Review Workbench Prototype#

The Survey review workbench prototype is a example-backed browser demo for inspecting one current/proposed ReviewItem and generating a ReviewDecision-shaped payload in local memory.

The reusable implementation lives under src/review-workbench/ and is exported as @kontourai/survey/review-workbench. The standalone browser demo remains under examples/review-workbench/ as a thin bootstrap plus page-owned assets. The workbench uses publicDirectoryReviewItemExample from a browser-safe data module that intentionally avoids package barrel imports and Node-only dependencies, and npm run check:review-workbench fails if its ReviewItem data drifts from the canonical example-data/public-directory-review-resource.ts example. The page renders the current and proposed candidates, source URL and source ref, locator and excerpt, extraction confidence, candidate values, reviewer note, decision controls, decision effect, a compact Surface preview, and generated JSON payload.

Running Locally#

Build the package first so the browser can load the compiled example module:

npm run build

Then open examples/review-workbench/index.html in a browser from the repo root. The first screen is the workbench itself.

The package check for prototype artifacts is:

npm run check:review-workbench

This is a static artifact and example-provenance guard. It confirms the built example files exist, the browser-safe example matches the canonical ReviewItem example, and Node-only imports are absent from the built example modules. It is not browser/UI evidence for responsive behavior; browser evidence is collected separately with Playwright.

npm run verify includes the content boundary check, typecheck, tests, and this static review workbench guard.

A sanitized copied producer proposal can be adapted into the same ReviewItem shape without adding product-specific workbench branches; the consumer examples under examples/review-workbench/ show the pattern.

The example queue also includes a generic regulated rule-conflict ReviewItem modeled after a downstream rules-management proof. It uses the same current / proposed candidate roles, source locator, extraction confidence, producer policy, and projection hints as the public-directory example. Survey code does not know the downstream product name; the workbench only consumes the portable resource shape.

Decision Mapping#

The prototype keeps reviewer note and selected decision state only in browser memory. It does not persist or send payloads.

All generated payloads use the resource contract from src/review-resource.ts: apiVersion, kind: "ReviewDecision", metadata.name, spec.reviewItemName, spec.candidateId, spec.status, spec.actor, spec.reviewedAt, spec.rationale, spec.projection, and status.appliedToClaimIds. The workbench also generates ReviewSession and ReviewSessionEvent payloads for session audit and replay; those are covered in consumer-integration-guide.md.

Surface Preview#

After a reviewer chooses a decision, the workbench builds a browser-safe Surface preview from the active ReviewItem plus the locally generated ReviewDecision. Accepting the proposed candidate and keeping the current candidate produce different previews because the selected canonical claim, source evidence, candidate history, and review event are derived from the decision candidate.

The preview is intentionally small and labelled. It shows:

• Selected canonical claim: candidate id, claim id, selected value, and review status.
• Unselected candidate history: the candidate values not selected by the decision.
• Source evidence: source URL/ref, excerpt, extraction details, and visible sourceAuthority metadata when the example provides it.
• Review event: actor, review time, status, rationale, and projected review outcome id.
• Integrity posture: candidate set, raw source, extraction, and checksum fields when present.
• Authority trace: an empty/not-provided neutral state unless portable authority trace data is actually present.

sourceAuthority is evidence metadata about the source, such as authority class, declaring system, and scope. It is displayed in source evidence but is not promoted into authorityTrace. An empty authorityTrace is not an error in this prototype; it means the example does not include portable actor/system authority trace data.

The preview records source and review posture for projection. It does not validate leaf truth or assert that the selected value is true in the real world. It is not a Surface Console replacement, trust graph view, persistence flow, or production API.

Non-Goals#

This prototype does not add production persistence, authentication, multi-product tenant administration, a queue/session workflow, a backend builder abstraction, or live downstream integration. Producers still own review UX, workflow state, assignment, policy, and adapters. Survey owns the portable review resource shape and projection hints.