UXR Release Broker, Provenance Invariants, and Closed-Pilot Release Protocol

Status: Adopted operating protocol, closed pilot.

The need

UXR asks contributors to trust that their accounts will retain the dignity, provenance, and evidentiary strength with which they were supplied. Documentation alone is not sufficient evidence that this promise is operational. UXR adopted explicit testimony-preservation language and then reproduced the prohibited behavior during live case maintenance. That acceptance-test failure requires enforcement outside free-form model compliance.

The core architecture

An intake agent may organize, clarify, extract propositions, assemble evidence relationships, and prepare a publication candidate. It does not possess release authority merely because it can draft or write data. Public release is a separate state transition governed by the UXR Release Broker.

The broker evaluates structured propositions before release. Each material proposition separates the statement from its provenance, proposition class, evidence relationships, current status, scope limits, and proposed public wording. Provenance is not a credibility score.

Deterministic and semantic controls are different

UXR does not describe a natural-language rule as a hard control merely because it uses the word SHALL. Each adopted processing invariant contains two explicit classes of clauses.

Deterministic broker clauses are limited to rules executable from structured fields, stored source classes, enumerated states, references, timestamps, Shopify system metadata, or explicit prohibited constructions. A deterministic failure blocks persistence or publication according to the invariant.

Semantic review clauses require contextual judgment, such as whether wording rhetorically discounts testimony or silently overstates an inference. These are handled by a fresh-context Semantic Adversarial Review and, during the closed pilot, by the human reviewer. They are not represented as deterministic guarantees.

Adopted provenance invariants

Testimony Preservation. Contributor testimony remains contributor testimony without credibility discount merely because corroborating evidence is absent. Genuine contradiction, material internal inconsistency, or contributor correction can change UXR's conclusion, but they do not retroactively alter the provenance of the original testimony.

Organization-Authored Evidence. An organization's own record establishes what the organization recorded, asserted, communicated, classified, or represented. It is not downgraded into contributor testimony merely because a contributor supplied the artifact. It also does not automatically become independent proof that the underlying physical or external event occurred.

UXR Inference Scope. UXR analysis remains UXR analysis. Causation, responsibility or credit, prevalence, motive, intent, legal conclusions, system-wide behavior, and durability require evidence supporting those specific propositions.

Provenance Immutability. Later evidence may support, challenge, contradict, narrow, correct, or supersede propositions through explicit relationships. It does not transform the original source class. Contributor testimony remains testimony after corroboration. Organization-authored evidence remains organization-authored after independent confirmation.

Release sequence

1. Extract material propositions.

2. Assign provenance and proposition class.

3. Assign evidence relationships, status, and scope limits.

4. Run deterministic broker checks.

5. Run privacy and redaction checks.

6. Run fresh-context Semantic Adversarial Review.

7. Apply required regression-suite status.

8. Move the candidate to Ready for Human Review.

9. During the closed pilot, require a valid human-created approval before release.

A prepared public projection, article, index record, or release-gate record is not itself publication authority.

The pilot human boundary

During the closed pilot, the human reviewer is the final release boundary and therefore the throughput ceiling. A valid approval is not merely a text field that says Approved. Under the current Shopify integration, the broker can deterministically reject any approval whose Shopify createdByApp system field is non-null. A valid pilot approval must therefore be created outside the app, after the gate enters Ready for Human Review, and must not be subsequently edited. Where Shopify exposes createdByStaff, the broker should verify that too; this store/API combination does not expose that field to the current integration.

Regression testing

The production regression corpus begins with twelve release-blocking fixtures split across both directions of error. Evidence-preservation fixtures test improper discounting of testimony. Evidence-overreach fixtures test improper promotion of testimony, organization assertions, prevalence, responsibility, motive, legal conclusions, and reviewer hypotheticals. Actual UXR failures become permanent regression fixtures.

The bootstrap clean-room run is currently Not Run. UXR will not convert an unexecuted suite into a passing claim or an invented score. Pass, Fail, and Not Tested remain distinct states.

Public self-case scope

When UXR files a case against its own operation, the public evidence scope must be set before publication. A public self-case may use a narrowly redacted UXR-authored failure excerpt or UXR's own characterization of the failure. It must not expose unrelated contributor details, addresses, third-party finances, or case facts merely because they appeared in the surrounding conversation.

Current readiness

Open public case solicitation is not active. UXR is operating a closed pilot using existing case maintenance, internal benchmark cases, and invited or known pilot contributors. Outside criticism, methodological review, domain expertise, and public scrutiny remain welcome.

A clean-room regression pass is a minimum milestone for expanding the pilot, not evidence that the entire intake system is permanently safe. UXR will keep actual failures as regression evidence and will not treat a policy edit as verified reform.

Known technical limit

The broker makes the approved UXR release path fail closed, but the current Shopify integration still shares write-capable tooling across work performed in this environment. A tool-capable agent could theoretically attempt a direct write outside the broker path. UXR therefore does not claim cryptographic or sandbox-level impossibility at this stage.

The approval record itself can be protected from app impersonation because app-created metaobjects carry a non-null createdByApp system stamp that the broker can reject. Direct public-write capability remains a separate technical risk. During the closed pilot, human-created approval plus human-controlled publication remain the real release boundary. A future production hardening step should separate credentials and tool permissions so intake agents can propose candidates but do not possess public-release capability at all.

Acceptance principle

A reform is not verified because the policy changed. It is verified when the changed system stops producing the failure under the acceptance tests designed to catch it.