read-only

OKF upstream relationship — complement at the format layer, competitor at the serving layer

0025-okf-upstream-relationship

decision read as Explain confidence asserted status active 2026-06-16 owner principal-architect
Reversibility
two-way door

DEC-0025 — OKF upstream relationship: complement at the format layer, competitor at the serving layer

Reversibility: two-way door — a positioning stance, not built code; the complement-vs-competitor framing is editorial synthesis that should be revised as the upstream evolves. The harder-to-reverse, durable commitment underneath is unchanged from Adopt OKF as Dossier's canonical knowledge format: stay canonical OKF and keep Dossier a legal superset, so any portability path to/from Knowledge Catalog stays open.

Context

This session (2026-06-16) we ran a deep-research pass on the upstream home of the format Dossier adopted in Adopt OKF as Dossier's canonical knowledge format: Google Cloud's GoogleCloudPlatform/knowledge-catalog repo — the reference implementation and spec home for OKF v0.1 (authors Sam McVeety, Amir Hormati; published 2026-06-12, Apache-2.0; ~2,878 stars within days; labeled "v0.1 — Draft"). The full, primary-sourced evidence base — what the repo ships, the format's minimal/git-native/no-central-authority stance, and the Knowledge Catalog serving path — is recorded as a reference atom: OKF Upstream — GoogleCloudPlatform/knowledge-catalog — June 2026 Scan.

This decision records the positioning call the research forces: what is Dossier's relationship to its own format's upstream, now that Google both publishes the spec and ships a vendor-coupled product that serves it? It confirms DEC-0001 and does not reopen the format choice; it adds the judgment layer DEC-0001 did not need to make.

Options considered

  1. No record — treat it as routine confirmation of DEC-0001. Rejected: the upstream is now an active, fast-moving competitor-and-collaborator at once; future-you will ask "why did we frame Google as both?" — that is a decision, not a log line.
  2. Rewrite / amend DEC-0001. Rejected: DEC-0001's choice (adopt OKF as canonical) is confirmed unchanged; rewriting it would lose history and violate atomicity. The new content is a relationship/positioning judgment, which belongs in its own atom that links to DEC-0001.
  3. Record the whole thing as one reference snapshot. Rejected for atomicity: the evidence (world-facts) and the judgment (the complement-vs-competitor synthesis + the positioning opening) are independently changeable. Per the Recalibrate the Dossier brand identity — demote color, promote type + restraint + craft — then build the showcase landing / Brand Identity & Premium-Surface Research — June 2026 precedent, evidence → reference, judgment → decision.
  4. Record a layered positioning judgment + a linked evidence atom (chosen). Captures the why, keeps DEC-0001 intact, and keeps the open follow-ups out of the "decided" column.

Decision

Adopt a layer-split stance on the OKF upstream, and re-affirm DEC-0001 unchanged:

  1. Format layer = COMPLEMENT. OKF-the-format (Google Cloud, v0.1) is git-native, SDK-free, and its spec explicitly disclaims central authority / a schema registry / required tooling — "if you can cat a file you can read OKF." This is philosophically convergent with Dossier's sovereignty stance (Dossier — Mission & North Star, Adopt OKF as Dossier's canonical knowledge format), not an opposed lock-in format. Dossier stays canonical OKF.

  2. Dossier is a deliberate, LEGAL superset — not a fork. Dossier's format layer is Google's OKF spec, plus: typed frontmatter edges (owner/uses/governed_by/produces/supersedes) on OKF's deliberately untyped-link base, confidence (verified/asserted/inferred), source/provenance, and a first-class decision judgment layer (Dossier — The Knowledge Model (v0)). OKF explicitly permits this ("Producers MAY include any additional keys"), so the extensions are legal OKF, not a fork. DEC-0001 is confirmed, not changed.

  3. Serving layer = COMPETITOR. Google "updated Knowledge Catalog to ingest OKF and serve it to our agents." Knowledge Catalog = rebranded Dataplex Universal Catalog (renamed 2026-04-10; API/CLI/IAM names unchanged), serving context to agents via MCP (remote dataplex.googleapis.com/mcp + local MCP Toolbox) — the same MCP primitive MCP agentic foundation — tenant-scoped GraphRAG over the OKF KB builds on — but coupled to GCP IAM/billing and the GCP data estate (BigQuery/AlloyDB/Spanner/Looker). That vendor-coupled serving path is the lock-in vector Dossier positions against. Dossier competes here: same primitive (MCP, explainable GraphRAG), opposite coupling (the client's own git, no GCP dependency).

Rationale

  • It is the only framing the evidence supports. The split is real: the format renounces central authority and is trivially portable (complement); the serving product re-couples that same OKF to GCP IAM/billing (competitor). Collapsing the two into a single "ally" or single "rival" would be false in one direction.
  • The superset is verified legal, so portability is not at risk by construction. Because OKF permits additional keys and Dossier only adds (typed edges, confidence, provenance, decision) on top of the canonical base, a Dossier bundle is still readable as OKF — which is exactly what keeps a future interop path to/from Knowledge Catalog open (follow-up 1).
  • asserted, not verified. The underlying facts in OKF Upstream — GoogleCloudPlatform/knowledge-catalog — June 2026 Scan are primary-sourced (verified), but the complement-vs-competitor verdict is editorial synthesis, not a sourced fact — so this judgment is confidence: asserted. (Honest caveats carried in the reference atom: extreme recency — OKF v0.1 is 4 days old / Draft; some upstream framing is Google marketing — "lock-in free", "three pillars: Aggregation/Enrichment/Search"; and the "serves OKF to Vertex AI" phrasing is an overreach — Google says only "our agents": Gemini-in-BigQuery, ADK, LangChain, external Claude via MCP.)
  • This is a positioning opening, not just a threat. The same primitive (MCP) plus the opposite coupling (sovereign git vs. GCP) is precisely Dossier's differentiation. Where Knowledge Catalog's "no central authority" is aspirational (Google solely owns/hosts/is primary adopter — follow-up 3), Dossier's git-native ownership is literal.

Consequences

Open questions (unresolved follow-ups — recorded, NOT decided)

  1. Two-way interop — build & test it. Can Dossier OKF bundles round-trip losslessly INTO Knowledge Catalog (and serve via dataplex.googleapis.com/mcp)? Can Dossier ingest OKF exported FROM Knowledge Catalog? This is the concrete test of the "legal superset / portability holds" claim.
  2. Typed-edge / confidence / provenance gap — watch convergence vs. drift. Does Google's OKF add typed relationships / confidence / provenance in a future version (converging on Dossier — good), or do Dossier's extensions drift from canonical OKF and hurt portability (bad)? Track each OKF version against Dossier — The Knowledge Model (v0).
  3. Governance watch — the positioning opening. OKF's "no central authority" is aspirational: Google solely owns / hosts / is the primary adopter of the spec. Whether it moves to a neutral standards body determines whether the shared format stays a true complement or becomes soft lock-in — and that is Dossier's positioning opening.

Review

Revisit on each new OKF release and whenever Knowledge Catalog's OKF support materially changes. Re-run OKF Upstream — GoogleCloudPlatform/knowledge-catalog — June 2026 Scan's facts against current primary sources (the recency caveat is real — v0.1 was 4 days old at capture). Promote toward verified only the parts that become testable — e.g. follow-up 1 (a real lossless round-trip to/from Knowledge Catalog) — leaving the positioning verdict asserted. Re-open the stance if OKF moves to a neutral standards body (collapses some competitor framing toward complement) or if Dossier's superset diverges enough to threaten portability (re-examine the extensions).