← Agentics Open Source Committee · Dashboard
v0 draft — a strawman for committee review, not a measurement. Most placements are classified from repo descriptions (see the per-row evidence tier); some are unverified and may be wrong. This is a working tool to structure the OIA intake discussion, not an official Foundation ranking of any project. Share feedback →

OIA Application Matrix · v0 · 2026-07-15

Where each application sits in the Open Intelligence Architecture stack. Click any application, or any filled cell, to drill down into the rationale, generate a gap-closing improvement prompt, and retest.
● centre of gravity · meaningful presence L3 · L8 = state-holding layers code ✓ verified in code   desc description only   desc ⚠ code differs

Click any row to expand its rationale, gap prompt, and rescan inline. Click again to collapse.

Methodology & limitations — how these placements were made

Two lenses

OIA (structural) answers where an app sits: 10 layers (L0 physical compute → L9 human interface) and 6 cross-layer spans (security, sovereignty, auditability, energy, identity, provenance). metaharness (operational) answers how well it runs: its score <repo> command reads a repo without executing it and rates harness-fit, build-likelihood, tool-safety, and cost-per-run.

What the cell values mean

ValueMeaning
● (2)Centre of gravity — a primary layer the app is fundamentally about.
· (1)Meaningful presence — the app touches this layer but it is not its core.
blank (0)Not addressed.

Scoring vocabulary carried from OIA

OIA's own Reader rates each layer COHERENT / GAP / NOT YET / N/A and flags vertical extension when one thing spans many layers. This matrix uses the footprint (● / ·) as a first pass; per-layer COHERENT/GAP maturity is a later refinement once a rubric exists.

The 12 evaluation dimensions

Structural: layer footprint · centre of gravity · vertical extension · novel-layer (L8) engagement · span coherence · layer maturity. Operational: harness-fit · build-likelihood · tool-safety · cost-per-run. Governance: openness / vendor-neutrality · qualification gate (issue tracker + docs + infographic/PRD/ADRs).

Evidence tier — description vs code

Every placement carries an evidence indicator. desc means it was classified from the repo's description/README only (the default, and most of this v0). code ✓ means the placement was checked against the actual source. desc ⚠ means a code check contradicted the description.

Worked example of why this matters: lean-agentic's description ("Lean4 formal verification") drove an L8 "strongest auditability" placement. A code check found a ~1.2MB Rust repo with a single src/ dir — no Lean4 proof system. The description oversold the layer. Without the indicator, that overstatement would sit in the matrix as fact. This is the same conformance-vs-performance discipline the committee applies elsewhere: a claim is not verified until checked at its source.

Limitations — read before citing

This is v0, classified from repo descriptions and READMEs, not code audits. Author = classifier, so it is a strawman for committee red-lining, not a measurement. Layer assignments are judgment calls at the margins (is metaharness's router L5 or L7). No composite score is offered on purpose: weighting the 12 dimensions is a committee decision, and one number would launder the judgment the committee exists to make. "Vertical extension" is descriptive, not a penalty. Cheapest next step: a second member independently re-tags, and a 1-page rubric with anchor examples per layer makes the tagging reproducible.

Sources

agenticsorg/OIA-Model docs/OIA-Model-v0.1-Digest.md · ruvnet/metaharness README + score · repo list generated 2026-07-15.