Proof — self-assessment
Everything below came from a run of headwater check on this repository, dated 2026-10-02. Nothing here was typed by a person: a script runs the engine and writes each figure into this page, and the same script fails when the page and the run disagree. The one thing we have not measured is labeled, and not omitted.
headwater generate --check, 2 untyped, 155 excluded, and 2 that are not a Markdown file at all.Principle 11 forbids publishing a number that no run produced. A measurement campaign ran on 2026-09-30, and the status below says what it found, but it measured this repository against itself with parts removed. What would fill this row is a measured comparison against the projects the comparison page names. Until a campaign runs that comparison, this row is the most honest thing on the site, and it stays.
The export layer declares 7 emitter targets. 2 of the 7 are built, and the other 5 parse and report the consumer each one waits on rather than emitting a plausible artifact nobody asked for.
| Target | State | What it is |
|---|---|---|
json | built | The native property graph, with no loss. |
jsonschema | built | A JSON Schema that constrains front matter. |
shacl | waiting | Parses, and reports the consumer it waits on. |
rdf | waiting | Parses, and reports the consumer it waits on. |
skos | waiting | Parses, and reports the consumer it waits on. |
okf | waiting | Parses, and reports the consumer it waits on. |
linkml | waiting | Parses, and reports the consumer it waits on. |
The conformance ladder measures what a repository wired up rather than what its documents say. This repository reaches L0 against headwater/standard 4.15.0, and the levels above it name the gap and the remedy rather than a grade.
The engine runs, and one campaign has measured part of what it claims. This repository types its own corpus, checks it on every commit, and scaffolds, explains, routes, generates and exports it. It consumes its own taxonomy as a pinned published package rather than a local file. The counterfactual campaign of 2026-09-30 measured it by component. The documents under docs/ have a large measured effect on whether an agent gives a sufficient answer. The hooks and skills showed no detectable gain. The evaluation states each figure, with its interval and its cost, and this page repeats none of them. No arm measured the MCP server or the quality of governance itself, and the efficacy claims stay open until one does. The work still open is on the milestones page.
v0.5.0 carries a static Linux x86_64 archive and a macOS arm64 archive, each with a checksum, and the release workflow runs both on a host with no Rust toolchain before it creates the release (#975). So the download is the lead route, and it needs no toolchain. cargo install headwater-cli is the labeled alternative, and the package registry entry #735 named is live at v0.5.0.
This is a self-assessment, and a self-assessment is the weakest form of evidence there is. We publish it under that name rather than as a proof, and this section says what would make it checkable.
Every figure above traces to one command over this repository. Each one is written into this page by tools/refresh-figures.sh, which reads headwater check --json, the obligation register, a generated verb index and the emitter enumeration of the engine source. Run with --check, that script writes nothing and fails when any figure on this page disagrees with a fresh run.
What is still missing is the part we cannot supply on our own. The repository is public, so you can run that script over it yourself, and nothing here asks you to take a number on trust. The benchmark row above stays empty and labeled, because a reading of this engine against a corpus that is not ours is one nobody has taken.