@@ -59,6 +59,8 @@ npm run model:start
 
 Downloads pinned llama.cpp and Qwen2.5-Coder 7B Q4_K_M. Keep that terminal running.
 No paid API key needed. Create a project/task in **Agent workspace**. [Setup details](docs/LANGUAGE-LAB.md#start-on-windows).
+Use `npm run model:inspect -- --hash` to verify the installed weight file without
+starting inference. File identity and provider/runtime readiness are separate checks.
 
 ## Language lab
 
 
@@ -16,19 +16,21 @@ it starts no model, tool, task or account write.
 
 ## Evidence vocabulary
 
-- **Documented**: an exact API ID appears in the linked provider's own catalog.
+- **Documented**: an exact API ID appears in the linked provider's own catalog,
+  or a local candidate has a pinned repository revision, file and digest.
   This does not show account access, adapter behavior, or suitability for the role.
 - **Unresolved model ID**: no exact API ID or chosen local artifact has been
   verified. `cz-local-coder` on Qwen roles is the existing application's local
-  alias, not a Qwen upstream model ID.
+  alias, separate from their upstream repository/file identity.
 - **Contract tested**: `node --test tests/agent-catalog.test.mjs` checks unique,
   disabled definitions and rejects mismatched, stale, oversized, or fabricated
   output envelopes. This is deterministic host validation, not model evaluation.
 - **Configured / runtime verified**: neither applies to the new catalog. The
   existing saved local agent path is separate and retains its earlier evidence.
 
-As of 2026-10-19, 12 entries reference documented API IDs and six need exact
-model selection. All 18 remain disabled. No access to Anthropic, OpenAI, Google,
+As of 2026-10-19, 12 entries reference documented API IDs, two Qwen roles share
+one pinned local artifact, and four need exact model selection. All 18 remain
+disabled. No access to Anthropic, OpenAI, Google,
 Mistral, Moonshot, or open-model inference was tested in this mission.
 
 ## Sources and decisions
@@ -39,11 +41,27 @@ Mistral, Moonshot, or open-model inference was tested in this mission.
 | [OpenAI API models](https://developers.openai.com/api/docs/models) | Luna, Sol, and Astra IDs. Record them as documented only; no API credential or availability is inferred. |
 | [Gemini API models](https://ai.google.dev/gemini-api/docs/models) | Gemini 3.8 Flash endpoint for Explorer. The role remains disabled until adapter and evaluation work. |
 | [Mistral Small 4](https://docs.mistral.ai/models/mistral-small-4-0-26-03) | Current `mistral-small-2603` ID for Editor. Earlier `mistral-small-2503` was rejected because it appears under deprecated models. |
+| [Pinned Qwen GGUF](https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct-GGUF/blob/13fb94bfda8c8cf22497dc57b78f391a9acb426a/qwen2.5-coder-7b-instruct-q4_k_m.gguf) and [model card/license](https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct-GGUF) | Explicitly choose the application's existing Qwen2.5-Coder 7B Q4_K_M candidate for both local roles. Record the official revision, file SHA-256 and Apache-2.0 license. Reuse the established pin; do not download or substitute new weights. File identity is independent of role or runtime qualification. |
 
-Local Qwen, Llama, Gemma, and DeepSeek roles need an explicitly selected model
+Local Llama, Gemma, and DeepSeek roles still need an explicitly selected model
 artifact, license, hardware check, and qualified adapter. Kimi needs a verified
 Moonshot API ID and account access. No silent substitute is allowed.
 
+The two Qwen roles declare repository `Qwen/Qwen2.5-Coder-7B-Instruct-GGUF`,
+revision `13fb94bfda8c8cf22497dc57b78f391a9acb426a`, file
+`qwen2.5-coder-7b-instruct-q4_k_m.gguf` and digest
+`509287f78cb4d4cf6b3843734733b914b2c158e43e22a7f4bf5e963800894d3c`.
+The file is 4,683,073,536 bytes; runtime assets and KV-cache/compute overhead are
+additional. Actual RAM/VRAM fit remains unqualified. The starter's 4,096-token
+window motivates a planned 2,496-token complete-input / 1,600-token output budget
+and a 6,000-character outer input ceiling. Character count does not prove token
+fit: a later host must tokenize the entire formatted prompt before activation.
+The catalog shows declared identity only and never inspects the user's files.
+`npm run model:inspect -- --hash` performs the separate read-only byte check;
+it does not read credentials, launch the provider, or qualify either role.
+The selected candidate is versioned; changing an installer pin requires an
+explicit role/fixture review and does not silently replace this identity.
+
 ## Contracts and activation
 
 Every candidate receives a bounded task, project/run ID, source revision and
@@ -114,6 +132,10 @@ shows the source-controlled catalog in a disposable scenario fixture. The
 scenario badge is part of the deterministic example; none of the proposed roles
 ran to produce this image.
 
+The [narrow Qwen details capture](media/qwen-identity-2026-10-19.png) is an actual
+Chrome element capture of declared artifact, alias, license and planned limits.
+It reports no local file availability or runtime qualification.
+
 The [fixture-section capture](media/agent-fixture-2026-10-19.png) is a real Chrome
 element capture from Fable Architect's read-only details. It shows authored input,
 unscored future criteria and the local JSON download; it is not model output.
@@ -82,7 +82,7 @@ and external AI CLIs require a direct user request. Never merge old work implici
 - Existing Commons run 49092854-f9bd-4b7a-8d12-4ac5ea0e5f70 retains the approved,
   unmerged safe-timer change from job 811bdb98-bcbb-4b65-80b9-f19693aef6b4.
   Preserve it. 29 earlier lab experiments were verified before this release work.
-- Public source: antropos17/commit-zero at b80d561 (06:58 UTC), CI green; use clean .cache/publication and preserve histories.
+- Public source: antropos17/commit-zero at 746729d (07:12 UTC), CI green; use clean .cache/publication and preserve histories.
   README GIF is a 15.8s condensed walkthrough of retained Sequence Kit work,
   not a continuous live recording. Static fallback included; desktop/mobile and all
   three CI jobs passed. Evidence: runtime/verification/github-readme-gif-2026-10-19.json.
 
@@ -40,6 +40,11 @@ Review the actual images before copying them into source; see [distribution](DIS
 `npm run agents:fixtures` checks authored host envelopes for all disabled roles;
 `-- --profile <stable-id>` selects one. It performs no inference and reports no
 model-quality score. [Fixture and holdout scope](AGENT-CATALOG.md#reproduce-host-contracts).
+The [ecosystem decisions](ECOSYSTEM-DECISIONS.md) connect adopted/deferred tools
+and provider choices to primary sources, tests and exact activation blockers.
+`npm run model:inspect` checks declared local artifact metadata; `-- --hash`
+streams its bytes without starting a provider or reading a token. Metadata-only,
+digest identity and runtime readiness remain distinct. [Local setup](LANGUAGE-LAB.md#start-on-windows).
 Focused suites are explicitly labeled feedback, not full acceptance. Select an area
 deliberately: changed-file inference is unreliable in this dirty checkout and can
 miss integration dependencies. The full gate remains serial to bound PGlite memory.
 
@@ -94,13 +94,13 @@ announcement posted elsewhere.
 Before publishing a release, replace the candidate version with an actual tag,
 link its exact commit and CI run, re-run the documented gates, and record real
 browser/model/sandbox coverage separately. No release was created by this note.
-Current implemented-code checkpoint: public `b80d561`, [CI run 39212997304](https://github.com/antropos17/commit-zero/actions/runs/39212997304)
+Current implemented-code checkpoint: public `746729d`, [CI run 39214513591](https://github.com/antropos17/commit-zero/actions/runs/39214513591)
 passed full gate, Chromium and runtime dependency audit. Local frozen gate at
-`85cacbd` passed 69/69 Node tests, DOM/axe 2/2, build and serving with source
-hash `f5cc6a6ad8bd461de2f856741a01eabaeb3e9333e9e00d46777e870b45ceb8b5`;
-separate Chrome coverage passed 11/11. This draft and capture tooling will require
-their own final source check before publication. The checked capture set is
-already published at `98389bd`; the current startup-recovery unit is pending its gate.
+`3282f75` passed 69/69 Node tests, DOM/axe 2/2, build and serving with source
+hash `de390f9b71d4801c7726de1bf5926225e977849626817ad52a6498079848234c`;
+separate Chrome coverage passed 12/12. Verify the final candidate revision before
+creating a release. The checked capture set is already published at `98389bd`;
+the current artifact-identity addition is pending its gate.
 
 ## Show HN fact sheet for a human author
 
 
@@ -0,0 +1,56 @@
+# Ecosystem decisions from the 2026-10-19 mission
+
+Research is bounded by the local product and the no-inference mission. Sources
+below were checked or attempted on 2026-10-19; documentation does not establish installed
+capabilities or provider entitlement. The [mission log](MISSION-2026-10-19.md)
+records implementation and verification timestamps.
+
+## Reuse a declared local artifact
+
+The official [Qwen model card](https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct-GGUF)
+and [pinned file page](https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct-GGUF/blob/13fb94bfda8c8cf22497dc57b78f391a9acb426a/qwen2.5-coder-7b-instruct-q4_k_m.gguf)
+provide the repository, Apache-2.0 license, revision and digest for the already
+used Q4_K_M candidate. Decision: bind the two disabled Qwen roles explicitly to
+that candidate and provide a read-only byte inspector. Expected benefit: another
+operator can distinguish an installed file's identity from an application alias
+and runtime readiness. No weights were downloaded or replaced. Metadata/digest,
+failure and cancellation fixtures verify the host tool; model quality and
+hardware/token fit remain unqualified. [Identity and limits](AGENT-CATALOG.md).
+
+The role pin is versioned. A later installer update must review the role identity
+and fixtures explicitly; it must not silently substitute new weights into a
+previously described role.
+
+## Defer an MCP transport until tool permissions are qualified
+
+The primary [MCP architecture reference](https://modelcontextprotocol.io/docs/2026-07-28/learn/architecture)
+separates the host/client/server, protocol features and transport. The current
+documentation resolved to version 2026-07-28, so a future integration must check
+that version's discovery and authorization requirements rather than assume an
+older handshake. Decision: retain Commit Zero's existing `ai-git/1` API and
+source-tested proposed grants; add no MCP server, SDK or transport in this pass.
+Expected benefit: avoid exposing executable tools before permission, cancellation
+and receipt handling are qualified. Current host fixtures check exact context
+and supplied evidence IDs; they are not MCP compatibility tests. Activation
+requires a concrete needed tool, reviewed executable/dependency provenance,
+capability discovery, narrow grants and real denial/cancellation fixtures.
+
+## Keep unresolved provider choices explicit
+
+Opening the official [Moonshot getting-started page](https://platform.moonshot.ai/docs/guide/getting-started)
+returned “not accessible via this tool.” No exact Kimi API ID or account access
+was verified. Decision: retain Kimi Navigator as unresolved and disabled. The
+smallest next action is to obtain an accessible current primary model-ID source,
+then qualify the selected adapter and account separately. Local Llama, Gemma
+and DeepSeek entries likewise need an explicit artifact/license/hardware choice.
+No substitute model is inferred from a familiar display name.
+
+## Other adopted decisions and their evidence
+
+| Question | Decision and verification |
+| --- | --- |
+| Hosted request/response patterns | Four disconnected text contracts share host context and evidence validation. [Primary sources, offline fixtures and activation gaps](PROVIDER-DRAFTS.md). |
+| Generated-code execution | Docker daemon remains unavailable. Keep the digest/mount/resource plan source-tested and the product blocked. [Setup path and unverified properties](SANDBOX-ADAPTER.md). |
+| Collaboration failure/recovery | Seeded actors exercise actual HTTP contracts in disposable stores. [Commands, seeds and limits](SIMULATIONS.md). |
+| Browser/workflow feedback | Use real Chrome, scoped fixtures and frozen full gates; preserve the production limiter. [Comparable measurements and limits](WORKFLOW-MEASUREMENTS.md). |
+| Open-source distribution | Prepare real captures and channel-specific material, respect channel restrictions, and publish only source/docs. [Rules, drafts and assets](DISTRIBUTION.md). |