Brief / Research packs / P03 / capability-taxonomy-and-industry-join.md
P03 · Supply · Rendered from source

P03 — capability taxonomy, 17-industry join, and overlap model

Curated capability shelf

64 lines5,913 bytessha256 dce47f333294

P03 — capability taxonomy, 17-industry join, and overlap model

Lane: ACTIONIST-S1-L2-SUPPLY · Run: 2026-08-27-sprint-1-fable · Observed: 2026-08-27 Status: synthesis from audited prior evidence; denominator surveys land separately in draft-top-repos.jsonl / draft-top-companies.jsonl.

1. Capability-kind taxonomy (the shelf's row unit)

A shelf entry is a CAPABILITY KIND, supplied by one or more sources (OSS repo, commercial vendor, local asset), each with a reuse shape. This inverts the prior corpora, which were repo-rowed. 24 kinds, grouped by plane:

Data & records: spreadsheet-database (Airtable-like), notes/docs (Notion-like), file/object management, search, analytics/BI embed. Workflow & authority: case/workflow state machine, approvals, forms/surveys, e-signature, project/task management, scheduling/calendar. Communication: support desk/ticketing, messaging/notifications, email infra, chat/comments/presence. Commercial spine: CRM, billing/payments, inventory/commerce. Boundary & platform: auth/identity, client portal shell, connectors/iPaaS, document generation/PDF, LMS/learning, field operations.

2. Industry → capability join (17 industries × capability kinds)

Derived from the Phase-8 archetype map (primary/secondary per industry) + the 12-atom demand table (niche-atom-block-join §2.3). Method: an industry needs a kind if its primary or secondary archetype's data spine requires it, or its catalogue atoms name it.

Universal (needed by 17/17): auth/identity, CRM-lite (party records), messaging/notifications, forms/intake, file/document management, reporting/analytics. Near-universal (≥12/17): case/workflow state machine (11 primary+secondary per archetype map; adding atom-level workflow demand → 15/17), scheduling (10/17), client portal (12/17 via portal-secondary plus document-collection atoms), approvals (13/17 via approval_publish atom), document generation (12/17). Clustered: billing/payments (8/17), e-signature (7/17 — law, mortgage, insurance, real estate, property, construction, accounting), inventory (3/17), LMS (2/17), field ops (3/17), support desk (5/17), spreadsheet-database as end-user surface (6/17), search beyond record lookup (4/17).

3. Commodity / scarce / missing classification

Commodity (deep OSS + deep commercial supply; selection problem, not sourcing problem): CRM, admin dashboards/scaffolds, forms, auth/identity, analytics embeds, support desk, notifications, email infra, docs/notes engines, spreadsheet-database engines, connectors (3 catalogues measured: OpenConnector 1,445 providers/15,156 actions; Activepieces 761 pieces; Nango 982 providers).

Scarce (demand-rich, clean-supply-thin — confirmed by Phase-8 shelf inversion finding):

  • Case/workflow spine: closest OSS fits are copyleft (Documenso AGPL-3.0, frappe/hrms GPL-3.0); permissive candidates are scaffolds without domain state machines.
  • Client portal (untrusted external identity + scoped read + request submission): no credible permissive JS/TS candidate cleared Phase-8 gates.
  • E-signature with clean rights: Documenso is AGPL; commercial (DocuSign-class) is priced per-envelope.
  • Field operations (offline capture + evidence): no shelf row at all in Phase-8 (0 rows).
  • LMS/learning with permissive license: frappe/lms AGPL, chamilo GPL, pupilfirst NOASSERTION — all holds.

Missing (no acceptable source in any lane at current evidence):

  • Industry-specific compliance overlays (trust accounting, PHI consent gates, statutory deadline clocks) — custom-delta territory by design.
  • Tenant-safe connector connection store (OpenConnector's failed the tenancy falsifier; must be owned).
  • Host chrome: Actionist identity/settings/navigation layer — definitionally owned, P09-P11 lanes' scope, cited read-only here.

4. Repo overlap model (composite supply)

One repo may supply many kinds; shelf must decompose. Overlap classes with known exemplars from audited evidence:

Overlap classPatternExemplars (evidence: Phase-8 shelf + connector research + SISOCRM receipts)
Full-app compositeconnectors + UI + data + workflow + host chrome in one repoTwenty (CRM+pipeline+notes+API), Chatwoot (desk+inbox+automation+reports), cal.com (scheduling+billing+workflows+embeds)
Platform compositedata engine + UI + plugin systemTeable, NocoBase, AFFiNE, Appsmith-class
Engine + catalogueruntime + provider/connector catalogueActivepieces (workflow engine + 761 pieces), OpenConnector (OAuth engine + 1,445 providers)
Scaffold compositeauth + billing + admin, no domain modelopen-saas, SaaS-Boilerplate, production-saas-starter
Single-capabilityone kind, extractableinvoify (invoice gen), formbricks (forms/surveys), react-email-class

Implication: counting repos undercounts supply for commodity kinds (each composite ships 3-6 kinds) and OVERCOUNTS for scarce kinds (the case_workflow inside Chatwoot is desk-specific, not a general spine). The overlap graph in the final report quantifies this from the two survey files: nodes = repos + kinds; edges = supplies(repo, kind); stats = kinds-per-repo distribution, repos-per-kind distribution, and the scarce-kind exposure list.

5. Provenance lanes (kept distinct per dispatch)

  1. phase2_expansion_corpus — 500 repos, immutable, metadata-grade.
  2. phase8_template_shelf — 17 pinned rows, license-verified 2026-08-27.
  3. s1l2_oss_survey — this run's live-verified OSS denominator.
  4. s1l2_commercial_survey — this run's commercial denominator.
  5. local_estate — SISOCRM donors, 21st stores (8,515 union / 3,506 source-bearing), AutoSaaS, Great Library.
  6. user_provided — operator/client-supplied SaaS/repo lists; NEVER merged silently; subagent B-task locates them.
  7. unresolved_large_corpora — 1.3M / 850k-80k pointers; discovery-only even if resolved.