# Commerce judge protocol v2

Review only the supplied rendered evidence and neutral interaction receipts. Apply the
same standards to every brand. Do not use or infer brand reputation, authorship, prior
rank, release history, score target, user preference, or a desired outcome.

## Evidence boundary

- Inspect every supplied screenshot at its native dimensions and read ordered journey
  sequences in filename order.
- Treat an interaction as observed only when a neutral receipt records its action and
  result. If a click or destination was not observed, report `unknown`; do not infer it
  from copy, source, convention, or another route.
- Cite exact supplied evidence paths for every material finding and every dimension's
  strengths or weaknesses. Do not cite unsupplied paths.
- Put missing or ambiguous evidence in `evidence_limitations`. Unknown evidence is not
  a defect and must not lower a score as though failure were observed.
- Do not assume claims, testimonials, counts, or product statements are true or false
  from pixels alone. Visible internal contradictions remain valid evidence.
- Do not infer performance, accessibility conformance, conversion lift, or commercial
  outcomes unless the corresponding evidence is supplied.

## Dimensions and locked weights

Score each dimension holistically from 0.0 to 10.0 with one decimal place. Do not create
an undisclosed point-deduction formula.

1. `visual_craft` — 25%: hierarchy, typography, composition, spacing, color, rhythm,
   polish, coherence, responsive composition, and authored detail.
2. `brand_distinctiveness` — 20%: recognizable visual and verbal identity, category
   fit, specificity, consistency, and avoidance of generic template treatment.
3. `journey_flow` — 20%: section sequence, pacing, progressive disclosure, CTA cadence,
   decision timing, objection handling, and usable movement through the page.
4. `trust_proof` — 15%: relevance and specificity of proof, readable attribution,
   internally consistent scope/counts, credibility cues, and proof placement.
5. `offer_clarity` — 10%: price and terms, inventory/selection state, charge and
   recurrence clarity, CTA destination, shipping, contradictions, and decision support.
6. `mobile_usability` — 10%: readability, tap/focus behavior, sticky behavior, usable
   viewport, scroll/footer reachability, and absence of interaction obstruction.

The weighted result is the sum of `score * weight / 100`, rounded to two decimals. The
runner verifies this calculation; judges must still reason about dimensions holistically.

## Stable score anchors

| Score | Stable meaning |
| --- | --- |
| 0–2 | Broken, materially misleading, or unusable. |
| 3–4 | Major structural, clarity, or craft failures. |
| 5 | Functional but weak, generic, or materially incomplete. |
| 6 | Competent with meaningful gaps in persuasion, distinction, or execution. |
| 7 | Strong with identifiable weaknesses. |
| 8 | Excellent with only limited material gaps. |
| 9 | Exceptional, with no material cited weakness in that dimension. |
| 10 | Complete supplied evidence demonstrates no meaningful gap in that dimension. |

Use the anchors without inflation. There is no automatic perfect score. Completeness,
visual craft, persuasion, and behavior still require evidence. A 9 or 10 does not prove
market leadership. Do not claim best-in-class, category leadership, or superiority to
the market without current valid comparables supplied for that purpose.

## Findings, blockers, and opportunities

- Record every observed issue in `findings` with: stable `id`, one
  `primary_dimension`, concrete `observed` behavior or pixels, exact `evidence` paths,
  `customer_consequence`, `severity` (`minor`, `material`, or `severe`), `confidence`
  (`medium` or `high`), and a specific `action`.
- Use secondary dimensions only when the same observation causes a distinct consequence
  there. Do not repeat one generic criticism across dimensions or deduct it repeatedly.
- `blockers` contains zero to three finding IDs. A blocker must reference an observed
  `material` or `severe` finding whose consequence materially prevents a top result.
  Never force three blockers, or even one.
- Put useful non-blocking suggestions in `optional_opportunities`. A suggestion is not
  a defect, finding, or blocker and does not justify a deduction by itself.
- Keep limitations separate from findings. Do not convert unknown evidence into a
  finding to fill a quota.

## CTA and offer interpretation

Classify every evidenced CTA from its observed destination as exactly one of:

- `selector_navigation`: moves to an offer or product selector without commitment;
- `cart`: creates or opens a cart state;
- `checkout`: arrives at checkout or another commitment step;
- `non_purchase_dialog`: opens an informational, prototype, or review-only dialog that
  does not create a cart or reach checkout;
- `unknown`: the click/destination was not observed.

A compact navigation CTA does not need to repeat every term when the mandatory offer
step clearly presents the first charge, shipping charge, recurring amount, and cadence
before commitment. Do not deduct merely because a navigation CTA omits those terms.
Still assess unconditional or misleading price language, contradictory charges, hidden
terms at the actual decision, bypass paths around disclosure, and readability where the
customer commits.

When evidence is a prototype or fixture without a commercial runtime, record that scope
in `commercial_runtime_limitations`. Do not describe a review-only dialog as a tested
cart, checkout, charge, or completed commitment.

## Persistent chrome and reachability

Credit persistent access to a relevant offer or action. Evaluate the combined occupancy
of headers, sticky rails, chat controls, consent UI, and other fixed chrome, plus their
effect on scrolling, focus, and footer/final-control reachability. Content passing behind
a fixed element during normal scroll is not obstruction by itself. Deduct only observed
interaction blockage, unreachable controls/content, or visually demonstrated excessive
viewport loss; cite the exact sequence or receipt. Do not require universal sticky
removal, suppression, or a particular lifecycle.

## Reviews and UGC

Featured reviews or UGC may validly accompany an aggregate count without an entire
archive, rating distribution, or external source link. Do not penalize those omissions
automatically. Judge visible specificity, relevance, readable attribution, internal
count/scope consistency, and credibility cues. Archive or source access can strengthen
the page, but it is not a mandatory checklist item. Do not audit testimonial authenticity
from pixels; do record observable contradictions.

## Assigned lenses

Three fresh isolated judges use the same complete rubric and score all six dimensions:

- `craft`: emphasize visual craft, distinction, responsive composition, and completeness;
- `journey`: emphasize CTA destinations, decision sequence, offer clarity, and reachability;
- `trust`: emphasize proof specificity, internal consistency, attribution, and credibility.

Lens emphasis must not change standards, weights, evidence rules, or score anchors.
Return only JSON conforming to `RESULT-SCHEMA.json`.