The Great Library of SISO · Research operating brief

Foundry Agency Intelligence

Foundry is the source-intelligence engine for discovering what already exists, testing what is actually reusable, and routing proven value to the SISO Work that should own it. SISO Agency OS applies those capabilities to business outcomes. The Great Library keeps the result findable, attributable, and versioned.

Boundary: Foundry remains an independent Research Work and Git repository. The Library records its stable identity, accepted releases, selection history, and promotion candidates. Foundry does not become the system of record for client operations, and the Library does not become a warehouse for Foundry's databases or copied upstream applications.

Why the repositories stay separate

Layer	Owns	Does not own
SISO Foundry	Source discovery, repository identity, adoption evidence, capability dependencies, verification receipts, and reuse intelligence	Client business records, product authority, or every extracted capability
Great Library	Work, Release, Snapshot, Assembly, Decision, Event, and Source Inventory lineage	Operational corpora, runtime databases, or duplicate source trees
SISO Agency OS	Business identity, tenancy, capability access, approval policy, audit, events, metering, and measurable outcomes	Hidden direct writes into independently owned module databases
Source-owned modules	Deep domain behavior such as tables, documents, signing, support, scheduling, analytics, or automation	Global SISO authority or cross-module policy

The application loop

Agency need or industry problem
  → Foundry discovery and direct-source review
  → adoption mode: whole app / API / sidecar / extract / reference / build
  → behavior, integration, cost, and outcome proof
  → Great Library Source Inventory and owner assignment
  → owning Work Release and Library Snapshot
  → Agency OS capability adapter, policy, approval, and audit
  → production evidence returned to Foundry

This loop is the practical link between the Foundry mission, the SISO Library, and the Agency product. It prevents two expensive mistakes: rebuilding mature software from screenshots, and copying source into one giant repository before its ownership and operating boundary are understood.

What has been recovered

A read-only review of the external Foundry data plane observed 1,358,200 repository cards, 23,778 bank repositories, 32,137 ranked liftable candidates, 2,722 adoption decisions, 70 total contract cards with a separately identified 20-card verified subset, a 104-node and 124-edge capability graph, 707 proof selections, 62 agent patterns, 26 prompt records, and 259 repository source signals.

These are 2026-08-02 metadata observations, not current-forever or downloadable-dataset claims. The payload stays external until schema, content-hash, rights, privacy, generator, and durable-copy receipts exist. 879 adoption records remain unresolved.

The Agency portfolio

The recovered application map spans structured work, documents, project delivery, signing, support, scheduling, automation, marketing, analytics, search and AI memory, hosting and data movement, and agent builders. Operator-nominated ideas include Teable, AFFiNE, Open Notebook, Plane, Documenso, Chatwoot, Cal.com, Activepieces, Listmonk, Postiz, Formbricks, Metabase, PostHog, Superset, Meilisearch, Qdrant, RAGFlow, Coolify, Airbyte, dlt, Meltano, Langflow, Coze, and Flowgram.

Those names remain candidates. Stars and open source are nomination signals; adoption requires current revision, license, security, maintenance, tenancy, data ownership, API, deployment, and business-outcome evidence.

The God Source value matrix

Foundry 0.4.0 replaces the repo-list question with a decision unit that can survive different clients and industries: repository × SISO use case × adoption route. A repository can be a God Source for one capability, a platform primitive for another, and the wrong global application shell.

The first cross-agency tranche scores twenty applications from the recovered 9,636-repository evidence universe. It separates gross value from feasibility, then records authority group, agent operations, ownership route, evidence maturity, and the next falsifiable proof gate. Twelve entries currently clear the evidence-qualified God Source rule; none is yet outcome-proven.

Role	Current leaders	What the label means
God Sources	Activepieces, Chatwoot, Teable, Plane, AFFiNE, Open Notebook, Documenso, SurveyJS, Metabase, Coolify, Cal.com, docassemble	Complete reusable capability sources with high cross-agency value and a viable evidence-backed ownership route.
Platform primitives	Meilisearch, Qdrant, Keycloak, OpenFGA	High-leverage substrate for search, retrieval, identity, and authorization; valuable because other modules and agents rely on it.
Specialist or interface bets	OpenContracts, ERPNext, AionUI	High value for a named trigger, industry workflow, or user interface, but not a universal Agency authority.
Evidence-gated candidate	Listmonk	Strong value shape, deliberately withheld from God Source status until direct source review exists.

Coverage receipt · Foundry v0.5.1

Foundry's public v0.5.1 coverage revision (b28b0f7) publishes COVERAGE.md, coverage-inventory.json, and a verified job-level capability-pillar-map.json. The receipt covers 628 projects, 497 applications, 189 capabilities, 30 frontier rows, 114 multi-vertical projects, and 89 source-read/reusable projects. Grade counts are 10 confirmed, 79 source-read, 408 metadata, and 131 inferred.

Published: coverage inventory + 20-row decision matrix. Analyzed: 628 projects with only 89 source-read and 10 confirmed. Specified: 10 integration proofs. Runtime-proven: 1 named script.

Job-level pillar mapping is verified in v0.5.1. Pillar counts: automation 86, communication 96, data 105, deployment 42, files 49, finance 67, identity 42, knowledge 66, legal 19, marketing 45, revenue 89, work 91. This remains a coverage receipt, not a source-code reuse grant or runtime adoption authorization.

This is a coverage receipt, not a source-code reuse grant or runtime adoption authorization. The Great Library records public locators, counts, and hashes only; it does not copy Foundry databases or upstream applications. Phase 2 must produce direct-source dossiers and the ten integration proofs, then obtain an Agency owner decision.

The complete scoring model and every proof gate live in the immutable Foundry release: human-readable matrix · machine-readable matrix.

Where SISO creates leverage

Source-owned modules already contain years of domain behavior. SISO's compounding IP is the business control plane that lets humans and agents use those modules as one governed operating system:

identity, tenant, role, and authority contracts;
agent-legible capability and action schemas;
approval policy, escalation, and reversible execution;
append-only audit, provenance, events, and read models;
usage metering, pricing, and entitlements;
reliable webhooks and cross-module data contracts; and
outcome measurement that improves future Foundry decisions.

The present evidence leaves six high-value control-plane gaps without a proven universal drop-in: usage metering, pricing and entitlements, multi-tenancy, append-only business audit, reliable webhook delivery, and OAuth authority. These are hypotheses for SISO-owned Works, not permission to build before one more focused source sweep.

Evidence discipline

Existing composition experiments support a narrow claim: structured source cards and proven validators can improve steering and catch invalid hand-rolled behavior. They do not prove that arbitrary repositories automatically assemble into a product. Every promotion still needs an owner, an integration boundary, behavior tests, and an outcome receipt.

Where to continue

Review the live candidate gates on Promotion.
Open the accepted SISO Foundry Work for its current public release.
Apply the value-matrix row contract to the remaining source-read dossiers and 707 proof selections instead of launching another broad GitHub scrape.
Register the Agency OS control-plane owner before extracting product-glue capabilities from Foundry.