Clean Health Lab · immutable R21.32 · commerce judge protocol v2
Fairer visual review, calibrated before scoring
Three fresh isolated GPT-6 Astra judges at high reasoning reviewed the unchanged public page through craft, journey and trust lenses. Each received the same 72 opaque rendered screenshots, one neutral corrected behavior receipt, the locked rubric and output schema—without prior scores, page history, user feedback, change notes, authorship, score targets or competitor results.
Scores
| Weighted overall | 8.58 |
|---|
| Visual craft | 8.80 |
|---|
| Brand distinctiveness | 8.63 |
|---|
| Journey | 8.43 |
|---|
| Trust/proof | 8.67 |
|---|
| Offer clarity | 8.00 |
|---|
| Mobile usability | 8.67 |
|---|
Lane spread
| Craft | 8.63 |
|---|
| Journey | 8.48 |
|---|
| Trust | 8.63 |
|---|
Range: 0.15. Lanes are averaged; agreement is consensus, not a multiplied penalty.
Weighted overall = visual craft 25% + brand distinctiveness 20% + journey 20% + trust/proof 15% + offer clarity 10% + mobile usability 10%. Dimensions are arithmetic means across three lanes.
Top supported blockers
- No material or severe blocker was returned.
Supported remaining issues
Across the lanes, the repeated offer issue is that the one-time option says $59.99 plus shipping without an amount or calculation basis (assets 050, 051, 060 and 061). Other minor observations concern small mobile recurrence/gallery-label text (assets 003, 029 and 036), repeated introductions across the proof sequence (assets 007, 008, 010, 011, 021 and 022), and a highly prominent testimonial with little usage duration or benefit-specific context (assets 010, 021 and 055).
What changed in the judge
Protocol v2 keeps the six dimensions and 25/20/20/15/10/10 weights, while allowing zero to three blockers, adding score anchors, requiring cited structured findings, separating visible pixels from tested behavior, and requiring a distinct customer consequence for any secondary impact.
- Compact FREE navigation prompts are judged with the mandatory disclosures in the buy box; they are not automatically penalized for omitting shipping or subscription terms.
- A sticky control is assessed for actual obstruction and offer access, not merely for being present in a screenshot.
- An aggregate review count paired with selected review excerpts does not require a complete review archive.
CalibrationThe accepted narrow calibration detected all three seeded counterexamples and produced zero findings on all three clean cases. This checks the repaired rules against six controlled fixtures; it does not prove broad judge correctness or commercial performance.
Evidence correction
The first page run is preserved but superseded: its neutral receipt omitted a subscription setup reset before the sticky-control test, so two lenses read independent scenarios as one continuous path. The corrected receipt states every setup/reset and scenario boundary. The same 72 screenshot bytes, locked rubric, schema, model and weights were used for this result.
The trust lane's reported total was deterministically normalized from 8.64 to 8.63 so it matched the unchanged six dimension scores and locked weighting formula. No judgment content changed; the normalization receipt records both hashes and the calculation.
Exact results
Craft lens — 8.63
Visual craft — 8.8
Strengths: 3
Weaknesses: 0
Brand distinctiveness — 8.7
Strengths: 2
Weaknesses: 0
Journey — 8.6
Strengths: 2
Weaknesses: 0
Trust/proof — 8.8
Strengths: 3
Weaknesses: 0
Offer clarity — 8.0
Strengths: 2
Weaknesses: 1
Mobile usability — 8.5
Strengths: 2
Weaknesses: 1
Findings
F-01 · offer_clarity · minor
Observed: The one-time option shows $59.99 plus shipping in both selectors and prototype dialogs, without a shipping amount or explanation of how it is determined.
Customer consequence: Customers can compare bottle prices but cannot determine the one-time delivered total from the supplied offer presentation.
Evidence: evidence/asset-050.png, evidence/asset-051.png, evidence/asset-060.png, evidence/asset-061.png
Suggested action: Show the one-time shipping charge, or explain when and how it is calculated beside the price.
F-02 · mobile_usability · minor
Observed: The upper mobile selector compresses renewal price, cadence, and cancellation language into small secondary text; gallery thumbnail captions are also markedly smaller than surrounding body text.
Customer consequence: Important renewal details and gallery labels require more concentrated reading at native mobile size.
Evidence: evidence/asset-003.jpg, evidence/asset-029.jpg, evidence/asset-036.jpg
Suggested action: Increase these text sizes and allow the renewal sentence to wrap naturally while preserving the primary button hierarchy.
Evidence limitations
- The hero subscription scenarios identify the route by label and setup, but their initial screenshots do not expose all selector terms or uniquely identify the clicked element. Pre-click disclosure timing is therefore unresolved.
- COA, label, gallery, and UGC screenshots establish visible states; without corresponding neutral action receipts, their opening, navigation, search, and playback behavior remains unverified.
- Mobile review expansion was not clicked. The hidden desktop-style control does not establish failure of the visible mobile carousel or access to its remaining cards.
- The blank diagnostic frames in evidence/asset-057.png and evidence/asset-070.png precede populated disclosure results. Their duration and cause are not established, so they are not treated as persistent rendering failures.
- Claims, review authenticity, aggregate counts, laboratory contents, and external source destinations were not independently verified.
- The receipt verifies selected focus targets and footer reachability, but does not establish comprehensive keyboard behavior or accessibility conformance.
Exact craft verdict JSON
Journey lens — 8.48
Visual craft — 8.8
Strengths: 2
Weaknesses: 0
Brand distinctiveness — 8.5
Strengths: 2
Weaknesses: 0
Journey — 8.1
Strengths: 2
Weaknesses: 1
Trust/proof — 8.6
Strengths: 3
Weaknesses: 0
Offer clarity — 8.0
Strengths: 2
Weaknesses: 1
Mobile usability — 8.7
Strengths: 3
Weaknesses: 0
Findings
F-01 · journey_flow · minor
Observed: The social-proof area uses successive introductory headings for community videos, customer reviews, a large featured review, and another customer-review collection before FAQs.
Customer consequence: Repeated introductions add scrolling and slow access to practical questions after the shopper has already encountered substantial social proof.
Evidence: evidence/asset-007.jpg, evidence/asset-008.jpg, evidence/asset-010.jpg, evidence/asset-011.jpg, evidence/asset-021.jpg, evidence/asset-022.jpg
Suggested action: Consolidate the written-review introduction and reduce the featured-review block's vertical footprint while retaining its attribution and the separate video format.
F-02 · offer_clarity · minor
Observed: The one-time plan consistently displays $59.99 plus unspecified shipping, including inside the tested review-only dialog.
Customer consequence: A shopper can compare product prices but cannot determine the one-time delivered total from the supplied offer.
Evidence: evidence/asset-050.png, evidence/asset-051.png, evidence/asset-060.png, evidence/asset-061.png
Suggested action: State the one-time shipping amount, or explain where and how it is calculated beside the price. Preserve that explanation in the selected-plan summary.
Evidence limitations
- Screenshots establish visible states, but only evidence/receipt-001.json establishes recorded interactions. Gallery changes, certificate opening/search, UGC playback, source links, most navigation links, and FAQ answers were not interaction-tested in that receipt.
- The receipt's scenarios are independent. Selected-plan continuity cannot be inferred across scenario IDs, and the sticky scenarios explicitly reset selection to subscription.
- Mobile review pagination was hidden and not clicked. The supplied evidence does not establish whether an alternate swipe route exposes further written reviews.
- The blank formula before-captures in evidence/asset-057.png and evidence/asset-070.png are followed by populated disclosure states. Their duration and repeatability are unknown; they do not establish a persistent rendering failure.
- The original certificate PDF, review authenticity, aggregate count accuracy, and external scientific sources were not independently verified.
- The 360-pixel evidence covers the initial hero only. Recorded interaction and footer-reachability tests cover 390x844 and 1440x900.
- The receipt identifies hero subscription purchase actions but does not separately identify the header button or the hero 'Choose your option' navigation control as clicked.
- One-time price and recurrence are explicit, but the shipping amount remains unspecified. No commercial commitment step is supplied to resolve that amount.
Exact journey verdict JSON
Trust lens — 8.63
Visual craft — 8.8
Strengths: 2
Weaknesses: 0
Brand distinctiveness — 8.7
Strengths: 2
Weaknesses: 0
Journey — 8.6
Strengths: 2
Weaknesses: 0
Trust/proof — 8.6
Strengths: 4
Weaknesses: 1
Offer clarity — 8.0
Strengths: 2
Weaknesses: 1
Mobile usability — 8.8
Strengths: 2
Weaknesses: 0
Findings
F-TRUST-01 · trust_proof · minor
Observed: The largest written testimonial emphasizes general quality and reordering, without duration of use or a specific experience relevant to the brain-fog positioning.
Customer consequence: Its prominent placement provides less concrete decision support than its visual emphasis suggests.
Evidence: evidence/asset-010.jpg, evidence/asset-021.jpg, evidence/asset-055.png
Suggested action: If available, feature an accurately attributed review with concrete usage context and duration; retain clear separation between anecdotal experience and product efficacy evidence.
F-OFFER-01 · offer_clarity · minor
Observed: The one-time option and its review-only dialog show $59.99 plus shipping without a shipping amount.
Customer consequence: A shopper cannot calculate the delivered one-time cost from the supplied offer states.
Evidence: evidence/asset-050.png, evidence/asset-051.png, evidence/asset-060.png, evidence/asset-061.png
Suggested action: State the one-time shipping charge or explain how it is calculated and provide an estimate before any future commercial commitment.
Evidence limitations
- Only supplied screenshots and receipt-001.json were assessed; testimonial authenticity, aggregate counts, laboratory findings, and scientific claims were not independently verified.
- Original PDFs, source links, video playback, gallery interactions, most FAQ and policy destinations, and report lookup behavior lack neutral action receipts.
- The receipt's scenarios are independent. Explicit subscription resets prevent interpreting the sticky tests as one-time selections being overwritten.
- Blank pre-disclosure frames in asset-057.png and asset-070.png do not establish persistent missing content; the corresponding after states show the expanded sections.
- Mobile review pagination was not clicked because its control was hidden. Alternative mobile review navigation and the extent of accessible review content remain untested.
- Receipt labels establish hero subscription CTA results but do not separately establish the promotional Choose your option or header CTA destinations.
- Focus receipts establish specific footer and dialog focus states, not comprehensive keyboard behavior or accessibility conformance.
Exact trust verdict JSON
Limits and comparability
This is a rendered-evidence heuristic, not a conversion, performance, efficacy, accessibility or checkout measurement. Static overlap shows what occupied a captured viewport; it does not by itself prove that content was unreachable. The neutral receipt records specific tested actions and max-scroll/focus states only. No current competitor ranking is reported because peers were not freshly scored under this exact v2 contract.
Earlier v1 scores remain immutable, but v1 and v2 are structurally different contracts; their numeric delta is not evidence that the unchanged page improved or regressed. The older AG1 result used a different lane and evidence contract, and no comparable scored IM8 record was found, so neither supports a current rank.
Receipts
First complete-input verdict accepted; no score-selected reruns. Page bytes were unchanged. Report generated 2026-09-16T03:34:59.678291+00:00.
