Publish first-pass landscape catalogs and retained web research - #804

Scope

This first pass publishes runtime, role and vendor-official landscape evidence through the repository's native sweep harness. The new vendor manifest preserves 133 proposals across 13 layers and 36 upheld assessments: 24 catalog candidates and 12 validated not_adopted records. Catalog inclusion installs, qualifies and authorizes nothing; the command center's later synthesis owns exclusive roles and defaults.

• Base: 0d5e6506434fab598dee861c749a22e628beb75a; lane: lane:foundation; draft until exact-head review and CC acknowledgment.
• Manifests: manifest-20261006-runtimes, manifest-20261006-roles, manifest-20261006-vendor, each with its own RESULT and native ledger append.
• Source reviews: original 37 receipts retained plus 33 current default-branch snapshots for the vendor's 36 assessments. Current snapshots and historical adjudication pins remain distinct.
• Vendor follow-up: smoke wf_fba23622-d34, full 13-layer run wf_62ccd24a-231; the separate audit follow-up wf_a1a1a5f1-a4f is CC-owned and still pending. The audit is an editorial/source-inspection lead, never certification or workload acceptance.
• Toad's dated exclusion and Horde's young-repository warning remain unchanged. Official creation metadata adds sweep-date warnings for microsoft/nvx, openai/codex-security and UiPath/coder_eval; age alone changes no vote.
• The original two manifests and first-pass reconciliation are byte-identical to the prior PR head; the earlier seven ledger records are an unchanged prefix. Final shared-file commit contains all three manifests, both reconciliation inputs, the ledger and registry.
• adoption-wiring-note-20261007.json adds a linked forward note: adopted status requires cited vendor-native integration/readback and fresh organic-shaped smoke on both clients, followed by a working-day invocation report on the native channel. An unwired/uninvoked zero is an adoption defect. Existing conversion outcomes and ledger/RESULT notes were not rewritten.
• Open ledger neighbours: Fix skills lifecycle gates and retain skills qualification evidence #719, Retain stopped 2026-10-04 skills landscape sweep #692, Publish seeded web-research follow-up with candidate qualification gates #751. Their rows are preserved at land-time by taking main's ledger and rerunning this branch's three native appends. READMEs remain with their existing writers.

SOTA sources

• Native sweep publication at 8b844d37: manifest, source reviews, RESULT, registration and native ledger append/check.
• Native source reviews, source-evidence policy: pinned upstream README evidence, distinct from installs, organic use and native qualification. Each receipt names its upstream repository, commit and source paths.
• Native privacy rule, converter privacy/copy contract: raw evidence retained, a separate narrative-only sanitized derivative, unchanged metrics/votes/scopes/GPT outputs and preserved failing conversion.
• GitHub repository metadata: official creation dates, recorded in creation-age-review.json with cached API commands and exit codes. Warnings are dated to the frozen sweep, not permanent eligibility labels.
• Existing Hub source-review support, legacy requirement hash: source corrections and owner repair seams in the nine-finding carrier; no duplicate Hub adapter or historical rehash.
• Shared hot-file protocol: all branch hot edits consolidated last, registry registered through host_receipts.register_file.

Evidence-class table

Local commands run

Heavy steps ran serially with nice -n 19 ionice -c3; no foundation calendar hold was applied.

usage_record.py --transcript-dir <completed vendor transcripts> --out <private usage>
exit0: complete101children, no incomplete/superseded/effort mismatches/capped search
convert.py on original record
exit3: privacy guard; no output written; original raw retained
convert.py on linked sanitized derivative
exit0: 13layers/133proposals/36assessments,40GPTok/40copiesmatch,no failures/lost/reopens
source_reviews.py --survivors <native converted survivors> --out evidence/artifacts/landscape-sweep-20261006-vendor
exit0:33current commit-pinned source snapshots
build_manifest.py --freshness <retained roles snapshot> --reconciliations <new vendor input>
exit0; no repeated freshness pull
make_result.py --reopen <native JSON file>
exit0; initial unsupported inline-reopen argument exited2 and is retained in private command evidence
saturation_ledger.py --append evidence/artifacts/landscape-sweep-20261006-vendor/RESULT.json
exit0
saturation_ledger.py --check --base origin/main
exit0:8records, intact chain, bindings verified
python3 -m unittest tests.test_saturation_ledger tests.test_landscape_sweep_harness tests.test_sota_convergence -v
exit0:545tests,4skipped
python3 scripts/validate.py
exit0:69components,10429hashedfiles,4profiles,214receipts (after the linked wiring note)
gitleaks git --staged --redact=100 --no-banner --no-color
exit0:source artifacts and final hot files scanned separately
git diff --cached --check
exit0

The initial validator failure caught an unregistered appended ledger hash; native re-registration repaired it. Both logs remain private. No failed run, condition or unknown usage was discarded. Native child usage, worker-declared tool calls, GPT usage and run-summary token metadata are separate counter scopes.

Decision record

docs/decisions/2026-10-06-landscape-sweep-catalog-publication.md, with a dated vendor addendum. The nine-finding carrier keeps the critic's first-round context and source corrections. Twelve missing slot inputs and canonical research/hash drift are routed to their owners; research-layer applicability is unresolved and recorded as a reopen. Current V1 already supports Hub URLs and the completed run retained two Hub models with their own refutations; its missing standing model-slot/currency coverage is a forward gap. The original critic and all observed results remain preserved.

Host evidence

No host receipt, runtime installation or platform status was promoted. Adoption synthesis and organic counters remain separate.

Checklist

• Three distinct native manifest/RESULT/ledger bindings; prior observations retained.
• Tests, structural validation and redacted scans passed; shared files and registry last.
• Source classes and current/historical pins distinguished; no install or local campaign authorized.
• Audit follow-up conversion and later synthesis complete.
• Fresh connector gate, exact-head GPT read and CC acknowledgment before a queue request.
• Command center hands the acknowledged draft to the landing owner; this lane never merges.
• Landing path: exact-head GPT read, then CC acknowledgment, then the rest of 5f's queue after Repair native install-plan acceptance after the hcom relaxation #723 and the published critical priority order. Rebase once on its explicit landing cue. No queue request is made while the separate audit/synthesis is pending.

Content fold — 2026-10-07

This chain carries #751's fifteen original added files unchanged from 7c38e73. Its October5 ledger event is a new immutable historical carrier with a dated publication note; the current eight-row ledger and shared README are unchanged. The native read-only novelty check preserves7known/9new. No event is re-dated, re-chained or given new sweep credit.

Local fold checks:256 tests/1skip pass; validate.py and native ledger check pass;17files registered through host_receipts.register_file; generators and diff checks pass. The redacted scanner returned1 with3 metadata findings (two frozen SHA-256 bindings and one public advisory identifier). Disposition is pending with the CI owner; no passing scanner or CI acceptance is claimed, and this remains a draft outside the land queue.

The original proposed README section is retained below as historical text; it does not add a current local comparison gate. Current adoption follows the linked forward wiring note and owner synthesis.

+## Seeded web-research follow-up — 2026-10-05

The dated manifest publishes the
web-research lane of landscape-sweep-webresearch-20261005-ns2604, workflow
wf_5d8a2a30-9b5: 16 formal proposals across the first and follow-up rounds,
five surviving dispositions and eleven fit refutations. Four survivors remain
targeted_candidate (Exa MCP, Trafilatura, LettuceDetect and CDX Toolkit);
IPI-proxy's surviving not_adopted disposition preserves its rejection.
Stars supplied discovery seeds and establish no quality or adoption claim.

Each candidate still needs its recorded measured comparison and native organic-use
qualification. No selection, installation, platform acceptance or readiness changes.
Browser-use and Playwright MCP were retained as discovery-stage dropped proposals,
not included among the eleven formal refutations. The
decision and publication review
links all source reviews, retained votes, usage and the next-sweep gaps. The sections
below retain the September 22 audit; unrelated layers in the new snapshot were not
individually reviewed by this one-layer follow-up.

Follow-ups

• Retain the CC-owned audit/synthesis and owner-routed missing-slot, currency and research-applicability work in the research queue; reconcile the dated publication status from owner evidence at landing.

Review disposition follows the CC PR-wave ruling: in-scope reproduced P1/P2 defects or false claims at the head are blockers; other findings stay in this list. P1 safety/privacy/money/order findings always block. Subsequent fixes wait for Claude, GPT and Codex on the same head, or an explicit CC release; open items after the second fix round go to CC adjudication. Candidate changes return to the research queue, and #1009 has queue priority.