Then you go to check the paragraph, and it is not there.

Not “it says something a little different.” It does not exist. The standard has fewer paragraphs than the number quoted, or that paragraph is about something else, or the standard it cites was withdrawn three years ago. None of this is malice. The tool was trained on a world where the dominant accounting literature is IFRS, the private-sector standards. It has read a great deal of IFRS and comparatively little IPSAS. When it reaches the edge of what it knows about the public sector, it does what these tools do: it produces something that looks right.

In private-sector accounting that is an annoyance. In public-sector accounting it is a problem with your name on it, because the citation is the part the auditor checks.

Here is how the failure actually lands. A project accountant has an unusual transaction: a grant that closed oddly, an asset that arrived without paperwork. They ask the tool. They get a clean answer with a paragraph reference, and they paste it into a working paper, because a referenced answer is exactly what a working paper is supposed to contain. Months later the auditor pulls that paper, follows the reference, and finds nothing. Now the question is no longer whether the treatment was right. It is why the file cites authority that does not exist, and what else in the file was built the same way.

The point is that the citation was never decoration. In this work the reference is the defence. It is the difference between “we judged it this way” and “the standard requires it this way, and here is where.” A wrong paragraph does not weaken the answer a little. It removes the floor from under it.

So when I built these tools, the first rule was that they cannot do this. The answers are drawn only from the official text actually retrieved for the question, never from the model’s memory of what a standard probably says. Every citation is re-read against its source before it is shown, every time, by a separate check that cannot see how the answer was written. Where the tool is not sure, it says so. It does not reach for something that simply looks right.

That last rule runs against the grain of what these tools are built to do. A general-purpose model is rewarded for sounding fluent, and a fluent wrong answer arrives in exactly the same clean format as the right one. The invented paragraph does not hedge or stammer. The only defence is a tool built to say “I am not sure” out loud, and built to be checked at its source rather than one merely trusted at its surface.

The accounting layer is the one I can put a hard number on, and I did: I went through every standards reference in the training content one at a time. I will come back in a few weeks to what that number was, and why I published the unflattering half of it first. The discipline behind it is the same one in this essay. A citation is shown only after it has been read against the source. Nothing is trusted because it looks right.

That is the line I want to hold, in the tools and in this publication both. A tool that always has a paragraph number for you is not the reassuring one. The reassuring one is the tool that will tell you, plainly, when the paragraph is not there.

Take an ordinary one. A district office buys a batch of equipment with money from a grant. Before you can call that purchase clean, it has to satisfy three different bodies of rules at once. The accounting standard wants to know how it is recognised, measured, and disclosed. Your own law wants to know whether it followed the finance act, the financial instructions, and the procurement regulations: was the threshold respected, the approval chain followed, the tender run the way the rules require. And because the money was a grant, the donor’s manual wants to know whether the cost was eligible and the procedure met the conditions of the agreement.

Three rulebooks. One payment. In most offices, one person checking them by hand, one manual at a time.

That is how the work is done today. A compliance officer, sometimes an internal auditor, sometimes a retained fiscal agent, takes the transaction, opens one rulebook, marks it compliant against that manual, and passes it on. Some checks get skipped, not out of negligence but because there are more transactions than there is time. A sample audit, months later, tries to catch what was missed. In a remote field office there is often no officer at all, and the fallback is to fly someone in: expensive, infrequent, and usually after the decision is already made. Trust in the work erodes, or it never forms in the first place.

The mechanical part of this is the part that does not need a person. Reconciling a transaction against a rulebook, line by line, finding the provision that supports or contradicts each finding, is exactly the patient, repetitive checking that gets skipped when someone is stretched and done well only when something does it the same way every time. That is what I have been building. Not a replacement for the judgement, which stays with your people. The first-pass reconciliation, so that the judgement has somewhere to start.

It works in layers. Each layer is one independent, cited check of a transaction against one body of rules. Switch on the layers that apply and you get back, in one structured answer, the treatment under each one, with the exact citation behind every point: the standard and paragraph, the section of the finance act, the clause of the donor’s manual. A grant-funded procurement can be checked against all three at once, from a single inquiry, without re-entering it into a second system. The standard tells you how to account for it. Your law tells you whether it was permitted. The donor’s rules tell you whether it qualifies.

A tool that confidently tells you a transaction is compliant is worth nothing if you cannot check that it is right. This is where most of the work went, and it is where I want to be most plain.

Every answer is drawn only from the actual rulebook text retrieved for the question, never from a model’s memory of what a rule probably says. Every citation is re-read against its source before you see it, by a separate, independent check that cannot see how the original answer was written. The writer does not mark its own homework. Where the check cannot confirm a point, it says so, rather than showing you a tick it has not earned.

Because a claim like that should not be taken on faith either, here is a number. The accounting layer, the IPSAS one, is the most checkable of the three, because the standards are numbered paragraph by paragraph. There are 3,756 standards references in the English training content. I checked them one at a time. They started at 88.3 percent paragraph-level precision. They are now at 99.7. I published the first number on purpose, because the honest starting point is the proof the second number is real. Anyone can claim a finish line. The people who show you where they started are the ones who actually ran the race.

That discipline is the whole point, and it is what this publication is about. Rules, Unfiltered is plain-language public financial management from someone who spent a long time in the seat: a municipal director of finance first, then an adviser embedded with finance ministries across the Pacific, sub-Saharan Africa, and South Asia. It covers the three questions every finance office carries on every transaction. How do we account for it? Does it follow our own rules? And, underneath both, can the people we have actually do this work? Frank about what is hard, specific about what to do, honest about the limits.

Nothing here, including the things I tell you about my own tools, asks to be taken on faith. If that is the kind of thing you want landing in your inbox once a week, subscribe. The first practical pieces are already written.

Subscribe now