/conjoint designs. Calculus auto-selects the right confidence-interval method by data shape. These are textbook psychometrics and choice theory implemented as service APIs.
• 04Conductor: metadata-grounded codegen — the model sees schema, field semantics, and canonical metric definitions, so the SQL/Python it produces is construct-honest by construction, not example-grounded. Bridges the gap between an AI writing queries and an AI writing queries that respect what the constructs are supposed to mean.
• 05Privacy is a service, not a setting. Data-anonymizer is cross-cutting — every spoke that surfaces team-level rollups calls min-N-check before responding. Anonymity-gated aggregations return blocked status below the floor; tokenization is deterministic and cache-backed; substitution strategies (mask / pseudonymize / synthetic-realistic) live in a registry.
• 06Systems × survey × behavioral-science join is first-class. Segmentation-studio normalizes HRIS canonical fields; data-anonymizer makes the join safe under k-anonymity; calculus enriches the joined records into MetricEnvelope objects. The same envelope shape carries data from a Workday extract, a survey response, or a derived rollup — consumers don't care which.
• 07Explicit contract versioning. Every spoke ships CONTRACT_VERSION; every additive change is a semver bump; every breaking change is a major bump with affected-consumer notes. Consumers vendor a copy and re-vendor on major bumps. The deploy boundary is clean.

Eight live spokes (reincarnation, preference-modeler, data-anonymizer, segmentation-studio, calculus, anycomp, forecasting, Conductor) plus one reserved namespace (job-family-agent). MCP gateway + HTTP routes shared across the spoke set; all health endpoints green; mcp.mcp_audit writing real rows; database migrations auto-run on Vercel production builds. First external consumer (Performix) migrated to MCP transport 2026-05-11. Solo build.