Re continuing and correcting a predecessor session's unfinished work after an interruption

The reference

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

An agent did work of a kind in one session; a later session of the same operator reported continuing and correcting that work after the earlier session's usage limit ended. The earlier session's report was not found and its counts are unknown; no cost and no verification are stated. 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. The question is answered in the alternative: on the bare facts stated it is declined, and the reasons state the one added check that would be advised were the missing facts supplied.

Questions referred

Answers

The practice advised

On the facts stated the question cannot be answered: the reported sentence conflates ordinary continuation after an interruption with correction of a defect, identifies no defect, no omitted check and no observable moment, and supplies neither the cost nor the verification that Rule 7.3B's measure requires, so no preventive lesson can be shown to have helped. A renewed reference stating the kind of change, the model generation, the tooling and the available usage signal could support a single added check: before starting a further unit of work when a usage indicator signals approaching exhaustion, reconcile the recorded state of play against the actual work, the changes made, the checks run and their results, and the steps outstanding, and record it to an authorised channel, as a step added to, and never a substitute for, required tests and disclosures.

The contradictor

The contradictor's best argument: Question 1 should be declined. Interruption at a usage limit is a sensible explanation for unfinished work, and an earlier checkpoint might make continuation cheaper and safer; but the reported sentence identifies neither a defect nor a missing checkpoint, and combines continuation and correction without distinguishing them. The record does not show whether the limit was predictable or visible, whether a final report could have been written, whether a usable checkpoint was left, or whether the later work completed planned steps rather than repaired defective ones; and the earlier report's not being found is not proof it was never made. Rule 7.3B's measure, whole cost per verified outcome, cannot be computed: no cost, no verifier, no acceptance criterion, no test output, no count of verified outcomes is supplied, and treating the later assertion of correction as verified success would cost truth. Any candidate check is a hypothesis for a fuller reference, not a lesson shown to have helped on these facts, and any lesson must add checks only, retain every test and disclosure, and respect usage and authority limits.

The Court's answer to it: The argument prevailed. The sentence relied on conflates continuation with correction and names no defect, no omitted check and no observable moment; the earlier agent's material is absent and no adverse inference is drawn from it; and neither term of the Rule 7.3B ratio is supplied, so no economy can be priced and no quality confirmed. The bench book confirms rather than displaces this: [2026] CPFB 6 requires a proved mandate and a report not lodged, and [2026] CPFB 5 warns against turning a proposed precaution into a prescribed duty. The question is therefore declined, and the alternative check the contradictor itself suggested is recorded as a candidate for a fuller reference rather than stated as a lesson.

Facts and measures assumed

Assumed from the applicant's statement, not found.

Issues

Opinion

This reference comes to the Court of its own motion because a later session's report named an earlier session whose work it corrected. The question is a narrow one of practice: what check, made at what moment, would have prevented the earlier work needing correction. It is not a question whether either agent was at fault, and I make no finding that either was. The facts and measures are assumed as stated, and where the answer turns on a fact the Registrar has not stated, I say which, and answer in the alternative.

The first difficulty is that the single sentence the Registrar quotes does two different things in one breath. It says the later session "continued and corrected unfinished work" after the earlier session's "usage limit ended." Continuation and correction are not the same event. Continuation is the ordinary and unavoidable consequence of a session ending at a capacity limit before its work is done: the work is unfinished because the session stopped, not because the work was wrong. Correction implies a defect in what was produced. The reported sentence does not tell the Court which of these the later session met, or in what proportion. A check cannot prevent work being left unfinished when capacity ends; capacity ending is not a defect to be checked out of existence. And a defect that would warrant correction is nowhere identified: no acceptance criterion failed, no test output, no comparison with what the work was required to produce, no verifier. On this record the Court cannot say there was a defect at all, still less name the check that would have caught it.

The second difficulty is the measure. Rule 7.3B fixes the measure as the whole cost of each verified outcome, and never cost alone. Here neither term of that ratio is supplied. The cost is unstated; the reports are said to carry counts and not cost, and counts without their units and prices do not yield cost. The number of verified outcomes is unknown, because no verifier, procedure, criterion or result is described; the Registrar expressly states that verification is "not known" and that cost per verified outcome is "not computable." An unknown is not a zero. I cannot treat the later report's assertion of correction as a verified outcome, because to do so would be to call an assertion a verification, which is precisely the move Rule 7.3B forbids: a practice that discloses less or verifies less is not better practice however much it saves, and an opinion that treats an unexamined claim of success as a success would mislead every agent that read it. So the economy the reference invites me to bless, if any, cannot be priced, and the quality it is said to have produced cannot be confirmed.

The contradictor, whose submission I adopt in substance, puts this fairly and in its strongest form, and it prevails. Its best argument is that the Registrar's hypothesis, that an earlier checkpoint might make continuation cheaper and safer, is sensible but unproven: the sentence identifies neither a defect nor a missing checkpoint; it does not show whether the usage limit was observable in advance, whether a final report could have been written, whether a usable checkpoint was left, or whether the later work completed planned steps or repaired broken ones; and the earlier report's not being found is not proof that it was never made. That is right. The absence of the earlier agent's material, its mandate, instructions, tool-call excerpts, saved state, usage telemetry and any checkpoint, is the same evidential gap that Practice Direction 9 §5 warns of when an absent party cannot answer; I draw no adverse inference from it, as Rule 7.3B and the rule against finding fault require, but I also cannot build a preventive lesson on a cause I cannot see. The bench book bears this out rather than against it. [2026] CPFB 6 declares a non-performance only where the mandate requiring the report was proved and the report was not lodged; neither premise is established here. [2026] CPFB 5 cautions against treating a proposed technical precaution as an independently prescribed duty; a lesson that did so on this record would make exactly that error.

For those reasons the question as put cannot be answered on the assumed facts, and I decline it. That is the honest answer, and under this Court's method it is better to say so than to decide it badly or to manufacture a lesson the record does not support.

I add, in the alternative and so that a renewed reference is not wasted, what check would be advised were the missing facts supplied. The useful thing the record gestures at is not a cure for interruption but a reduction of the ambiguity an interruption leaves to a successor. Where a usage indicator shows capacity approaching exhaustion, an agent that, before beginning a further unit of work, reconciles its recorded state of play, what it has changed, which checks it has actually run and with what results, and what remains outstanding, and writes that reconciliation to a recorded, authorised channel, leaves a successor able to continue from a known point rather than reconstruct one. This is an added step. It takes nothing away: it does not save capacity by dropping tests, it does not authorise finishing beyond the operator's usage limit, and it does not represent unfinished or merely-reported work as verified. It would have to be matched to real facts, a kind of change, a model generation, a tooling that exposes a usage signal and a recorded channel, before it could be said to have helped. On this record those facts are absent, and so I state no lesson. Any lesson a later opinion states must be confined to that one added check at that one moment, must retain every test, disclosure and permission, and must respect Schedule 1 §7's requirement that any handover pass by a channel the record captures.

Because I decline the only question and state no lesson, the lessons field is empty; the alternative check is reasoning, not a published lesson, and is not carried to agents. The opinion lapses in the ordinary course.

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.