Generated knowledge spine · knowledge/03-EVIDENCE-MAP.md
03 · Evidence map
The high-signal source packets behind each current domain decision.
Generated from knowledge/03-EVIDENCE-MAP.md · regenerate with node site/system-map/scripts/generate-docs.mjs
Evidence and prior-work map
This map points to high-signal source packets. The generated source-inventory.jsonl accounts for the complete inventoried tree.
Core synthesis and framework
| Source | Contribution | Domains |
|---|
| Initial research synthesis | Market structure, last-mile failures, atoms, evidence gates | D01-D05, D17 |
| Original first-principles framework | Problem tree, capability inventory, block contract, lifecycle, eval and economics | D01-D18 |
| Expansion synthesis | Company, GitHub, public-signal, standards and niche joins | D01-D05, D17 |
| Phase 8 synthesis | Universal Block Foundry, local corpus, templates, mechanics and composition | D04-D17 |
| Claude lane synthesis | Connectors, Lovable, templates and UI preference science | D12-D16 |
| First-principles domain map | Current 18-domain decomposition and challenged assumptions | D01-D18 |
Demand, industries and archetypes
| Source | Measured contribution | Limits |
|---|
| Industry atom specifications | 17 industries, 12 teams, 66 use cases, 72 ideas, 12 atoms | Demand structure, not validated willingness to pay |
| Niche→atom→block join | Crosswalk from catalogue to atoms and repository baseline | Repository relevance is not capability readiness |
| Industry contradictions | Contradictions and absence codes by industry | Public evidence has denominator and access gaps |
| B2B template shelf | 10 archetypes, 17 pinned rows, 34 industry-archetype variants | Shelf is a seed, not a production template catalogue |
Source discovery and corpus
| Source | Measured contribution | Limits |
|---|
| Permanent capability source registry | Joins expanded-universe seeds, P03 candidates, recovered Foundry/operator nominations and Bykonz runtime precedents into one identity projection | Registry status is not qualification or admission; evidence classes remain separate |
| GitHub expansion report | 500 unique candidates; 17 industries; 9 capability families | Metadata/reference corpus |
| Phase 7 corpus integrity | 1,700 target rows with 270 complete and 1,430 explicit deficits | Completeness invariant was research-specific |
| Pair completion forensics | Proved the 270/1,700 denominator and evidence gaps | Did not execute or qualify repositories |
| Local corpus join | 8,515 UI identities; source/bundle/overlap distinction | Mini corpus unresolved |
| Local corpus block join | Joined local assets to block stages; zero strict Stage-0/admission assets at that snapshot | Strict gate outcome, not absence of useful code |
Capability contracts and reuse mechanics
| Source | Contribution | Limits |
|---|
| Block Contract v1 | Full governed capability dossier and synthetic fixture | Too monolithic for composition interface |
| Universal Block Framework | Packaging modes, evidence families, lifecycle and contradictions | Research design; no admitted block |
| Repo-to-block pipeline | Source shapes, stages, deterministic/human split, negative paths | Mixes extraction, qualification and admission |
| SISOCRM ownership/data decisions | First-principles service-vs-own and data ownership reasoning | One product context |
| Teable absorption brief | Concrete host-session/data/runtime seam questions | Teable-specific |
| Source | Contribution | Limits |
|---|
| Reusable-block framework report | Typed domain port and Postgres adapter guidance | Design recommendation, not production benchmark |
| Builder architecture | Control plane, data adapter, policy, runtime and deployment decomposition | Inherited scoped-Postgres assumptions need testing |
| SISOCRM ownership/data decisions | One server/separate schema and event-read-model rationale | Donor stack constrained the choice |
Connectors and external actions
| Source | Contribution | Limits |
|---|
| Connector summary | OpenConnector, Activepieces, Nango and Jentic comparison | Provider breadth does not equal uniform depth |
| OpenConnector spike | Local generation, actions, encrypted credentials and OAuth receipts | Storage/tenancy unsuitable as inherited |
UI, design and product shell
| Source | Contribution | Limits |
|---|
| Design grammar and preference protocol | Current join of validity gates, design axes, elicitation rounds, DesignDNA and focused experiments | Exact axis levels, correlations and round policy remain unexecuted |
| Local corpus join | Two complementary 21st stores and precise counts | Quality and source readiness vary |
| Claude lane synthesis | Preference learning, UI knobs and deterministic UI findings | Conversation-derived synthesis |
| AutoSaaS UI requirements | Required dashboard primitives and metadata | Early minimal registry |
| AutoSaaS method | Five-area shell and product-factory flow | Shell universality is unproven |
Composition, verification and economics
| Source | Contribution | Limits |
|---|
| Composition architecture | Plan-then-fill, deterministic solver, model roles and hypotheses | Unexecuted |
| Composition eval suite | 26 fixtures, 8 axes, 6 tiers and kill rules | Requires real capability contracts |
| Evaluation/admission plan | Seven gates, four fixtures and 20-task model matrix | Designed, not run |
| Decision ledger | Ranked backlog, stop rules and hold/kill criteria | Earlier phase assumptions |
| Source | Contribution | Limits |
|---|
| Platform deep dives | 67 initial builder surfaces plus expansion candidates | Field depth varies |
| Platform depth triage | 117 ranked records and source-depth priorities | Authenticated behavior mostly unknown |
| Strict competitor census | Complete 118×144 structural denominator | Feature-specific direct evidence extremely sparse in strict snapshot |
| Phase 9 Wave 6–12 deep dossiers | Repeated 10-surface evidence tranches | Not yet synthesized into one canonical current census |
Evidence-accounting rule
The source inventory is exhaustive by file path for configured roots. This curated map is intentionally selective by meaning. A file being inventoried does not make every claim in it current; canonical synthesis must cite the direct receipt when exact counts matter.