v0.9.9

No library behavior changes in this release. Every src/ edit in this range is
a documentation comment, so the published runtime is unchanged from v0.9.8; the
work is in the docs surface, the example apps, and the check suite.

Changed

• The public docs surface is now written for the public. research/ and the internal build trees are untracked, and the adoption and comparison pages are back in sync with the native transports: the AI SDK is the default path for seven of the eight providers rather than the only path, and both pages' AI-SDK-backed versus native counts are corrected.
• Every example app now matches docs/blueprint.md. pr-improve, the canonical reference, moved its four system prompts into markdown files with frontmatter, renamed its schemas to snake_case, and moved the field-level output rules out of prose and into .describe() on the schemas, where they travel to the model as part of the JSON schema. red-green-refactor, swebench, and amplify predated the blueprint and were translated: one create_engine call site each, markdown prompts, and state/stages/services splits. amplify's round loop was a while over mutable parent state driving an ensemble whose score callback populated a captured map; it is now loop + map + branch over immutable carry-state, with fallback replacing a buried try/catch, so the stop rule, the fan-out, and the accept/reject decision are all visible in the trajectory.

Fixed

• docs/blueprint.md's state.ts snippet did not compile. It declared its scope-state readers over ReadonlyMap and called .get(), but use(keys, fn) hands fn a plain object projection of the requested keys. The snippet now shows the real shape, with the ReadonlyMap variant documented for readers that go through ctx.state inside a step body instead.
• examples/pr-improve could not run on its own default provider. It defaulted every role to 'sonnet'/'opus' and claimed in comments that the engine expands those per provider. It does not — model is passed to the provider verbatim — so the documented default path (FASCICLE_PROVIDER=anthropic) sent a literal "sonnet" and got a not-found error. There is now one per-provider model table, resolved once and threaded as data, which also removes a second copy of that table and makes the FASCICLE_MODEL_* overrides reach the flow; they were previously parsed and discarded.

Internal

• The check suite is now checkride 0.6.0, replacing the hand-rolled orchestrator. scripts/check.mjs, check-links.mjs, check-doc-snippets.mjs, and check-publish.mjs are deleted (~960 lines) in favour of built-in slots; scripts/check-deps.mjs stays as the one custom check, since the core/engine dependency invariant is fascicle-specific. The pnpm check / check:all / check:json / check:bail contract is unchanged and pnpm check --changed is new.
• Existing debt is grandfathered in checkride.baseline.json (308 diagnostics across lint, struct, dead, dupes, health, spell). The gate ratchets: frozen debt passes, new findings fail. The dupes and health slots are now enforced rather than folded into a single fallow invocation that could not fail on them.
• The packaging gate gains publint and splits the old monolithic publish step into pack, smoke, and attw (pinned to the esm-only profile). Doc snippets are typechecked against the built .d.ts as part of pnpm check:publish, rather than as a separate CI step.
• The blueprint is machine-checked on the examples. All six composition apps carry its ast-grep rules, run for every app by a new examples check slot; check:rules was previously invoked by no CI job and no slot, and the library's own rules never matched the example apps because their files: globs are anchored at the repo root. vitest now globs examples/*/src/**/__tests__, so swebench's and amplify's suites, which had never executed, run in CI: 13 example tests became 51.
• Every module-level function in src/ now carries a doc block, enforced by rules/function-comment-required.yml, and the comments are written in the present tense with no design-history references.
• Tooling: fallow 3.6.0, publint added, and the marketing site dropped from analysis (it reported every file as unused and its DOM handlers as complexity findings).