Creatidy/model-intelligence
5
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Agree capability identity and calibration ownership with Router and Kernel #14

New issue
Closed
opened 2026-10-04 22:19:25 +00:00 by kilo · 7 comments
kilo commented 2026-10-04 22:19:25 +00:00
Member
Copy link

Parent: #11. MI owns raw public evidence; Router owns matching/selection and its existing calibration unless owners explicitly agree otherwise; Kernel owns task requirements and adapter verification. G05 semantic boundary, G10. Source: shared architecture v1.0 (2026-10-04), supplied sections 1–5, 7–10, 12–18; MI brief §4 identity/capability/calibration/Kernel and §5. Source availability/hash in #11.

Problem and Evidence

Develop fb7299810fdc4612d4e0559465faef571662c6ef, merged #9/PR #10 model/surface evidence and A/C synthetic tests preserve context and assertion strength. This does not establish a shared MI–Router–Kernel vocabulary or prove mapping from model/alias/version/provider/channel/configuration to Router profiles. A benchmark result cannot automatically become coding=5, opaque variant cannot establish effort, and advertised surface support cannot prove local harness capability or entitlement.

Bounded Contract Research

Inspect current producer/consumer types and actual calibration/profile evidence. Deliver a small versioned mapping/ownership decision record and executable cross-boundary fixtures; do not transfer all Router calibration into MI, create a third scale, global ontology or common package for everything. If ownership remains disputed, present concrete alternatives for owner approval, not a fictional specification.

Acceptance

  1. Distinguish physical model, aliases/lineage, version, provider and serving channel, configuration parameters and versioned surface evidence. Cover same model/different channels, provider-managed unknown physical model, alias ambiguity, unknown effort vs literal none, and incompatible mapping versions. No silent cross-channel entitlement/model substitution.
  2. Preserve original benchmark metric/unit/date/method/version/parameters/provenance; fixtures demonstrate that incomparable evidence is not silently converted into a universal intelligence score or a quality profile. Public observations do not classify private tasks.
  3. Identify who produces evidence, who approves a derived mapping/calibration, who consumes/validates it and what local adapter proof is still required. Kernel requirement → Router matching using admitted MI artifact → knowledge reference in result evidence is traced without a second ranking engine. ZCode evidence/Router support does not establish Kernel ZCode runtime.
  4. Producer–consumer tests cover supported mapping, unsupported/unknown/conflicting/stale evidence, changed aliases/channels/configuration and backwards incompatibility; fail explicitly without bypassing existing admission/effort safeguards. Explain diagnostic consequences and migration of persisted mapping versions/configuration, retaining unresolved history.
  5. Return approved boundary or OWNER_DECISION_NEEDED with evidence and options; implement only the separately accepted local contract scope, not consumer product functions.

Reuse, Dependencies and Priority

Reuse actual Router profiles/calibration and Kernel requirement/adapter contracts, plus MI typed model/surface objects. Compare exact revisions, licenses where material is reused, negative guarantees, security and maintenance; don't mint another scale just for independence. Coordinate MI publication and real-source proof tasks under #11 and actual Router/Kernel counterparts, linked after discovery. Priority P1: prevents incompatible semantics and duplicate calibration before integration.
No private telemetry/tasks/credentials, task inference, live inventory/admission, harness controller, benchmark execution or permission grant. Offline fixtures need no inference; owner agreement for a contested calibration boundary and any later service/effect/inference access are separate gates.

Parent: #11. MI owns raw public evidence; Router owns matching/selection and its existing calibration unless owners explicitly agree otherwise; Kernel owns task requirements and adapter verification. G05 semantic boundary, G10. Source: shared architecture v1.0 (2026-10-04), supplied sections 1–5, 7–10, 12–18; MI brief §4 identity/capability/calibration/Kernel and §5. Source availability/hash in #11. ## Problem and Evidence Develop `fb7299810fdc4612d4e0559465faef571662c6ef`, merged #9/PR #10 model/surface evidence and A/C synthetic tests preserve context and assertion strength. This does not establish a shared MI–Router–Kernel vocabulary or prove mapping from model/alias/version/provider/channel/configuration to Router profiles. A benchmark result cannot automatically become coding=5, opaque variant cannot establish effort, and advertised surface support cannot prove local harness capability or entitlement. ## Bounded Contract Research Inspect current producer/consumer types and actual calibration/profile evidence. Deliver a small versioned mapping/ownership decision record and executable cross-boundary fixtures; do not transfer all Router calibration into MI, create a third scale, global ontology or common package for everything. If ownership remains disputed, present concrete alternatives for owner approval, not a fictional specification. ## Acceptance 1. Distinguish physical model, aliases/lineage, version, provider and serving channel, configuration parameters and versioned surface evidence. Cover same model/different channels, provider-managed unknown physical model, alias ambiguity, unknown effort vs literal none, and incompatible mapping versions. No silent cross-channel entitlement/model substitution. 2. Preserve original benchmark metric/unit/date/method/version/parameters/provenance; fixtures demonstrate that incomparable evidence is not silently converted into a universal intelligence score or a quality profile. Public observations do not classify private tasks. 3. Identify who produces evidence, who approves a derived mapping/calibration, who consumes/validates it and what local adapter proof is still required. Kernel requirement → Router matching using admitted MI artifact → knowledge reference in result evidence is traced without a second ranking engine. ZCode evidence/Router support does not establish Kernel ZCode runtime. 4. Producer–consumer tests cover supported mapping, unsupported/unknown/conflicting/stale evidence, changed aliases/channels/configuration and backwards incompatibility; fail explicitly without bypassing existing admission/effort safeguards. Explain diagnostic consequences and migration of persisted mapping versions/configuration, retaining unresolved history. 5. Return approved boundary or OWNER_DECISION_NEEDED with evidence and options; implement only the separately accepted local contract scope, not consumer product functions. ## Reuse, Dependencies and Priority Reuse actual Router profiles/calibration and Kernel requirement/adapter contracts, plus MI typed model/surface objects. Compare exact revisions, licenses where material is reused, negative guarantees, security and maintenance; don't mint another scale just for independence. Coordinate MI publication and real-source proof tasks under #11 and actual Router/Kernel counterparts, linked after discovery. Priority P1: prevents incompatible semantics and duplicate calibration before integration. No private telemetry/tasks/credentials, task inference, live inventory/admission, harness controller, benchmark execution or permission grant. Offline fixtures need no inference; owner agreement for a contested calibration boundary and any later service/effect/inference access are separate gates.
kilo commented 2026-10-04 22:22:50 +00:00
Author
Member
Copy link

Actual owners/counterparts: Router architecture plan BioMedical-IT/scarcity-router#173 and Kernel requirements plan Creatidy/creatidy-kernel#46. Reuse existing Router effort/calibration work, not relocate it automatically: BioMedical-IT/scarcity-router#57 (explicit effort/null vs none), /issues/79 (per-effort evaluation), /issues/107 (dated curated bindings), /issues/126 (restricted tracks). They cover concrete mappings, not the shared MI contract. Coordinate MI #13 publication and #12 source evidence before #15 production mapping and #19 installed acceptance. No independent third scale is approved.

Actual owners/counterparts: Router architecture plan https://forgejo.creatidy.com/BioMedical-IT/scarcity-router/issues/173 and Kernel requirements plan https://forgejo.creatidy.com/Creatidy/creatidy-kernel/issues/46. Reuse existing Router effort/calibration work, not relocate it automatically: https://forgejo.creatidy.com/BioMedical-IT/scarcity-router/issues/57 (explicit effort/null vs none), /issues/79 (per-effort evaluation), /issues/107 (dated curated bindings), /issues/126 (restricted tracks). They cover concrete mappings, not the shared MI contract. Coordinate MI #13 publication and #12 source evidence before #15 production mapping and #19 installed acceptance. No independent third scale is approved.
kilo commented 2026-10-04 22:30:36 +00:00
Author
Member
Copy link

Final actual counterparts: Router BioMedical-IT/scarcity-router#175 (rich Kernel task/harness requirements into monotone eligibility) and Kernel Creatidy/creatidy-kernel#49 (approved versioned ordinary-issue specifications). Identity/protocol boundaries: Router BioMedical-IT/scarcity-router#179 (opaque variant vs explicit effort and honest execution identity), BioMedical-IT/scarcity-router#181 (versioned harness/protocol compatibility), Kernel Creatidy/creatidy-kernel#47 (harness/gateway compatibility). These scope matches preserve requirements/unknowns and current calibration ownership; no delivered shared vocabulary is claimed.

Final actual counterparts: Router https://forgejo.creatidy.com/BioMedical-IT/scarcity-router/issues/175 (rich Kernel task/harness requirements into monotone eligibility) and Kernel https://forgejo.creatidy.com/Creatidy/creatidy-kernel/issues/49 (approved versioned ordinary-issue specifications). Identity/protocol boundaries: Router https://forgejo.creatidy.com/BioMedical-IT/scarcity-router/issues/179 (opaque variant vs explicit effort and honest execution identity), https://forgejo.creatidy.com/BioMedical-IT/scarcity-router/issues/181 (versioned harness/protocol compatibility), Kernel https://forgejo.creatidy.com/Creatidy/creatidy-kernel/issues/47 (harness/gateway compatibility). These scope matches preserve requirements/unknowns and current calibration ownership; no delivered shared vocabulary is claimed.
kilo commented 2026-10-04 22:35:16 +00:00
Author
Member
Copy link

Completeness of G05 public capability evidence: the agreed mapping must also cover public capability/limit evidence relevant to Kernel/Router requirements (for example context/output/interface limits), preserving native units, configuration/channel applicability and known/unknown/conflicting state. Current R2 Capability is only nullable boolean, not proof of this broader representation. Add fixtures where the same model has different channel/configuration limits, a requirement exceeds an evidenced limit, units are incompatible or no current limit evidence exists. Consumer rejects/records unsupported comparison without inventing a number or weakening a hard requirement. No global ontology or new public schema is pre-approved; #15 acquires only accepted lawful evidence.

Completeness of G05 public capability evidence: the agreed mapping must also cover public capability/limit evidence relevant to Kernel/Router requirements (for example context/output/interface limits), preserving native units, configuration/channel applicability and known/unknown/conflicting state. Current R2 `Capability` is only nullable boolean, not proof of this broader representation. Add fixtures where the same model has different channel/configuration limits, a requirement exceeds an evidenced limit, units are incompatible or no current limit evidence exists. Consumer rejects/records unsupported comparison without inventing a number or weakening a hard requirement. No global ontology or new public schema is pre-approved; #15 acquires only accepted lawful evidence.
kilo commented 2026-10-04 22:49:06 +00:00
Author
Member
Copy link

Review correction adds exact requirements consumer Creatidy/creatidy-kernel#51, preserving hard constraints/no L0 fallback and specification/request/decision provenance from approved intake #49. Router BioMedical-IT/scarcity-router#175 owns interpretation. Runtime route identity consumer is Kernel Creatidy/creatidy-kernel#52, while #47 remains feasibility proof and #53 conditional harness lifecycle delivery. No issue supersedes #49/#47 and no MI ownership transfer or implemented mapping is inferred.

Review correction adds exact requirements consumer https://forgejo.creatidy.com/Creatidy/creatidy-kernel/issues/51, preserving hard constraints/no L0 fallback and specification/request/decision provenance from approved intake #49. Router https://forgejo.creatidy.com/BioMedical-IT/scarcity-router/issues/175 owns interpretation. Runtime route identity consumer is Kernel https://forgejo.creatidy.com/Creatidy/creatidy-kernel/issues/52, while #47 remains feasibility proof and #53 conditional harness lifecycle delivery. No issue supersedes #49/#47 and no MI ownership transfer or implemented mapping is inferred.
kilo commented 2026-10-06 07:16:01 +00:00
Author
Member
Copy link

Selected by the refreshed full canonical queue as the remaining P1. Post-selection inspection of actual current Router 1dae1948f372e0f1739896655bb7db97d6b08460 and Kernel 9dfa20931a46931d1e97d1f0efddcf2692afb7a7 found mature bounded research seams, not a blanket prerequisite barring #14. The actual Kernel narrow reference translator still emits L0/empty minima; rich #51 preservation and installed #176 adoption are not claimed or fabricated, and #51 ordering follows semantic agreement.

Bounded research delivery: PR #31, base 2f106e8424db5b55fa4f43279a9ba138f2131baa, HEAD 2199bc2521e33c7df1ed725a310114037296a3d8. Existing-boundary versioned record plus original offline identity/native-limit/benchmark/unknown-negative fixtures; no production mapping schema, third scale, copied foreign material, calibration transfer or consumer implementation.

Sanitized offline make check PASS 102; exact-pin guarded offline consumer proof PASS 7, historical #13 regression proof PASS 16. No product GO, installed acceptance, live inference, deployment or runtime/entitlement attestation. Fresh independent whole-PR review follows; issue remains open and PR unmerged pending acceptance/currentness gates.

Selected by the refreshed full canonical queue as the remaining P1. Post-selection inspection of actual current Router `1dae1948f372e0f1739896655bb7db97d6b08460` and Kernel `9dfa20931a46931d1e97d1f0efddcf2692afb7a7` found mature bounded research seams, not a blanket prerequisite barring #14. The actual Kernel narrow reference translator still emits L0/empty minima; rich #51 preservation and installed #176 adoption are not claimed or fabricated, and #51 ordering follows semantic agreement. Bounded research delivery: PR #31, base `2f106e8424db5b55fa4f43279a9ba138f2131baa`, HEAD `2199bc2521e33c7df1ed725a310114037296a3d8`. Existing-boundary versioned record plus original offline identity/native-limit/benchmark/unknown-negative fixtures; no production mapping schema, third scale, copied foreign material, calibration transfer or consumer implementation. Sanitized offline `make check` PASS 102; exact-pin guarded offline consumer proof PASS 7, historical #13 regression proof PASS 16. No product GO, installed acceptance, live inference, deployment or runtime/entitlement attestation. Fresh independent whole-PR review follows; issue remains open and PR unmerged pending acceptance/currentness gates.
kilo commented 2026-10-06 08:04:19 +00:00
Author
Member
Copy link

Completed the accepted bounded ownership/semantic research through merged PR #31.

  • Reviewed base: 2f106e8424db5b55fa4f43279a9ba138f2131baa; approved HEAD: e5d1cfa0480748eb8336fddf91749e1424d8ead9.
  • Actual normal merge: c2d84f36cc314352d3c1ccef35257d94cbc78673, exact parents [reviewed base, approved HEAD]; merged tree equals approved tree 9ea4127c0a46f2887f04f9fd372fa5ff21e6b258; approved ancestry verified.
  • Fresh independent full-PR review ordinal 4/10, task ses_eefc98a36ffe6Q3mNnFT4jmZZe: APPROVE, findings empty. Earlier actionable fixture defects were reproduced and fixed in normal commits, preserving history/counter.
  • Integrated clean canonical develop make check: PASS107, Ruff lint/format33, basedpyright0, offline locked sync and staged/unstaged whitespace. Independent current-pin guarded proof PASS15 and historical #13 regression PASS16.
  • Four substantive files: ARCHITECTURE.md versioned boundary/ownership record; original tests/test_capability_boundary.py; exact-pin tests/verify_capability_consumers.py; reusable unchanged audit guard in tests/verify_router_consumers.py. Existing production types, R2 temporal semantics and #13 public grammar unchanged.

Boundary retained: MI public evidence/identity relationships/raw attributed capabilities; Router reviewed mapping/calibration/admission/private economics/matching; Kernel approved requirements/adapter/execution proof. Serialized identity/configuration/surface-provider changes, omitted/null effort, units, unsupported versions and missing/unknown/not-yet-acquired/stale/conflicting evidence are explicit, with distinct known-effort cuts and #13 utilized decision references. No benchmark becomes a universal rating or private requirement.

Receipt is bounded research/offline conformance, not approval of owned-capability-boundary/1 as a production schema. AC5 separately accepted production implementation remains reserved; no installed MI adoption, rich Kernel producer/translation, emitted Router MI extension, local runtime/entitlement attestation, product GO, live inference, release or deployment. Future acquisition/consumer work must independently verify its actual accepted contract, producer, rights and installed gates; issue closure does not grant those approvals.

Completed the accepted bounded ownership/semantic research through merged PR #31. - Reviewed base: `2f106e8424db5b55fa4f43279a9ba138f2131baa`; approved HEAD: `e5d1cfa0480748eb8336fddf91749e1424d8ead9`. - Actual normal merge: `c2d84f36cc314352d3c1ccef35257d94cbc78673`, exact parents `[reviewed base, approved HEAD]`; merged tree equals approved tree `9ea4127c0a46f2887f04f9fd372fa5ff21e6b258`; approved ancestry verified. - Fresh independent full-PR review ordinal 4/10, task `ses_eefc98a36ffe6Q3mNnFT4jmZZe`: APPROVE, findings empty. Earlier actionable fixture defects were reproduced and fixed in normal commits, preserving history/counter. - Integrated clean canonical develop `make check`: PASS107, Ruff lint/format33, basedpyright0, offline locked sync and staged/unstaged whitespace. Independent current-pin guarded proof PASS15 and historical #13 regression PASS16. - Four substantive files: ARCHITECTURE.md versioned boundary/ownership record; original tests/test_capability_boundary.py; exact-pin tests/verify_capability_consumers.py; reusable unchanged audit guard in tests/verify_router_consumers.py. Existing production types, R2 temporal semantics and #13 public grammar unchanged. Boundary retained: MI public evidence/identity relationships/raw attributed capabilities; Router reviewed mapping/calibration/admission/private economics/matching; Kernel approved requirements/adapter/execution proof. Serialized identity/configuration/surface-provider changes, omitted/null effort, units, unsupported versions and missing/unknown/not-yet-acquired/stale/conflicting evidence are explicit, with distinct known-effort cuts and #13 utilized decision references. No benchmark becomes a universal rating or private requirement. Receipt is bounded research/offline conformance, not approval of `owned-capability-boundary/1` as a production schema. AC5 separately accepted production implementation remains reserved; no installed MI adoption, rich Kernel producer/translation, emitted Router MI extension, local runtime/entitlement attestation, product GO, live inference, release or deployment. Future acquisition/consumer work must independently verify its actual accepted contract, producer, rights and installed gates; issue closure does not grant those approvals.
kilo closed this issue 2026-10-06 08:04:29 +00:00
kilo commented 2026-10-06 08:18:55 +00:00
Author
Member
Copy link

Bounded completion consistency follow-up PR #32 merged; #14 remains closed and was not restarted.

Exact independently approved documentation HEAD eb0067253709a8167553e8eca4988255e8811513, reviewed base c2d84f36cc314352d3c1ccef35257d94cbc78673, fresh whole-PR ordinal5/10 task ses_eefbaa30bffeirj1O200H0Umaa: APPROVE/findings empty. Actual merge a1d985e3635da30416aea898dccebd0017bfe49a; exact reviewed parents and identical approved tree cf908089851841df92e6c6c1b413a6a19b3ed9d6 verified, ancestry confirmed. Same normal checkout returned clean by fast-forward to canonical develop; integrated sanitized offline make check PASS108, Ruff lint/format33, basedpyright0, locked sync and staged/unstaged whitespace.

README/ARCHITECTURE/ROADMAP now distinguish completed bounded #14 research from unaccepted production mapping/producer, actual source/copying rights, rich Kernel requirements and installed adoption. One offline documentation regression; no production/contract/verifier/fixture behavior changes. Delivery review counter remains5/10 across PR31/32, preserving normal commits and previous histories. No product GO, installed acceptance, source acquisition, runtime/entitlement attestation, publishing/release/deployment or live inference. The loop returns to a full canonical queue; closure alone does not approve #15's actual production contract/source route.

Bounded completion consistency follow-up PR #32 merged; #14 remains closed and was not restarted. Exact independently approved documentation HEAD `eb0067253709a8167553e8eca4988255e8811513`, reviewed base `c2d84f36cc314352d3c1ccef35257d94cbc78673`, fresh whole-PR ordinal5/10 task `ses_eefbaa30bffeirj1O200H0Umaa`: APPROVE/findings empty. Actual merge `a1d985e3635da30416aea898dccebd0017bfe49a`; exact reviewed parents and identical approved tree `cf908089851841df92e6c6c1b413a6a19b3ed9d6` verified, ancestry confirmed. Same normal checkout returned clean by fast-forward to canonical develop; integrated sanitized offline `make check` PASS108, Ruff lint/format33, basedpyright0, locked sync and staged/unstaged whitespace. README/ARCHITECTURE/ROADMAP now distinguish completed bounded #14 research from unaccepted production mapping/producer, actual source/copying rights, rich Kernel requirements and installed adoption. One offline documentation regression; no production/contract/verifier/fixture behavior changes. Delivery review counter remains5/10 across PR31/32, preserving normal commits and previous histories. No product GO, installed acceptance, source acquisition, runtime/entitlement attestation, publishing/release/deployment or live inference. The loop returns to a full canonical queue; closure alone does not approve #15's actual production contract/source route.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
develop
issue-47-terminal-bench-evidence-isolated
issue-45-coding-benchmark-evidence
issue-43-zai-source-refresh
issue-19-installed-router
docs/conversational-review-policy
issue-18-public-distribution
issue-16-offline-operator-contract
issue-15-public-evidence-producer
issue-17-sharing-decision
issue-14-status-consistency
issue-14-capability-boundary
issue-29-post-selection-gates
issue-13-completion-docs
issue-13-publication-contract
issue-12-post-merge-docs
issue-12-real-source-value
workflow-loop-self-remediation
issue-9-knowledge-effective
issue-7-explicit-lineage
issue-3-temporal-proof
No results found.
Labels
Clear labels   
bug
Something is not working

  
duplicate
This issue or pull request already exists

  
enhancement
New feature

  
help wanted
Need some help

  
invalid
Something is wrong

  
question
More information is needed

  
wontfix
This won't be fixed

No labels
bug
duplicate
enhancement
help wanted
invalid
question
wontfix
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
Creatidy/model-intelligence#14
No description provided.