Work up VCL-UT: repair CI, reconcile the level ladder, and set direction

[hyperpolymath]

Two parts, as asked: deal with the immediate issues first, then work up where this goes — theory and VeriSimDB application kept separate.

Part A — immediate issues (all measured, all verified)

Two CI faults, independent:

• dogfood-gate.yml was malformed YAML. The embedded Python started at column 1 while the run: | block scalar was indented 10, so YAML closed the scalar early. The workflow had failed every run in 0s since at least 2026-07-06 — it never once validated anything — and it broke Dependabot's github_actions updater with dependency_file_not_parseable. Verified by executing the extracted step against valid and invalid manifests, not just by parsing it. (~80 of 266 estate copies carry the identical fault.)
• dependabot.yml declared mix/npm/pip/nix with no manifests present — four of the five weekly failures. Also: this repo has five cargo workspace roots and the single / entry covered one, so the parser and vclt-gate were invisible to Dependabot.

Two crates had silently stopped compiling. ast::Statement gained the consonance field verb; attest and recompute-wasm were never updated (E0063). Only their test modules were stale, so the libraries still built and nothing complained — 12 tests had just stopped running, in the crates carrying the attestation and recompute boundaries. They were covered by no CI job at all, which is exactly how they rotted. Now fixed and gated.

Coverage gaps closed: e2e.yml ran 69 of the root workspace's 102 tests — every e2e and fuzz test was skipped. Also removed two git clone echidna steps that cloned a whole repo per run for nothing (proven: 102 tests and e2e.sh 19/19 pass with no sibling present).

Correction to an earlier finding of mine: I had recorded "0 unit tests, root cargo test is a fake gate". That was wrong — an artefact of truncating the log. The root crate is a re-export facade whose whole surface lives in tests/; the tests-102 pass badge was exact. Noted in the workflow header so the misreading isn't repeated.

Documentation truth pass — the level ladder disagreed with the code three ways:

	was	now
Numbering	1–10	L0–L10 (they're wire tags — requesting 7 got Effect-Tracking, not the Cardinality documented)
Vocabulary	FETCH/REMOVE	SELECT/RETRACT — the documented verbs are rejected by the parser
Count	ten	eleven — L10 EpistemicSafe appeared in no document anywhere

Part B/C — the work-up

docs/2026-07-21-workup-consonance-and-verisim.adoc, two strands kept separate.

Theory. The strongest argument for the consonance reframe came from the code, not from taste: the eleventh level isn't a query-safety property at all, and it's the one level not inherited from TypeLL (a UI panel has no federation, so no warrant question). The paper is titled "A Decalogue…" — its own title can't house its own eleventh level. The gap in the world is stated honestly: type-safe query languages are a crowded, mature field and a ten-levels paper reads as synthesis. What none of them do is type warrant — every one assumes a single source of truth that cannot disagree with itself.

The whitepaper reframe is a proposal, not an executed rewrite. The 1,909-line .tex is untouched. Structure + a full draft abstract are offered for approval first — how a paper argues its claim is yours to settle, and it's the least testable change available here.

Application. Opens with the uncomfortable fact: VCL-UT delivers nothing to VeriSimDB at runtime today — the producer is finished and the consumer was never written (verisim-api/src/vcl.rs has zero vclt-gate references). Two VeriSimDB defects located precisely, including VCLTypeChecker.res:60 building the default context with six modalities though the type admits eight — Provenance and Spatial silently absent from every default type-check. Those are VeriSimDB changes and are not made here.

Closes with a falsification section, including one thing that should be measured before the wiring work rather than after: subprocess latency on the hot path.

Verification

reuse lint                       -> compliant, 460/460
23 workflows                     -> 0 malformed
cargo test --workspace --locked  -> 102 passed
cargo test (parse)               ->  36 passed
cargo test (attest)              ->   9 passed   (was: did not compile)
cargo test (recompute-wasm)      ->   3 passed   (was: did not compile)
bash tests/e2e.sh                -> 19/19, no echidna sibling
                                    150 tests total

Note

The three unpushed main commits (sweep2/3/4, Mistral Vibe) are not carried here. They break reuse lint — a genuinely fail-closed gate — by adding an AGPL licence file to a repo with zero AGPL-declared files, and they inject @metadatastician as owner/sponsor into a hyperpolymath repo while duplicating correct .github/ files. Recommend main be reset to origin/main with a sweep-quarantine branch kept for reachability. This branch is based at origin/main and is unaffected.

🤖 Generated with Claude Code