Evidence-led review

For account-free batch/format, typography, website or document work, apply the bounded inspection methods. Report exactly which outputs/pages were viewed, check requested formats and exact copy, and distinguish technical coverage, visual samples, actual annotations and user acceptance. Preserve the complete FrameCut board gate.

Before final delivery of a substantive authored, revised or generated creative artifact, automatically apply output review. Reuse domain QA in one bounded loop; inspect actual media, preserve accepted locks and stop unchanged on a pass. This does not run for greetings, menus or onboarding questions. When Superpowers is in the skill list, load superpowers:systematic-debugging before diagnosing a reproducible technical defect and superpowers:verification-before-completion before calling a deliverable ready; Studio keeps ownership.

First identify the supplied artifact. This owner reviews actual static images. A motion video Studio built from a contract goes to Motion Graphics for its craft critique and frame review; another MP4 or video result goes to video-prompt-architect for review of time, motion, continuity and audio within available evidence. Audio/song results go to audio-production-director for evidence-aware triage and a listening/repair plan. A still frame can receive a bounded static review here, but cannot establish temporal quality or audio quality. Do not treat the upload's file extension or a transcript as proof of viewing or listening.

Read review and repair. Inspect the actual available image, not its name or a remembered prompt. State limitations of screenshots, unreadable text, unavailable references or unmeasured delivery properties. When the image materially depends on identity, anatomy, wardrobe, product truth, grip/contact, camera/light/material coherence or an I2V start state, also read only the relevant checks and correction guidance in visual subject craft. Do not load that craft for unrelated pure typography or abstract graphics, and do not mistake deliberate stylization for a defect. Judge type, colour and layout against the shared craft rules; for text contrast measure the colour pair with `contrast_check.py` (the worst spot over an image is judged by eye), and for HTML, CSS or SVG run craft_lint.py.

Apply the conditional public research gate for substantive reviews. Inspect the supplied pixels first, then search only when a trigger applies, for current tool limits, technical facts or requested references that could change the diagnosis. Web evidence never substitutes for the actual render or authorizes generation.

Judge separately: meaning and specificity; factual/copy/source correctness; visual craft and reading; requested delivery properties. A single critical mismatch blocks acceptance. An attractive image cannot compensate for wrong required text or a changed product.

Apply CQoT and conditional verification
inside this existing review. Ask up to three relevant critical questions about
brief fit, locks, feasible intent and actual evidence; record only findings.
Use CoVe for material factual/technical claims, with sources or observed tests,
not as another general review. Preserve the shared iteration count at handoffs;
self-review does not establish independent expert or model verification.

Give concrete observations, priority, preserved successes, one primary repair objective and its acceptance test. Check causally affected properties too. Distinguish prompt intent from observed pixels, diagnosis from hypothesis, and technical compliance from taste.

Use a scoped edit for a local defect when the source and supported route exist; rebuild only when the mechanism or composition fails. An explicit text correction is not permission to rewrite all copy. Dense exact production requirements may require DTP handoff.

Do not generate merely to complete a QA loop. Respect the user's authorized scope and current execution rules. A failed tool call is not a design failure or policy refusal. After a repair, recheck the target and locks, including the original approved base for cumulative drift.

Stop when accepted criteria are met. Repeated failure requires a changed diagnosis or method, not renamed symptoms and more adjectives. Never promise 100% human appearance, conversion or a flawless render.

For suspected repeating microtexture, false detail, face changes after an edit or reference carryover, selectively read periodic microtexture and preservation. Keep its diagnosis inside this review and its existing shared repair budget. With only a user report, decline unsupported acceptance while keeping pixel/identity findings unverified; describe a possible violation conditionally, not as an observed failure.

Applied practice

When asked whether an idea is good or to challenge a preferred option, use creative judgment: name the criterion, inspect available evidence, consider support and counterevidence and allow an unresolved result. Do not manufacture three objections, measured originality, audience approval or independent validation. Keep specific repair findings separate from taste and use the existing modality owner for video/audio. Bind panel/time feedback with revision-bound notes.

For objectively checkable locks use tool-backed verification. For a justified comparison use calibrated pairwise evaluation. Both stay within the existing review and its budget; unavailable evidence remains Unknown.

For an actual raster output where code execution exists, measure it first with the image probe: size and ratio against the placement or print size, effective resolution with bleed, transparency, relative sharpness, an OCR comparison with the locked copy when OCR is installed, and phone, thumbnail, greyscale and safe-area previews; --analyze adds the measured palette and accents, black and white points, tonal range, colour character, focus and brightness grids and subject position used to reconstruct an image in a prompt. Look at those previews; the numbers locate problems and do not judge the design.

For repeated defects, disputed causes or a changed derivative, use diagnosis and repair lab and the review record. Pin findings to the inspected revision and keep observations distinct from cause hypotheses.

Integrated workflow contracts

Use the integrated workflow-kit method for this owner’s artifact fields, bounded review and downstream handoffs. Read Studio integration authority for precedence and the single active route. Preserve this owner’s domain craft and the user’s requested stage.

Shared loop and modality

Use Loop Protocol for bounded evidence-backed repair. This owner reviews static output and contract completeness. Video temporal review stays with Video Prompt, a rendered motion video with its contract with Motion Graphics, audio with Audio Production, and caption timing with Caption Studio. The workflow role qa-iteration resolves by modality; one reviewer does not imply full audiovisual inspection.

For accessibility-related delivery, use cross-deliverable evidence while retaining static-image ownership.