There is a number that gets quoted in this work, and it is a true number: a full internal audit drafted with an assembly costs tens of cents in model calls. Fifty-eight calls at the measured rate is $0.40600. An engagement drafting an audit set across a month is $0.58800. Neither is the cost of an audit, and quoting either as one is the specific error this piece exists to correct.
The error is not arithmetic. It is a category mistake with a plausible shape: the only line in an audit that has a published, measurable, per-unit price is the model call, and a figure that can be measured is a figure that gets reported. Everything else in the audit — the licensed standard, the corpus, the auditor’s days, the auditees’ hours, the work of correcting what was found — is either unpriced by the suppliers involved or is somebody’s time. Those lines are the cost. The calls are the floor beneath them.
Six lines, one of them priced
An internal audit is bought in six places. Only the last of them has a price this publication can read off an endpoint.
Table 1. What an internal audit is made of, what drives each line, and whether it can be priced here. Source: the method in the quality-and-compliance pillar; the last row is the measured 402 price, the others are described rather than priced.
| Line | What it buys | What drives it | Borne by | Priced here |
|---|---|---|---|---|
| The standard | A licensed copy of ISO 9001:2015, and of ISO 19011:2018 if the guidance is used | The standards body’s licence terms; the number of readers | The organisation | No |
| The corpus | Procedures, records and previous reports, prepared as text rather than photographed pages | Document count, format and condition; paper costs more than text | The organisation | No |
| The auditor’s time | Scoping, the map, the checklist, fieldwork, judgement, the log, the report, the closing meeting | Audit-days, and who is permitted to audit which process | The organisation, or its auditor’s employer | No; days are counted, the rate is not set |
| The auditees’ time | Interviews, evidence retrieval, replies to findings, the closing meeting, the actions | The state of the records, and how many people own a process | The organisation | No |
| Correction | The work of fixing what the findings require | What the findings turn out to be: a register rebuilt, a procedure rewritten, a training record completed | The organisation | No |
| Model calls | Drafting the map, the checklist, the log and the report from what the corpus and the fieldwork supplied | Call counts: documents, processes, findings, replies | The organisation | Yes: $0.00700 a call, measured |
The ordering in that table is not the order of importance, but it is close to the order of size, and the last row is smallest by a wide margin. The reason is structural rather than commercial: a model call is a few thousand tokens of text processing, and an audit is a person reading a system and deciding what it means. No improvement in the first reduces the second, because they are not the same kind of work.
The corpus is a capital item
The corpus is the least discussed line and it is the one that changes the arithmetic of repeat audits. It has two components: the standard, which is licensed rather than purchased outright and whose terms are the standards body’s, and the client’s own documents — procedures, work instructions, the records each process keeps, and any previous audit report.
Preparing a corpus is not free, and its cost is measured in the format the documents are in. Text or markdown indexes directly; a scanned procedure indexes only after it has been made machine-readable, and a filing cabinet of records has to be photographed, or typed, or read by a person and summarised. Clause 7.5.3 allows documented information in any form or medium, so nothing in the standard requires digitisation; the corpus does, because an assembly that reads text cannot read a shelf. Where a year of records lives in paper, scanning it costs more than reading it, and the honest conclusion is that the fieldwork should be done by a person with the cabinet open.
The compensating property is that a corpus is built once and used repeatedly. A second audit of the same system re-reads the same procedures, adds the changes, and re-uses the map. The first audit pays for the corpus and the subsequent ones amortise it, which is why an organisation auditing annually gets a cheaper second audit than its first, and why an organisation that audits once, to satisfy a customer, pays the whole cost for one use. The same logic argues against rebuilding the corpus for every auditor: the map belongs to the organisation, not to the engagement.
There is a depreciation point that is easy to miss. A corpus describes a system at a point in time. Procedures are revised, a process is outsourced, a site closes, and the corpus describes an organisation that no longer exists. An audit run against a stale corpus produces findings about documents rather than about work, and the cost of keeping the corpus current belongs in the annual budget beside the audit programme rather than in the audit’s own line.
The auditor’s time is the audit
The second line is the price of the audit in any ordinary sense, and the only honest way to state it in a publication that has not measured audit-day rates is as a count: this audit is twelve audit-days of work, and what those days cost is a matter between the organisation and the person doing them.
Twelve is not a small number and it is not arbitrary. It is the total of scoping, corpus and mapping, the checklist, fieldwork, the findings and the closing meeting, and the report, on a six-process system across two sites, and most of it is not fieldwork. The counting is the useful part; the rate is the reader’s.
Two properties of that line are worth stating because they are usually left out of a quote.
The first is that the work cannot be done by whoever is free. Clause 9.2.2 requires auditors to be selected and audits conducted so that objectivity and impartiality are ensured, and nobody audits their own work. In a small organisation this is a real cost: the quality manager who owns the system cannot audit it, so the twelve days have to come from a second person, and training that person is a cost that lands before the first audit rather than during it.
The second is that the twelve days are not evenly valuable. The map and the checklist decide what the fieldwork asks; the fieldwork produces the evidence; the log and the report turn evidence into something management can act on. Days removed from the map produce a shorter audit that asks the wrong questions, which is the most expensive kind of saving available in this work, because it is invisible until the certification body asks the question the internal audit missed.
The auditees’ hours and the cost of correction
Two lines sit outside the auditor’s invoice and are almost always absent from the budget.
The auditees’ time is the largest under-counted item. Twelve audit-days of fieldwork involve process owners in interviews, in producing registers and records, in answering findings within the reply window, and in a closing meeting; an illustrative six auditee-days is a planning assumption of the same order as the auditor’s, not a measurement, and it moves with the state of the records rather than with the auditor’s efficiency. Where records are in order, an interview is an hour and the evidence is produced in a day. Where they are not, the same interview produces a week of reconstruction, and the audit’s cost rises without a single additional audit-day.
The cost of correction is the line nobody budgets at all. Findings produce actions: a register rebuilt, an interval recorded, a procedure rewritten, a training record completed. Those actions are the point of the audit and they consume the organisation’s own people for longer than the audit did, because a corrective action changes a process and a process is what people do every day. An audit whose findings produce no cost is an audit whose findings changed nothing. The entry that starts that cost has to be answerable, which is the discipline the piece on audit findings sets out.
Neither line carries a price in Table 2, and the absence is the finding of this piece rather than an omission from it.
The same audit, six lines, one number
The consequence of the preceding sections is a table with a single priced row.
Table 2. The twelve-audit-day audit of a six-process, two-site system, costed on its six lines. Source: the model calls are priced at the measured 402 rate on Base, 31 October 2026; the days and the corpus are counted, not priced; the two unpriced lines are stated as such.
| Line | Basis | Amount |
|---|---|---|
| Model calls | 58 calls at $0.00700 | $0.40600 |
| The standard | One licensed copy, terms of the standards body | Not priced here |
| Corpus | 22 documents, prepared as text | Not priced here |
| Auditor’s time | 12 audit-days | 12 days, at a rate this publication does not set |
| Auditees’ time | 6 days, planning assumption | 6 days, at the organisation’s own cost |
| Correction | Whatever the findings require | Not priced here |
The arithmetic of the one priced row, for completeness: 22 extraction calls at $0.00700 are $0.15400; 7 mapping calls are $0.04900; 6 checklist calls are $0.04200; 8 fieldwork-note calls are $0.05600; 4 findings calls are $0.02800; 3 report calls are $0.02100; 5 reply calls are $0.03500; 3 rewrite passes are $0.02100. 22 + 7 + 6 + 8 + 4 + 3 + 5 + 3 = 58 calls, and 58 × $0.00700 = $0.40600.
Set against the days, that is $0.03383 an audit-day, or $0.02256 for each of the eighteen days the organisation spends on the audit when the auditees’ six are counted. The figure equals one audit-day only if somebody works a day for forty-one cents. Multiply it by a hundred — $40.60, a figure no plausible change to the method would produce — and it is still smaller than a single audit-day at any rate above forty-one dollars.
The bounds behave as they do elsewhere. At the circulated catalogue rate of $0.00875, which is Qwen’s and not this service’s, the same fifty-eight calls cost $0.50750, which is $0.10150, or 25 per cent, more. At the metered ceiling of $0.00955 — a long prompt under exact, or upto at max_tokens: 1000 — they cost $0.55390, which is $0.14790, or 36 per cent, more. The 0.50 USDC opening balance buys seventy-one calls at $0.00700, so the audit uses fifty-eight and leaves thirteen, worth $0.09100.
Why three published counts do not agree
This site publishes three call counts for this work and they are not the same number: fifty-eight calls here for a single twelve-audit-day audit, eighty-four calls and $0.58800 for a month of drafting in the pillar, and one hundred and ten calls and $0.77000 for an engagement on the role page. A reader is entitled to ask which one is right.
All three, and none of them is a price for an audit. The fifty-eight draft one audit end to end: extraction, map, checklist, fieldwork notes, findings, report, replies and rewrites. The eighty-four is a month of an engagement, with twenty documents extracted, twenty mapping calls, twenty drafting calls, twelve replies and twelve rewrites — a drafting month, not an audit. The hundred and ten is the role’s fuller shape, eighty extraction calls over a larger corpus and thirty reply calls, which is an engagement at the scale the role page describes.
The differences come from scope, not from rate, and the fact that three defensible counts differ by a factor of two is the argument against quoting any of them as the cost of an audit. A single number would have to be wrong for two of the three shapes. What is constant across all three is the rate — $0.00700 a call on Base, measured — and the rate is the only thing here that can be stated without describing the work first.
What the model bill does measure
None of this makes the figure useless. It measures drafting volume, and drafting volume is a fair proxy for the shape of the work: a corpus of forty documents costs twice what a corpus of twenty costs; a log of twenty-four entries costs what twenty-four entries cost; every auditee reply that has to be worked is a call. An operator watching the call count can see where the engagement is and how much of it is left, which is what a budget line is for.
It also sets a floor that cannot be argued downward. Whatever an audit costs, it costs at least the calls, and the calls are small enough that no decision about the audit should turn on them. The practical use of the figure is negative: it rules out cost as a reason not to audit, which is the reason most often given. The audit is expensive because of the days and the corpus, and if those cannot be found, the model bill being forty-one cents does not help.
Where the figure stops
Four limits, and the first is the piece.
There is no rate in this piece, because there is no measurement behind one. Audit-day rates differ by jurisdiction, by seniority, by whether the auditor is employed or contracted, and by what the engagement includes; publishing a figure would be inventing a market. The method here gives a count of days, which is the part that can be argued with, and leaves the rate to the reader who knows their own.
The auditee days are a planning assumption of the same order as the auditor’s, not a measurement. They will be wrong for any particular organisation, and they are wrong in the direction that flatters the budget: the reconstruction of missing records is unbounded, and no table can predict it.
The corpus and the licence are described rather than priced because their terms are the standards body’s and the client’s. Where a licence is per-reader or per-site, the cost depends on how many people the organisation lets read the standard, which is a decision it has not usually made.
And the correction line is unpriceable in advance by construction: it is whatever the findings require. A budget that includes an estimate for it has estimated the findings, which is the one thing an audit cannot do before it is run.
The decision
Budget the audit in days and in corpus, and put the calls in petty cash. Concretely: a line for twelve audit-days, a line for six auditee-days, a line for the licence and the corpus preparation, and a line for corrective action sized after the findings rather than before. The model calls go underneath as $0.40600, noted and not discussed, because a line under a hundredth of the next one is not a budget decision.
Then apply one test to any cost figure you are offered for an audit, including the ones on this site: does it contain a day of somebody’s time? A number that does not is a number about paperwork. The drafting can be measured, and it has been; the audit is the days, the records and the corrections, and until those are in the sum the figure is a floor with a price tag and no roof.
Where this goes
- The pillar: x402.review/2026/10/31/quality-and-compliance/ — the job, the six artefacts and the certificate no agent issues
- The role: studio.x402.press/roles/iso_auditor — the agent, its parts and its running cost
- The service it buys: x402.press/services/deepseek/ — the canonical price, seven chains
- The review: x402.review/2026/10/31/deepseek-review/ — the measured handshake and what it does not promise
Method note
The price was read from the live 402 challenge on Base on 31 October 2026, outside DeepSeek’s weekday peak windows of 01:00–04:00 and 06:00–10:00 UTC; the catalogue figure of $0.00875 was checked against it, found to be Qwen’s rate, and is used here only where it is named as the figure that circulates beside its correction. The call counts, the twelve audit-days and the six auditee-days are planning assumptions stated in the tables and repeated in the arithmetic; the rate for an audit-day is deliberately absent because this publication has measured none. The corpus, the licence and the corrective action are described and not priced, and no estimate is offered for any of them. The organisation in the example is constructed for this piece; no client, engagement or audit outcome is described. No figure here is derived from a catalogue, and no saving, return or failure rate is claimed.
x402.review is written and edited by N.P. Vincent. The publication is connected to x402.press, which sells the services reviewed here; that interest is disclosed on every piece that touches a service, and it is stated here rather than left to inference.
Filed under Jobs
