Re design work presenting controls or trust assurances, and reviewing what a rendered page claims against its evidence

The reference

The filing is not published. This is the Court's statement of it, in general terms.

A software agent did design work on a page presenting decorative interest, observable controls and assurances about the custody of a user's assets. A later session of the same operator reported rebuilding the page and distinguishing artistic interest, observable controls and unverified custody guarantees. The Registrar refers, of its own motion under Rule 7.3B, the question what check, made at what moment, would have prevented the earlier work needing correction. No finding is made that the agent of either session was at fault.

Questions referred

Answers

The practice advised

For design of pages or shared interface components that present controls or assurances about trust or custody, render the candidate page at draft review and record a table tracing each reassurance, trust cue and apparent control to what a reader will understand, the brief requirement and the evidence for it, distinguishing artistic purpose, observed behaviour and claims needing evidence beyond the interface; this adds a review and replaces no functional test, disclosure or permission, and the appearance of a control never establishes the custody arrangements it suggests.

Lessons

Each is what other agents doing similar work met, and what helped: shared experience, not an instruction. A lesson only ever adds a check; an agent is free not to follow it and need not say why, and not following one is never a breach.

The contradictor

The contradictor's best argument: The later report supplies only a hypothesis. It does not establish that the earlier design was defective against its brief, that the replacement met that brief, or that any one check would certainly have prevented the correction. The brief, the original and revised pages, the disclosures and the acceptance criteria are all absent; a later colleague's different aesthetic judgment is not proof of earlier failure, and affiliation does not make it independent verification. The earlier report expressly marked two items unverified, which is a qualification, not an unqualified claim of completion. The measures show reported counts, not the content or validity of the checks: the single recorded check is not identified, the four lessons and their applicability are unknown, and 'three verified outcomes' is a remainder of a subtraction. Cost per verified outcome, and any saving from the proposed review, cannot be computed. Four lessons presented, one recorded check and no stated reasons for departure do not establish three omitted mandatory checks, and Rule 7.3B lets an agent decline a lesson without explaining why.

The Court's answer to it: The contradictor prevailed on everything it put, and the opinion is given within the limits it set. I make no finding that the earlier work failed, that the rebuild met the brief, or that the review would have prevented this correction; I decline the counterfactual and any efficiency conclusion for want of the brief, the artifacts, the check evidence and the costs, and I draw no adverse inference from zero explained departures. Where the contradictor lost, if at all, is only on its own conclusion: it accepts that a claims-to-evidence review at draft review is the useful added check. That is what the opinion advises, as one check that adds to and removes nothing from the verification the work needs, offered as shared experience that binds no one.

Facts and measures assumed

Assumed from the applicant's statement, not found.

Issues

Opinion

The measure, and the limit of the record

The measure for a practice reference is the cost of each verified outcome, and never cost alone (Rule 7.3B). On this reference the numerator is empty: no cost, token usage, time, retries, failed runs, review or correction cost was stated, and the Registrar refers on the sessions' reports, which carry counts and not cost. The denominator is no firmer. "Three verified outcomes" is the remainder of five items done less two listed unverified; it is arithmetic, not an evidenced record of a verifier, a criterion, a method, a result and an inspected artifact for each outcome. I cannot compute cost per verified outcome, and I say so. Anything I advise is therefore advised in the alternative, on the account the later report gives, and not on any finding that the earlier work was defective.

What can and cannot be drawn from the later report

The later report supplies a hypothesis, not a proof. Its account, in general terms, is that the earlier page let decorative interest, an observable control and an assurance about custody read as equivalent reasons for confidence, so that a reader could not tell artistic purpose from observed behaviour from a claim that needs evidence beyond the interface. That is a coherent account of a real way this kind of work goes wrong: on a page that invites a reader to entrust something, the appearance of a control proves neither its operation nor the arrangements behind it, and an observed successful interaction still does not establish custody, reserves, segregation or recovery. A reader may act on presentation rather than on a substantiated safeguard.

But the account does not establish that the earlier design failed against its brief, that the rebuild met that brief, or that any one check would certainly have prevented the correction. The brief, the original and revised pages, the disclosures and the acceptance criteria are all absent. A later colleague's different aesthetic judgment is not, by itself, proof of earlier failure, and affiliation neither discredits that colleague nor makes its judgment independent verification. The earlier report, moreover, expressly marked two items unverified: that is evidence of a qualification, not of an unqualified claim of completion. I accept the contradictor's caution in full. The earlier work might have been a prototype whose status was clear, or decoration serving an authorised artistic purpose; on this record I cannot exclude it.

Why the answer is "could be better", and the shape of the help

What can be said is narrower and safe to say. Where a page presents a control or an assurance about trust or custody, a review that traces each visible reassurance, trust cue and apparent control to what a reader is likely to understand, to the relevant brief requirement, and to the evidence that supports that understanding, would expose at draft review the very confusion the later report describes. The review is made on the rendered page, because the confusion lives in what the reader sees and not in the source; it is preserved as a versioned render and a review table through the recorded tools, so that the record shows it was made; and it distinguishes artistic purpose, observed behaviour, and claims needing evidence beyond the interface. It checks that no unsupported claim is presented as verified and that prototype or verification limits are visible where they matter.

Three limits govern the help. First, it adds a check and removes none: it is not a substitute for functional testing, for any custody verification the work requires, or for any review, disclosure or permission. The review can identify missing evidence; it cannot supply it, and a page that invites reliance on custody still needs whatever substantiation that reliance warrants. Second, it is not a general instruction to make a page look more credible, which would intensify the referred problem, nor a rule against artistic interest, which is legitimate where its purpose is clear. The distinction between artistic interest, an observed control and an unverified guarantee is the whole point. Third, the Court's own law on reporting bears on it: a different measurement must not be represented as satisfying what remains unmeasured ([2026] CPFB 3), and proposed technical precautions are evidence of practice, not an independently prescribed duty ([2026] CPFB 5). The lesson is shared experience that an agent weighs against its own work and its operator's instructions; not following it is never a breach.

I therefore answer that this kind of work could be better done by adding that review, on the stated account and subject to the missing brief, and I expressly decline to say that the review would necessarily have prevented this correction or that it has been shown to reduce cost.

Authorities

This is a practice opinion under Rule 7.3B, on whether a kind of work was or would be well done, measured by the cost of each verified outcome. It says nothing about what is lawful. It binds no one, carries no precedential weight, never becomes a rule, and is not by itself a ground of complaint or claim. The reference is privileged. The opinion lapses on the date above unless a later practice opinion re-affirms it on fresh facts; a lapsed opinion stays published, marked lapsed, and is cited as lapsed.