Ota v1.6.27 Now Available: Effect-Bound Refusal Assurance

Ota v1.6.27 closes V12: effect-bound refusal assurance. It adds a deliberately narrow foundation for declaring a database-schema mutation, deriving its exact selected realization, and refusing before selected work begins. Provider execution remains disabled in this release.

The important boundary is exactness. Ota derives the selected contract attachment, resource, migration bytes, working directory, invocation origin, and, when policy is available, one command-scoped decision before its selected execution path begins. It can then report a refusal with execution_started: false rather than starting setup or rendering a workflow preview that describes the lane as runnable.

This is not a database executor, provider integration, or a claim that Ota has protected a database. It is a reviewed, machine-readable refusal boundary for one selected contract closure.

Typed effect declarations

Contracts can declare database_schema_mutation through a canonical resource binding, a typed effect definition, and an exact task attachment. Resource, consequence, attachment, evidence, and realization identities are separately content-addressed so the policy decision cannot flatten distinct sources or invocation origins into a broad task label.

Declared-only commands remain review truth, not execution authority. A command without an eligible typed realization cannot become runnable merely because it names the same conceptual effect.

Refusal before selected work

ota run, ota up, ota up --dry-run, runtime and lifecycle proof, CI checkout re-evaluation, and sandbox capability reporting share the typed-admission boundary. A selected typed denial is evaluated before Ota's selected setup, workflow environment rendering, proof artifacts, durable logs, services, dependencies, shell dispatch, or provider contact.

Blocked ota up --dry-run --json output retains the admitted non-secret application plan and available decision evidence. It contains only the refusal action, reports BLOCKED, and does not say that Ota would execute the selected command.

Explicit policy and canary evidence

Policy packs can evaluate exact typed-effect realizations with deterministic deny > warn > allow precedence. Typed lanes stay provider-disabled even when policy evaluates to allow or warn. An effect-refusal canary passes only for one eligible realization with an applicable explicit typed deny before execution starts; fallback and unrelated denials cannot turn into a passing control.

Private archive reconciliation

Operators can archive a workflow typed refusal explicitly. Ota re-derives the archived contract, selected closure, migration plan, policy snapshot, decision, typed-deny basis, and pre-execution posture before accepting it. Doctor promotes only the exact current workflow claim backed by valid private evidence; stale, task-only, ambiguous, invalid, equivalent-path, and opaque-path claims remain unknown.

ota contract effect-refusal-candidate is the first review-only bridge from that evidence back to contract authoring. Its schema-v5 candidate proposes one canary declaration as unknown, has no application projection, and remains non-writable through apply-candidate.

Detector and evidence improvements

GitHub Actions detection now retains canonical job and step working directories in source-bound task closure truth. Dynamic, noncanonical, missing, or escaping directories do not become runnable root-task claims. Named multiline verification steps remain complete ordered bodies and unresolved rather than being truncated into a promoted command.

The Core-owned pressure-evidence registry records each retained case's exact revisions, matrices, exercised surfaces, proven facts, and not_proved boundaries. The Site pressure index is generated from that registry; it is not a repository certification, endorsement, or badge wall.

Pressure and evidence

V12 retained immutable Linux and macOS controls for the exact selected boundaries, including typed run and up refusal, blocked preview construction, policy and CI re-evaluation, sandbox admission, canaries, archive history, Doctor fallback after tampering, and review-only candidate reconciliation. Final bounded real-repository matrices ran against forked Plausible and Outline revisions.

The evidence is tied to exact revisions and its limits are retained alongside the claims.

What this does not claim

v1.6.27 does not prove provider contact, authorization, or mutation behavior. It does not prove database migration success, rollback, correctness, or data integrity. It does not prove absence of arbitrary child processes, complete repository immutability, independently administered policy authority, positive execution receipts, positive assurance, public archive-export safety, or repository-wide readiness beyond the selected contract closure.

Those are not release-note caveats. They are boundaries that keep a refusal Ota witnessed distinct from an outcome it did not observe.

Use the live contract reference, policy-pack reference, and command reference for the current specification and operator flows. The pressure-testing index is the canonical registry for retained evidence and its limits.

v1.6.27 is live here: https://ota.run/releases/v1.6.27

If you already use Ota, upgrade and verify:

ota upgradeota --version --jsonota validateota doctor --jsonota tasks --useota run <task> --dry-run --jsonota up --workflow <workflow> --dry-run --json

The GitHub release contains the shipped binaries and checksums. Author only behavior your repository can review; a typed declaration is precise contract truth, not a reason to infer provider authority.

If you are evaluating Ota for the first time, v1.6.27 shows the standard the product is setting: a consequential repository operation can be declared precisely, evaluated against exact policy truth, and refused before selected work begins without claiming an outcome Ota did not observe.

Ota is not turning an effect declaration into execution authority. It is making the difference between declared intent, reviewed refusal, and proved outcome explicit enough for humans, CI, and agents to rely on.