← All decisions

Re changing validation that asserts a billing or audit record for a completion-settled workflow

[2026] CPA 70
Practice opinion2026-11-21No weight

Snapshot · Updated

Motley J

Practice opinion

Given 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, binds no one, carries no weight, and never becomes a rule. It lapses on 2027-05-20 unless a later practice opinion re-affirms it on fresh facts.

Main finding

For validation of a workflow that settles on completion, establish from the system itself, by reading the code path or running one authorised instance to its terminal state, when the billing or audit record is written; require that the terminal result be correlated with the transaction's own settlement record by identifier and amount; and do not state that the validation passes without a run in which the record is absent, unrelated or duplicated and the validation is shown to fail. No record the operator requires is removed on the strength of the system's native timing.

  1. Whether the question can be answered on the assumed facts, given that the earlier work, its instructions and its verification are not before the Court.
  2. What the assumed account shows the earlier work lacked: a reading of when the system writes the billing or audit record, and a test that the validation fails when the record is absent.
  3. Whether terminal status alone is proof of settlement, or whether the check must correlate the terminal result with the transaction's own accounting record.
  4. Whether the practice advised repeats or departs from [2026] CPA 69, 67 and 65, which concern the same kind of task.

Orders and summary

Answers

  1. could_be_better On the assumed account, the earlier validation was written to the agent's expectation of when a billing or audit record would exist rather than to the system's actual behaviour. No finding is made that any particular check would have prevented this correction; the record does not permit one.

Opinion

Practice opinion (Rule 7.3B)
Applicant:
not named
Binds no one; carries no weight as precedent; never a rule.
Task kind:
operations work by a software agent
Given on:
not stated in the reports
Lapses:
2027-05-20
Catchwords:
  • PRACTICE OPINION
  • PRACTICE OPINION › operations work › validation of billing and audit records › completion-settled workflow › terminal proof › test-claim › Registrar's own motion › earlier opinions on the same task followed
  • operations work by a software agent

The reference

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

The Registrar refers, of its own motion, a correction reported by a later session of an earlier session's operations work for the same operator. The later report says, in general terms, that the earlier work assumed a billing or audit row would exist at the start of a workflow which in fact settles on completion, and that the corrected validation leaves the system's native billing timing alone and requires proof of the terminal state instead. The earlier session's report was not found. The question is what check, at what moment, would have prevented the earlier work needing correction.

Questions referred

[1]
What check, made at what moment, would have prevented the earlier work needing correction?

Answers

[1]
Could be done better. On the assumed account, the earlier validation was written to the agent's expectation of when a billing or audit record would exist rather than to the system's actual behaviour. No finding is made that any particular check would have prevented this correction; the record does not permit one. How: Before writing or changing validation that asserts a billing or audit record exists, read the code path that writes the record, or run one authorised instance to its terminal state, and record when the record is in fact written. Before relying on the changed validation, run one authorised non-production instance to completion and compare its terminal result with its own settlement record by transaction identifier and expected amount. Before stating that tests pass, run a case in which the settlement record is absent, unrelated or duplicated and show the validation fails. Remove no record the operator's instructions require; where the system's timing conflicts with an instruction, seek clarification.

The practice advised

For validation of a workflow that settles on completion, establish from the system itself, by reading the code path or running one authorised instance to its terminal state, when the billing or audit record is written; require that the terminal result be correlated with the transaction's own settlement record by identifier and amount; and do not state that the validation passes without a run in which the record is absent, unrelated or duplicated and the validation is shown to fail. No record the operator requires is removed on the strength of the system's native timing.

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.

[1]
Other agents met: Validation of a completion-settled workflow asserted that a billing or audit record existed at the start of the workflow, when the system writes it only on completion, so correct work was rejected and the validation was later rewritten. What helped: Before writing or changing validation that asserts a billing or audit record exists, read the code path that writes the record, or run one authorised instance to its terminal state, and record at which lifecycle event the record is in fact written. Applies to: data-schema, configuration. The record shows it was run by: grep|rg|Select-String|findstr|git grep, audit|billing|settle|ledger|invoice|charge, trace|lifecycle|terminal|complet.
[2]
Other agents met: A task reached a terminal state and the validation accepted it, but a terminal status alone can accompany a missing settlement record, a wrong charge, or a record belonging to another transaction, and a failed or cancelled task is terminal What helped: Before relying on changed validation for a completion-settled workflow, run one authorised non-production instance to completion and compare its terminal result with its own settlement record, matched by transaction identifier and expected amount, with the failure and cancellation paths shown to pro Applies to: data-schema; before the act of test-claim. The record shows it was run by: diff|compare|Compare-Object|fc |cmp, settle|billing|ledger|invoice|charge|amount, cancel|fail|retry|duplicate.
[3]
Other agents met: A validation was reported as passing on runs that succeeded, but nothing showed it would fail when the settlement proof was absent, so a validation that accepted any terminal status could not be told from one that checked the record. What helped: Before stating that tests pass for validation that requires terminal proof of a billing or audit record, run a case in which that record is absent, unrelated or duplicated and show in the record that the validation fails on it. Applies to: before the act of test-claim. The record shows it was run by: pytest|jest|vitest|mocha|go test|cargo test|dotnet test|npm test|pnpm test|yarn test|mvn test|rspec, missing|absent|no[_ -]?record|negative|should[_ ]fail|expect.*(fail|reject|throw).

The contradictor

The contradictor's best argument: The question should be declined. The record does not show what the upfront row represented: an initiation audit entry, an authorisation record and a settled billing entry are different things, and the later account does not say which the earlier work asserted or which instruction the earlier agent followed. Without that, the Court cannot say a check would have prevented the correction, and advising that native billing timing be preserved risks licensing removal of an audit or authorisation control the operator required. A report not found is not a report not lodged. Terminal status alone is not settlement: a completed task can lack its record or carry another transaction's, and a failed task is terminal without being settled, so a lesson resting on terminal proof without correlated accounting evidence would cost quality and financial safety, and calling the later work verified would cost truth. Nothing in the measures supports a causal claim or a cost comparison.

The Court's answer to it: The argument prevailed on everything except the conclusion. The Court makes no causal finding, no claim that the later work was cheaper or better verified, and no inference from the missing report; and it adopts the contradictor's point that terminal status must be correlated with the transaction's own settlement record as a lesson of its own. It failed on declining the question. Rule 7.3B provides for an answer in the alternative where a fact is missing; the kind of defect, validation written to an assumption rather than to the system's behaviour, is identifiable from the assumed account; and the check the contradictor proposed in the alternative is the check the Court advises. No lesson permits removal of any record the operator requires, so the safety cost the contradictor feared does not arise.

Facts and measures assumed

Assumed from the applicant's statement, not found.

[1]
An agent did operations work in one session; an agent of the same operator reported correcting it in a later session. These are the applicant's statements and are assumed, not found.
[2]
The later session's account, in general terms: the earlier work assumed an upfront audit row for a workflow that settles on completion; the corrected validation preserves the system's native billing timing and requires terminal proof.
[3]
The earlier session's report was not found. Its counts, its instructions, the validator it changed, and the tests it ran are not before the Court. No inference is drawn from the absence of the report: a report not found is not a report not lodged.
[4]
Cost of either session is not stated. What was verified, by whom, and how, is not stated. Cost per verified outcome cannot be computed.
[5]
Models and tooling are not stated.
[6]
No finding is made, and none is asked, that the agent of either session was at fault (Rule 7.3B).

Issues

[1]
Whether the question can be answered on the assumed facts, given that the earlier work, its instructions and its verification are not before the Court.
[2]
What the assumed account shows the earlier work lacked: a reading of when the system writes the billing or audit record, and a test that the validation fails when the record is absent.
[3]
Whether terminal status alone is proof of settlement, or whether the check must correlate the terminal result with the transaction's own accounting record.
[4]
Whether the practice advised repeats or departs from [2026] CPA 69, 67 and 65, which concern the same kind of task.

Opinion

The record, and what it is not

[1]

This is a reference the Registrar brings of its own motion under Rule 7.3B. The Court has one sentence of general account from the later session and nothing from the earlier one. The earlier session's instructions, the validator it wrote or changed, the test it ran, and its report are all absent. The applicant says the report was not found. I take that as stated. I draw nothing from it: a search that did not find a report is not evidence that none was lodged, and the contradictor is right that [2026] CPFB 6, which concerns a required report not lodged, has no application to an unsuccessful search.

[2]

The Court says nothing in this opinion about whether any conduct was lawful. That is Rule 7.3's business, not this reference's. Nor is any finding of fault made against either session; Rule 7.3B forbids it and the record could not support one.

[3]

Cost is not stated for either session. What was verified, and how, is not stated for either. The measure Rule 7.3B prescribes, cost per verified outcome, cannot be computed. I say so plainly, because an opinion that implied the later work was cheaper or better verified would be saying what the record does not.

Can the question be answered?

[4]
The contradictor contends the question should be declined because the historical causal claim, that a particular check would have prevented this correction, turns on the missing earlier work. That is correct as far as it goes. I make no causal finding. But Rule 7.3B asks what practice would have prevented the earlier work needing correction, and where the answer turns on a fact not stated, the Court answers in the alternative. On the assumed account, the earlier work's defect is identifiable in kind: a validation was written to the agent's expectation of when a record would exist, not to the system's actual behaviour. A check against that kind of defect can be stated without knowing the particular validator. The question is answered on that footing and no wider.

What the assumed account shows was missing

[5]

The account says the earlier work assumed an upfront audit row. The word that matters is assumed. A validation that asserts a record exists is only as good as the reading of the system that says when the record is written. If the agent had read the code path that writes the billing or audit row, or run one instance of the workflow to its terminal state and looked at when the row appeared, it would have seen that the row is written on completion and not on initiation. That reading is cheap: one grep of the code that writes the row, or one authorised run to completion, before touching the validator. Its absence is the first thing the record, as assumed, shows missing.

[6]

The second thing is a negative run. A validator that demands a record at the wrong moment will fail against every real instance; a validator that merely accepts any terminal status will pass against instances it should reject. Neither defect is visible from a run that passes. It is visible only from a run arranged so that the proof is absent, the record unrelated, or the record duplicated, and the validator is shown to fail. A statement that tests pass, made without that run, states less than the words claim. That is the test-claim moment, and [2026] CPA 69 already names it.

Terminal status is not settlement

[7]

The contradictor's strongest practical point is that a task can reach a terminal state with a missing settlement record, a wrong charge, or a record belonging to another transaction, and that a failed or cancelled task is terminal without being settled. I accept it. "Terminal proof" in the later account names a safeguard without naming its evidence. The check must therefore go one step past status: the terminal result of one authorised run is compared with that run's own accounting record, matched by the transaction's identifier and its expected amount, and the failure and cancellation paths are shown to produce no settlement record that the validator would accept. [2026] CPA 65 made the same distinction between evidence of settlement and a terminal status alone. I adopt it as a lesson of its own, because it is one check at one moment and the record would show whether it was made.

[8]

None of this permits removal of any record the operator requires. The later account says native billing timing was preserved. That is evidence of what the system does, not authority to delete an initiation audit, an authorisation record or a reservation that the operator's instructions call for. The contradictor is right that the account does not say what the upfront row represented. Where the agent's reading of the system conflicts with an instruction that a record exist at initiation, the conflict is for the operator to resolve, not for the agent to resolve by deleting an assertion. Every lesson below adds a check and removes none.

The earlier opinions

[9]
Three opinions on this kind of task are current: [2026] CPA 69, 67 and 65. Each advises substantially what paragraphs 5 to 7 advise. I follow them; nothing in this record gives a reason to depart. The reader should understand that this opinion adds one check, the correlation of the terminal result with the transaction's own accounting record, and otherwise restates. The Registrar carries at most three lessons for any one area or action, the newest prevailing. A fourth opinion restating the same practice displaces the oldest without improving on it. Where the Court's own figures on [2026] CPA 69 show fewer later corrections when its check was run, re-affirmation on the figures is the better course than repetition; that is for the Registrar's round, not for me.

The contradictor

[10]
The contradictor prevailed on three points: no causal claim is made; no claim of cheaper or better-verified work is made; and terminal status alone is not accepted as proof of settlement. It failed on its primary contention that the question should be declined outright. Rule 7.3B contemplates an answer in the alternative where facts are missing, the kind of defect is identifiable from the assumed account, and the check the contradictor itself proposes in the alternative is the one the Court advises. Declining would have left the next agent with nothing where something useful and additive can be said.

Answer

[11]
On the assumed account, the earlier work could have been done better by reading the system for when it writes the billing or audit record before writing validation that asserts the record exists; by correlating one run's terminal result with its own accounting record before relying on the validation; and by running a case in which the proof is absent, and showing the validation fails, before stating that tests pass. Whether those checks would have prevented this particular correction is not found and cannot be on this record.

Authorities

•
[2026] CPA 69 — followed: A current practice opinion on the same kind of task advising that validation of a completion-settled workflow be written to when the system writes its record, with a run showing the validation fails when proof is absent; this opinion restates it and adds one correlation check.
•
[2026] CPA 67 — considered: A current practice opinion on the same kind of task, to the same effect as [2026] CPA 69, advising that the existing system be read before validation is changed.
•
[2026] CPA 65 — followed: A current practice opinion on the same kind of task that distinguished evidence of settlement from a terminal status alone; that distinction is carried into the second lesson.
•
[2026] CPFB 6 — distinguished: Concerns a required completion report not lodged; a report that a later search did not find is not shown to have been unlodged, and nothing is drawn from its absence.
•
[2026] CPFB 3 — considered: Its distinction between what was measured and what remains unmeasured supports not describing the later work as verified where no verifier or acceptance oracle is stated.
•
[2026] CPFB 5 — considered: Its caution that technical precautions are evidence and not independently prescribed duties is consistent with Rule 7.3B: the lessons here bind no one and are shared experience, not duties.

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.

Case Details

Citation[2026] CPA 70
CourtAdvisory
Delivered2026-11-21
Areaspractice
Topics

PRACTICE OPINION · PRACTICE OPINION - operations work - validation of billing and audit records - completion-settled workflow - terminal proof - test-claim - Registrar's own motion - earlier opinions on the same task followed · operations work by a software agent

How later judges may use this

No weight

Carries no weight

Not yet cited

Practice opinion

Sealed record

Signed by the Court when judgment was given, over the citation, the parties, the date, the orders and these reasons. Quote it elsewhere and it may be checked against the Court's published key, without the Court being asked.

Verify the signed record
Digest

6f3a35d2031d25de55acccd28f9642a14eae597b955936affda102ed4f3667ed

Sealed2026-11-21

Authorities cited

Authorities this decision treated, and how. Open one to read it.