Consolidate catalog evidence, organic controls and foundation records

[seathatflowsinourveins]

Current FIX4 repair and checks

Named-tool exclusion now extracts original Unicode name spans and compares complete tokens, configured names and MCP components by the language's canonical caseless rule, NFD(NFD(value).casefold()). Combining marks remain inside a token; invisible format characters remain distinct. Existing punctuation, possessive and longer-name behavior is preserved.

Cross-kind command agreement now requires both the documented serialization and equal resolved shell_reads projections, with zero unknown counts on both sides. It guards RTK serialization and exact Sh/Bash/Zsh wrappers in the agreement predicate. The read parser, native ledger, raw/provisional path, typed duplicates and request/completion enrichment are unchanged.

Current validation:

• The native unittest counter module ran 71 synthetic tests, rc0, with every earlier57 test method byte-identical and14 added controls. The unchanged d99 code returned rc1 with144 failing subcases/no errors under the new14-method control run; the two nonvacuous preservation controls also ran. Actual source/test hashes, observed clock boundaries, elapsed time, logs and snapshots are in the new FIX4 receipt.
• Local nice -n 19 ionice -c3 python3 scripts/validate.py: rc0,69components/4profiles/229receipts/10403hashedfiles. Retained stdout SHA256 5881c181176dc889ec03041c47e3a5fcb322988f968d29e9614c419093ed1447; empty stderr. Diffcheck rc0.
• Independent same-family GPT source-class review:0unresolved findings within both requested classes, source inspection only; no reviewer test/network/native trial. Record SHA256 9eba6422bf1e81a4b7f5ef85107d03ef8bdaabbc6ea289b1a64f3d04e1881f32.
• Hosted validate has not run: the PR conflicts with main and has no merge ref. Local consistency/synthetic checks remain separate from hosted CI, organic use and upstream acceptance.

Publication: source/receipt commit 8f45700353a638a8d6183e0df4e9f72f2501deff, followed by registry-only 1068cd644141bb0e89e230c7328ea9d0802ea470. This registry-last chronology is stated with commit objects; a tree snapshot alone does not prove it. One push, no rebase. Prior failed observations/receipts and decided text are preserved; the decision has a dated amendment.

Canonical Unicode reference: python/cpython@v3.12.0 and @v3.13.0:Doc/howto/unicode.rst:465-480, same verified byte source SHA256 c5cf5081618e2b06a519d2ede00fb52a53922c5a8d636c4be337810fed2bd17c; Python 3.13 category API. The installed runtime's Unicode database15.1.0 is retained separately from latest-doc metadata.

Landing path remains: co-op exact-head delta, CC ACK, required CI/refutation at the single relayed landing rebase, and5f's retained-catalog queue. This PR stays DRAFT until its read. The slot-defaults/N1 recommendation, terminal-role amendment and canonical origin-URL extension are a stacked follow-up; none enters this repair tree. No deployment, installation, model/provider call, native-data counter scan, order or broker operation.

Retained implementation, fold conservation and historical check records

The records below retain their dated scope; FIX3 and earlier counter checks are historical. The current result above supersedes their repair-status wording.

Scope

Consolidate the finalized-catalog scaffold and three owned foundation drafts into one reviewable #820. The catalog remains pending_synthesis: 79 role slots, no selected repositories and a null census. The typed contract joins official release metadata, component pins, registered evidence, native wiring/fresh smoke/daily reports, four family/role refutations and dated exclusions; none of these declarations promotes a local check into adoption.

Folded originals and frozen source heads:

• Preserve organic evidence and version saturation requirement bindings #750 6cd4f403ff9c6dae18cfd560385d508ca8922ea6: organic evidence carrier/guards, additive requirement binding and Codex provenance-control fixtures. The original unchanged prototype/test pin 7d20f02732f85ede3521f324efd4a11c5d250413 remains reachable on its preserved branch.
• Carry portable NativeStack2604 Terminal profiles and source receipt #768 3a02630adeb1d75f14bb3856d14a2477220d6fe4: portable Windows Terminal fragment, explicit GUIDs, profile policy checker and exact historical receipt.
• docs: record NativeStack2604 Dagu loopback operator access #787 dff1c03ab3059b1e00710ee12d166e8967519277: dated Dagu loopback/auth-none record; its owner classification retains all eight DAGs as non-paper-lane.

Base retained head: 6eb5ed83914f37c3f0baca40cba89c8172cfa9e3. Conserved fold snapshot: 77d35e41911c0f8c28a5561790cd8d8a8cd68175. Repair head: dee7cae97c74a6657cb40c7bf8324ccbd41f9111. Repair source commits: 5530276d1df426db36ce2ddbd0f92e59105c0f1e and 0fa1b5cf4763f0f432b4ee79815c5c3b6dff37f4; the last commit dee7cae97c74a6657cb40c7bf8324ccbd41f9111 changes only manifests/evidence.json, verified locally by Git history. This author report is separate from the tree-only reader's observations. Lane exactly lane:foundation; DRAFT. No rebase for FIX2. The author records the prior fold's source commit 80d3c42244d872a04897d693083258cbd4c0131a followed by registry-only 77d35e41911c0f8c28a5561790cd8d8a8cd68175; the source-only tree review did not independently establish this commit chronology.

Landing path: the co-op verified fold conservation and closed #750/#768/#787 with records under review-ns2604-coop-20261007T023012Z section 3; their origin branches stay preserved. The retained #820 now needs command-center exact-head delta/read and ACK, CI/refutation, then 5f's foundation queue. Root remains writer until queued. One atomic Git push published the retained and preserved #750 refs; no merge or original closure performed.

The co-op's byte check and source-only GPT review confirm conservation of all three folds; the co-op has closed the originals with records (not merged), as recorded in pr-triage-closures-20261007.md at 2026-10-07T06:59:45Z, 06:59:50Z and 06:59:55Z respectively. FIX2 repairs named-tool punctuation, contradictory completed-command attribution and inherited original-judgment provenance. New fixture results are recorded separately in evidence/receipts/ns2604-catalog-fix2-20261007.json and its immutable-linked alias supplement evidence/receipts/ns2604-catalog-fix2-alias-20261007.json. The first snapshot remains reachable at 5530276d; subsequent linked-ID and sibling-alias counterexamples are retained with their actual outputs and source hashes. Historical receipts and decided text are preserved.

SOTA sources

• native-agent-stack@77d35e41911c0f8c28a5561790cd8d8a8cd68175:tools/invocation-monitoring/codex_counter.py:299-307,405-425,577-657, scripts/validate_foundation.py:123-144,702-753,877, scripts/saturation_ledger.py:439-495 and docs/acceptance-evidence-policy.md:26-33,42-69: the reviewed original controls, inherited provenance and evidence-class boundaries.

• Python 3.12 regular-expression reference: literal escaping and prose boundary assertions used by the counter repair.

• native-agent-stack@6eb5ed83914f37c3f0baca40cba89c8172cfa9e3:scripts/validate_foundation.py:79-196,326-340,457-611; catalogs/foundation/finalized-sota-catalog.json; docs/decisions/2026-10-07-finalized-sota-catalog.md: existing typed catalog/adoption contract.

• The three source heads above bind the carried files. Their unchanged receipts and decisions retain exact upstream pins and evidence classes.

• native-agent-stack@a40a083172af588f4b97646db87dfc8ef3c0b60e:tools/skill-usage/skill_usage.py:1314-1655 and examples/claude-native/workflows/child-usage.mjs:2854-2878: reuse the native call ledger.

• openai/codex@d27764b82f7118f674371e6d6e76271d9d606edb:codex-rs/protocol/src/models.rs:1061-1155; codex-rs/protocol/src/protocol.rs:1948-1960,2179-2184,3301-3307; codex-rs/history/src/rollout_payload.rs:30-68; codex-rs/rollout/src/recorder.rs:2061-2092: verified native identity/turn/wire source for rust-v0.160.1, published2026-10-05T18:29:37Z. Full source references are in the new contract artifact.

• nativestack-practice@a7b8018524575cabbd979e6f2e1c84ee6e94700f:checks/check-terminal-profiles.py; microsoft/terminal@v1.25.2733.0(dc8ae365847d41f62d1ebca93815bb00959bc12f), Profile::_GenerateGuidForProfile; Microsoft fragment extensions, dated2025-11-12/checked2026-10-06.

• Native registry/generators at retained base: scripts/host_receipts.py, scripts/component_matrix.py, docs/lanes.md:94-128. Merge and history dispositions are registered in evidence/receipts/grand-catalog-folds-20261007.json.

Exact overlap dispositions

Path	Merge
scripts/validate.py	Native three-way merge keeps organic receipt validation, narrow profile.guid scanning and #820's finalized selection guard.
component-evidence-matrix.json / its Markdown	Existing builder regenerates the union; existing rows/observations remain and organic information flips no status.
manifests/evidence.json	Retain all219 prior receipt bytes/metadata; union five source-owned receipts and the fold record; actual checksums committed last.
2026-10-05-omniroute-gateway-composition.md	Explicit A34 exception: original decided prefix and line47 unchanged; exact #768 F1 correction appended as dated amendment atline129. This file intentionally differs from #768. Its source branch and historical receipt are untouched.
tests/test_finalized_foundation_catalog.py	Explicitly mock missing owner and separately reject outside-owner references now that #750's helper is present. Production guards unchanged.

Byte-identical carried content (registry excluded): #750 26/29, #768 8/10, #787 7/7. Every file/hash/mode is listed in the fold receipt. Other differences are the documented merges above.

Open item: version-2 producer handoff

The additive requirement_binding is complete and carried here. Convergence-practice owns #821's producer paths:

• tools/sota-convergence/landscape-sweep/build_inputs.py
• tools/sota-convergence/landscape-sweep/convert.py
• tools/sota-convergence/landscape-sweep/build_args.py

Open before any future text-bound qualification: carry the same frozen snapshot with binding_version:2, identity:{catalog,layer_id}, exact requirement_text, requirement_hash and retained legacy_hash. Include it in original/expanded field digests when present; preserve the exact absent-snapshot legacy formula. Grand-catalog changes none of those producer files and claims no activation, new sweep or historical rehash. The original five ledger records/95 scopes remain byte-preserved.

FIX3 repair: case and completed command representations

Two remaining counter defects are repaired: configured and extracted prompt names now match case-insensitively, and completed command originals preserve argv/shell representation semantics. Every completed pair must agree within the existing native identity groups; a shell-text record cannot bridge conflicting argv records. Supported enrichment, exact duplicates and narrowly documented wrapper/RTK serialization remain valid.

• Synthetic controls: 57 tests pass, including all 44 earlier methods byte-for-byte unchanged. The same corrected rejection controls fail at dee7cae97c74a6657cb40c7bf8324ccbd41f9111: 11 methods, 108 failing subcases. Earlier red stages and the first failed repaired run (55 tests, 4 fixture-declaration failures) remain distinct in evidence/receipts/ns2604-catalog-fix3-20261007.json.
• Source preservation: prior receipts, original legacy scan/prompt-name extraction, maintained native adapter/ledger and decided text are unchanged. The decision is extended with an appended dated amendment. The source snapshot carrier links the exact prior hashes.
• Structural validation: local nice -n 19 ionice -c3 python3 scripts/validate.py passes with 69 components, 4 profiles, 228 receipts and 10,402 hashed files. git diff --check passes. Final registry checks and commit chronology are reported below.
• Hosted validation has not run: this draft conflicts with main and has no merge ref. Local results establish local consistency only; they do not establish hosted CI acceptance.

scan's older prompt filter remains raw/provisional; the controlled oracle rebuilds complete original owner/turn context. Actual native counter qualification, equivalent Claude proof and the adoption synthesis remain pending. Malformed command value types outside documented native shapes remain an explicit source limitation. No organic counts or adopted defaults are published.

Scope remains this repair on the same branch, with no rebase and one push. Slot defaults and the single north-star-rnd entry remain a stacked follow-up after this repair's publication/read; they are absent from this tree. Landing path: co-op exact-head delta → command-center ACK → hosted CI/refutation at the single landing rebase → 5f's retained catalog queue. Draft status remains.

Sources: native-agent-stack@dee7cae97c74a6657cb40c7bf8324ccbd41f9111:tools/invocation-monitoring/codex_counter.py:49-99,299-309,405-432,611, tools/skill-usage/skill_usage.py:926-936; openai/codex@d27764b82f7118f674371e6d6e76271d9d606edb:codex-rs/core/src/shell.rs:20-30; rtk-ai/rtk@v0.51.0:README.md:6,313; Python casefold, subprocess arguments, and shlex.join.

Evidence-class table

Claim	Class	Evidence
Earlier execution/results	Their unchanged original classes	Source receipts/artifacts
Source-byte, mode, metadata and AST conservation	source_review / local_integration	Fold receipt and scoped critic
Control filters and typed counterexamples	synthetic	34 Codex control fixtures; native counts remain null
Generator/checker/registry consistency	local_integration	Actual commands below
Real organic proof, daily owner carrier, adoption synthesis	pending	No qualified native oracle or finalized selections

Historical FIX2 repair checks

The three requested findings are repaired and the bounded same-family source micro-review has zero remaining findings within those seams. It is source review, not independently executed upstream or native acceptance.

• P1: each original judgment's hash-resolved document/pointer ancestry must declare resolved non-synthetic provenance. A single synthetic original or ancestor cannot qualify adoption; unknown/deferred observations remain non-adopted. Red seven methods/12 failures; focused green seven; full catalog suite 67 passed. The positive fixture declares original source_review provenance in a separate registered artifact while its ancillary frozen input remains synthetic.
• P2a: terminal punctuation is excluded from prose tokens; internal punctuation and distinct longer names remain literal. Exact Use serena./Use rtk: plus quoted/bracketed/repeated punctuation cases and positive counterparts are retained.
• P2b: known completed commands/statuses must agree under storage keys and the exact one-hop native identity group used by qualification. Linked-ID red had six failures/six errors. Sibling-alias red had 16 failures, including eight false organic-count cases at the pair-only guard. Sticky incomplete identity now produces unknown attribution; agreeing aliases, duplicate completions and supported enrichment still pass. Final counter suite: 44 passed.
• Source/test bindings: counter 92f83389d1131553c1e06050779876ead04941616948e7b357ad2273f6299249, counter test d4973be70b384bf31f90c660c46c95dd9a44f2172835c77701707f274f837be3, validator 8201cd367bf142c785c240bd8feb768499cdf1b69a9d33c0c88d92a7788e1700, catalog test a6b800588e3699cf97f41dd03c7c9b3955d0728ae08b52d84991b58de76d79f3.
• nice -n 19 ionice -c3 python3 scripts/validate.py: 0; 69 components, 4 profiles, 227 receipts, 10401 hashed files. git diff --check:0. Required test steps were serialized at nice19/ionice3; hash-matching worker catalog results are reused and final worker counter bytes match the integrated root.

No native original-counter scan, organic count, adoption result or rebase was performed. FIX2 uses one ordinary push to the same branch; no merge or pin promotion. Hosted CI at the repair head is not claimed by these local checks; the earlier 77d35e4 read reported CodeQL green and validate not run due DIRTY/no merge ref. Native/Claude oracle proof and adoption synthesis remain pending. Prior fold checks below are historical.

Local commands run (fold snapshot)

Heavy steps use nice -n19 ionice-c3; test scratch is in a private external cache.

• Combined touched modules: first692 tests/42.210s retained two failures and one optional-jsonschema skip. The fold-related absent-owner test is fixed with explicit controls. The existing installed-client auth_storage_failure notification drift remains open under A21; no notification table/overlay changed.
• Final tests.test_finalized_foundation_catalog tests.test_terminal_fragment: 0;83tests/2.561s.
• Codex counter focused module: 0;34tests/5.804s, including three unchanged baseline tests. Earlier ToolSearch and nine source-derived window/alias/unsupported-completion subcase failures remain retained.
• component_matrix.py --write/--check:0;32rows/no flip violations.
• new_host_grand_list.py --check:0;32layers/66winners.
• check-terminal-profiles.py:0;16repository profiles.
• saturation_ledger.py --check:0;5records/bindings verified.
• python3 scripts/validate.py:0;69components,4profiles,225receipts,10399hashedfiles.
• git diff --check:0. One atomic push and exact remote-head readback:0.

These are local checks, not upstream acceptance or an independent native E2E. The first unsafe TMPDIR preflight exited1 before tests; its output and every subsequent failure remain durable. Same-family scoped source critics cleared the repaired deltas; command-center review remains required.

Decision record

docs/decisions/2026-10-07-finalized-sota-catalog.md retains pending selections. The organic control decision and original Terminal/Dagu decisions retain scoped alternatives and overturn conditions. The dated gateway amendment preserves decided history.

Host evidence

Repository-only. No deployment, client/config change, ordinary counter/native-data scan, new live service probe, install, model run, credential access, order or broker connection. Existing Windows apply/read-back/rollback and special-launcher contracts are carried unchanged in windows/README.md; the co-op applies them after review.

Checklist

• Historical observed results and source branches preserved.
• Three originals named; one lane label; draft; exact overlap record.
• A34 dated amendment, checksums/registry last and one retained push.
• Co-op byte/range conservation verification and source-only fold review.
• Co-op original closures with records; all three origin branches preserved.
• Command-center exact-head read/ACK, CI/refutation and 5f landing.

FIX3 publication and check provenance

The two-commit publication is source/receipt commit 3ce2e74e175753e913c6ee67be47531fb01d5a01, followed by registry-only commit d99e8a2e84f531c2c6be2dd5e890e801ec8c76ce. This registry-last statement is backed by those commit objects; a tree snapshot alone does not establish chronology. No rebase was performed.

The final local nice -n 19 ionice -c3 python3 scripts/validate.py returned 0 from 2026-10-07T10:14:51Z to 2026-10-07T10:15:19Z: 69 components, 4 profiles, 228 receipts and 10,402 hashed files. Its retained stdout SHA256 is f25ecd71b5628c0b9eee8c3b74d20d526bce2e6f2b60efda6b2756cdbb29a154; stderr is empty. These are local integrity checks. Hosted validate has not run because the PR conflicts with main and has no merge ref.

A separate same-family GPT source critic, completed at 2026-10-07T10:11:48Z, found no unresolved finding within the requested case/command-representation variant axes. Its record SHA256 is 5099e721939d5984adb338019ae02cfda5b7df27292081f12316c8114f9c35d8, and its exact source/test bindings are in the new FIX3 receipt. It performed source inspection, not independent execution or cross-family acceptance. The co-op's exact-head delta and the command center's ACK remain pending.

[seathatflowsinourveins]

Coordination for a user-requested memory/RAG upstream-quality catalog update, based on main e712bae. I am preparing an additive dated source review under catalogs/foundation/memory-rag-stacks-20261007.json, docs/decisions/2026-10-07-memory-rag-source-stacks.md and evidence/artifacts/memory-rag-stacks-20261007/, linked from catalogs/foundation/README.md. Scope: final architectural recommendations by use case plus immutable upstream source/CI/benchmark evidence; no installed-selection changes, no runtime qualification, no modification or reopening of blind trials. I will leave this PR's finalized-catalog scaffold, Foundation manifest, root catalog navigation and generators to its existing owner. Shared evidence registration will use current main plus registration of this branch's own files. The resulting scoped PR will be linked here for your catalog synthesis; these recommendations are inputs, not a claim that the 79-slot synthesis is complete.

[seathatflowsinourveins]

The additive memory/RAG source-review recommendations are now in PR #828: #828 (head 88a4627; base e712bae). It owns the new dated foundation leaf/decision/review, a small foundation README link, generated us-equities decision-index entries, and its own evidence registrations. No changes to your catalog synthesis, foundation manifest, component matrix or generators. Local validation, 33 relevant tests and all 3 exact-commit pre-push gates passed; required CI is running. The five architecture choices and 18 repo source reviews retain deployment, blind-trial, benchmark and full-stack limits. Please include the new registered supplement when regenerating the decision index after integration.

[seathatflowsinourveins]

Final handoff: the memory/RAG contribution in PR #828 is merged to main as 18150f10b061ca3fb74a63cdb783dbaaf31c4057. All seven required checks passed at reviewed head 6b2a9dd452d2b1da8a7e1eb93585b23d079c1d58; the merged tree matches it exactly. The dated decision is docs/decisions/2026-10-07-memory-rag-source-stacks.md, with five source-quality architecture recommendations, 18 unchanged upstream audit records, and the registered catalogs/foundation/memory-rag-stacks-20261007.json#/repository_decisions supplement. The wording now follows the merged October 6 upstream-evidence policy while preserving installed selections and historical artifacts. This remains additive input to your catalog synthesis; complete-stack execution is not claimed, and your owned synthesis files were not changed.