The Vault Has Two Separable Functions: Substrate and Methodology
The Vault Has Two Separable Functions: Substrate and Methodology
The triangulate Phase 0 experiment surfaced an architectural distinction that's load-bearing for how the newsroom uses the brain vault. The vault provides two separable functions, and they contribute to newsroom output in different ways. This distinction directly informs the Phase 1 design of packages/vault/ — the typed reader should expose both functions as first-class API surfaces, not just sources/wiki content.
The two functions
Substrate
Topic-specific knowledge accumulated in the vault as typed pages:
wiki/entities/<slug>.md— specific named things (people, companies, places, programs).wiki/concepts/<slug>.md— synthesized ideas, patterns, principles.wiki/mechanisms/<slug>.md— multi-step causal arguments with per-step evidence.wiki/questions/<slug>.md— testable hypotheses and research gaps.sources/andvault/sources/— immutable raw material citations chain back to.
Substrate strengthens an article when it exists. The denser the substrate on a topic, the less an article needs to lean on fresh fetches.
Methodology
Domain-general operating discipline encoded in the vault as meta-documents and structural conventions:
vault/_meta/JOURNALISTIC_STANDARDS.md— operational checklist disciplining contested-topic synthesis.vault/_meta/SOURCE_RELIABILITY.md— append-only fetchability tracker with reliability categories and mirror map.<context>/SCOPE.mdfiles — triangulation posture for politically-charged topics; in/out-of-scope topical declarations.vault/_meta/RESEARCH.md— typed page schemas (entity / concept / mechanism / question).- The parallel-checker discipline itself, encoded in triangulate-4-phase-application-protocol.
Methodology generalizes — it applies to any topic regardless of vault substrate density.
The Phase 0 evidence
Vault-carries percentages across the five Phase 0 topics, with their corresponding Q1-Q3 pass status:
| Topic | Register | Vault-carries % | Q1 (catches) | Q2 (audit log) | Q3 (send-to-friend) |
|---|---|---|---|---|---|
| 1 (AB 2624) | political-news | ~30% | ✅ 5 F1, 3 F3 | ✅ | ✅ (after v2 House-Style) |
| 2 (MN childcare freeze) | political-news | ~50% | ✅ 6 F1, 5 F3 | ✅ | ✅ single-pass |
| 3 (TSMC → Intel) | financial-thesis | ~80% | ✅ 6 F1, 5 F3 | ✅ | ✅ single-pass |
| 4 (PA Medicaid) | political-news | 0% | ✅ 6 F1, 5 F3 | ✅ | ✅ single-pass |
| 5 (Bronze Age Collapse) | serious-history | ~95% (via autoresearch) | ✅ 4 F1, 6 F3 | ✅ | ✅ single-pass |
Topic 4 is the critical data point. Zero PA-specific vault substrate. The article still cleared Q1–Q3 at the same quality as Topics 1, 2, 3, 5. From 2026-05-13-phase-0-retrospective: "The vault as a substrate repository was insufficient, but the vault as a methodology repository (triangulation posture, source-reliability tracker, journalistic-standards checklist, scope-defined editorial discipline) was load-bearing."
Why the distinction matters
The Plan's explicit pivot-point question, posed as a Phase 0 failure mode, was: "If the brain vault is too thin per topic → newsroom becomes an autoresearch wrapper." Topic 4 was the explicit test. The protocol survived because the methodology layer carried what the substrate layer didn't. From 2026-05-13-phase-0-retrospective: "The vault-as-methodology layer is what differentiates the newsroom from an autoresearch wrapper. The vault-as-substrate layer is what strengthens a given article when it exists."
This is also the architectural anchor for the user-articulated cross-domain-pattern thesis. The typed primitives (entity / concept / mechanism / question) and standards discipline are part of the methodology layer. They're domain-general. The same mechanism page format that captures [[tsmc-saturation-to-intel-anchor-stack]] captures [[lba-multi-stressor-collapse-chain]]. Density-on-substrate compounds into newsroom drafting power; density-on-methodology is what makes that compounding possible.
Implications for Phase 1 of the Plan
packages/vault/ (brain's typed reader, ~1-week build) should expose both functions as separable API surfaces:
Substrate API:
readSource(path) → SourcelistSourcesForContext(context) → Source[](wraps_lib/multi_context.py)readWikiPage(path) → WikiPage(with typed discrimination on entity / concept / mechanism / question)getLogEntries(context) → LogEntry[]
Methodology API (the addition this concept page proposes for Phase 1):
readJournalisticStandards() → StandardsChecklist(parsed fromJOURNALISTIC_STANDARDS.md)readSourceReliability() → ReliabilityTracker(parsed fromSOURCE_RELIABILITY.md; queryable by domain)readScope(context) → ScopeDeclaration(parsed from<context>/SCOPE.md; in/out topics, sourcing posture)readSchema() → SchemaDefinitions(parsed fromRESEARCH.md; typed page shapes for validation)
The newsroom's Reporter and Misinterpretation Editor agents will call the substrate API; the orchestrator and checker agents will call the methodology API. Phase 2's packages/orchestrator/ will compose them.
Evidence
- From 2026-05-13-phase-0-retrospective: "The vault has two separable functions: substrate (what we know about specific topics) and methodology (how we evaluate any topic). Topics 1-3 exercised both. Topic 4 exercises only the methodology layer. Both worked."
- From 2026-05-13-phase-0-retrospective on the Plan's pivot question: "'If the brain vault is too thin per topic → newsroom becomes an autoresearch wrapper.' Confirmed false."
- From 2026-05-13-phase-0-retrospective on Topic 4 specifically: "the vault as a methodology repository... was load-bearing... the standards-discipline catches (especially the OIG-vs-fraud F3) and the triangulation posture... added real value."
Related
- triangulate-4-phase-application-protocol — the protocol the methodology layer enforces
- transferable-f3-misinterpretation-patterns — these patterns are part of the methodology layer
- editorial-register-classes — register conventions are part of the methodology layer