Buildout corpus

Consolidated 20 October 2026. Six domains, three notes each, plus two reviews. This is a decision input, not a plan of record; the consolidated program is in program.md.

Status update

Status at 2026-10-19T23:38Z. The reviews observed the harness gate at 12 passed / 1 failed (citations) at their review time; all 12 were false positives, because backticked API route strings and media file extensions were read as repository-relative paths. The corpus spans were corrected at 2026-10-19T23:38Z by rewriting the spans while keeping their meaning; the citations check was not changed and the buildout directory was not skipped. The full gate now passes 13 of 13 checks with 0 warnings and 0 failed (exit 0). The typed-citation improvement (P0.1) remains an open recommendation, not a blocker.

Status at consolidation (historical)

• The harness gate was red at the reviews’ observation (historical). The reviewed run was 12 checks passed and 1 failed (citations), and every broken citation was inside this corpus: route strings and media extensions read as file paths. The corpus could not pass until those spans were corrected on 2026-10-19T23:38Z (see Status update); the fix was the corpus, not a lowered check, and skipping this directory was never acceptable.
• The reviews correct four material claims. Radar’s research counts are wrong by roughly a factor of two; acumen’s Anthropic translators exist unwired; iptvcatalog’s backoff and demotion claims contradict the code; the harness analysis’s 13 of 13 is a pre-corpus snapshot.
• No product gate was re-run for this index. Product and external facts are the notes’ claims, not re-verified here. The only gate state at consolidation was the reviews’ observation; the current gate line is in Status update.
• A green product gate is evidence about a tree, not proof a feature works. Nothing in this corpus should be read as product health.

Layout

Directory	What it holds
research	External landscape per domain: market, competitors, pricing signals, technology, gaps, sources and confidence. Dated 2026-10-19. Treat as a snapshot; the analysis corrects it.
analysis	Product reality check against the tree: what exists, what is claimed, mechanism, scorecard, risks and unknowns. Dated 2026-10-20, except harness (2026-10-19, stale on the gate).
expand	Eleven or twelve expansion ideas per domain, a ranked top three, kill criteria, sequencing and what not to build.
reviews	The adversarial review (claims challenged, required corrections) and the coverage-gap review (matrix, missing coverage, cross-domain gaps, recommended additions).

Each domain has one note in each of the first three directories, named for the domain.

Domains

Domain	One line
acumen	OpenAI/Anthropic-compatible router in one Go binary and one SQLite file: plan-aware ledger, learned ranking, failover, audit log. Strongest maturity; compatibility, cost and learning claims outrun the mechanism.
massprox	Proxy supply and SearXNG search pipeline: gather, verify, rank, export. Real pipeline and product-grade plumbing; differentiation is commodity until CONNECT selection and trust are packaged.
radar	LLM price and intelligence desk: 661-source registry, truth ladder, four unit-separated boards. Deep data, weak provenance and rights; data ends 2026-10-05 and committed credentials block exposure.
iptvcatalog	IPTV probe, rank and export tool: six commands, one SQLite file, ranked and dead outputs. Working batch pruner; “alive” means reachable and the rank is deterministic, not known predictive.
dock	Stale mirror of the Pelican Wings game-server daemon: no local delta, panel, release or user. Archive candidate; no buildout.
harness	Ops gate and evidence layer: 13-check gate, product records, wayfinder, consent-gated local memory, read-only DSH surface. Best evidence discipline in the portfolio; was red on its own corpus at the reviews’ observation (corrected to 13/13 at 2026-10-19T23:38Z) and is not an operating system.

Reviews

• Adversarial review — challenges specific claims, separates overclaims from defects, and lists 14 required corrections. Read it before quoting any note.
• Coverage-gap review — scores each domain on market, competitor pricing, own economics, technology, risk and expansion; names the missing commercial half and the cross-domain gaps.

Reading order

1. program.md — the consolidated roadmap, dependencies, risks and open decisions.
2. This index.
3. The analysis notes for the domains you own; the analysis is the correction layer over the research.
4. The research notes for market context, with their dates and confidence caveats.
5. The expand notes for options and kill criteria.
6. The reviews before publishing or quoting any note.

Rules of the corpus

• Label every claim observed, derived or unknown.
• Cite external claims with URLs; no credentials, tokens, private keys or absolute host paths.
• Do not put dot-prefixed or slash-prefixed path strings in backticks; the citation gate reads them as file paths.
• A research note is a dated snapshot; an analysis supersedes it where they disagree; an expansion is a set of options, not a commitment.
• A gate verdict is evidence about a tree, not proof a feature works.

Known gaps

• Commercial half absent. No product has a cost-to-serve, a price, a margin or a willingness-to-pay test.
• Portfolio scope unsettled. Three of six active projects are outside the stated thesis; dock’s lifecycle is wrong and its disposition ticket does not exist.
• Corpus gate was red at the reviews’ observation (resolved 2026-10-19T23:38Z). Twelve false positives from route strings and media extensions; the corpus spans were corrected and the gate now passes 13 of 13. The broken-citation count changed once (four to twelve) before the repair. The typed-citation model (P0.1) remains the durable fix, not a blocker.
• Radar exposure. Credential material is committed and the API is unauthenticated with wildcard CORS; no rotation or history rewrite is recorded.
• External facts not re-fetched. Several primary pages were blocked or unreadable to the research workers, and vendor pricing is volatile.
• No product gate re-run for this index. The programme’s product facts are inherited from the notes and reviews.