Skip to content

11: Standards and Ecosystem Strategy

11.1 IFCX conformance posture

Every published layer is a valid IFCX document; ifclite:: is one extension namespace (manifest, tombstones). Composition without the namespace degrades gracefully except deletions, hence bake (09). Nothing in the system requires a fork of IFCX.

11.2 buildingSMART path (the actual star)

Sequence, using the existing paid IFCX panel engagement (Evandro) and the geometry-tiers collaboration (Thomas):

  1. Publish the spec set (02-05) implementation-neutral in the repo; reference implementation = IFClite
  2. Bring deletion overlays to the panel first: smallest, most obviously-missing piece, already acknowledged as future work in IFClite's own code, and a need any layered-IFCX tool will hit
  3. Bring changeset + provenance manifest second, framed as "interchange of model changes" (the MVD-vs-IDS lesson: standardize the artifact, not the workflow)
  4. Feed the derived-tier merge rule (05 ยง5.6) into the geometry-tiers branch as a consuming use case: merge correctness is a strong argument for the P/B/M separation
  5. Target: an official IFCX changeset/provenance part with IFClite as reference implementation. The position "proposer with working code and benchmarks" is the strongest one available in standards bodies

11.3 Ecosystem integrations (priority order)

Partner Integration Why
Motif Their native IFC import (current SOW) emits base layers; their edits emit layers; their UI embeds the review viewer via @ifc-lite/embed-sdk Turns the SOW relationship into platform adoption; their ex-Autodesk enterprise buyers are exactly who pays for audit
IfcOpenShell / Bonsai (Dion) A python writer emitting layer changesets; Bonsai as a desktop authoring client publishing to the registry Instantly cross-ecosystem; fits the already-discussed IFClite-as-web-viewer collaboration
buildingSMART validation svc Their checks as registry check providers Required-checks credibility
Lignum / DBL / GS1 (Hansueli, Thomas G.) Manufacturer DPP data as signed vendor layers The Porto "last mile" talk, productized
BFH Courses on the public registry; thesis topics (heuristic identity suggestions, merge UX studies) Pipeline + research credibility

11.4 Competitive positioning

  • Speckle: versions object streams; no IFCX, no layer composition, no agent provenance, no capability scoping, no required checks. They version data; this versions intent under policy
  • ThatOpen/Fragments: display-oriented, no change model
  • Autodesk/Nemetschek: will need an agent-safety answer within ~18 months; an existing open standard with a reference implementation forces adopt-or-visibly-reject
  • Moat ordering: (1) standards position, (2) working merge engine + benchmarks, (3) registry network effects. Speed on (2) buys time to land (1)

11.5 Narrative assets

Benchmark posts ("merge in under a second, in the browser"), the agent fire-safety demo video, a "No agent writes to main" manifesto post, bcftimemachine.com relaunch on the layer DAG, conference talk (next bSI summit): "Pull Requests for Buildings".