Samples

The work, shown.

What the work actually looks like when it is done.

Every case here is constructed. What is established and what is not is set out in Evidence.

Most firms describe the work. This page hands it to you — the full deliverable, unabridged, before a single conversation. Every sample here is a constructed case — invented names, real method — and says so on its face. When real, anonymised reports exist, they join this page.

Before you read them

Couldn't an AI just write this?

It could write something that looks like this. It would arrive in a minute, read confidently, and cost nothing. That is exactly the problem — and it is the risk most companies are looking right past.

The numbers here come from systems, not from a model. Every figure in a teardown comes out of the client's own ERP, CRM and inboxes, and the method that produced it is printed beside it so it can be checked. An AI without that access does not say "I can't know that" — it fills the gap with something plausible. And plausible-and-wrong gets acted on, because it looks identical to right.

Every claim tells you how much to trust it. Known means I can point at the source. Inferred means I reasoned to it and show how. Guessed says guessed — and what cannot be known is recorded as NOT-CAPTURED, never papered over. Raw AI output carries no such ladder — every line arrives equally sure of itself. The danger was never the wrong answer. It is the wrong answer delivered with the same certainty as the right ones.

The definitions hold still. Ask a chatbot twice and you get two different reports, with no way to compare them. Everything here is measured against a fixed schema — the same definitions for every client, every quarter — which is the only thing that makes "before and after" mean anything.

None of this comes off a shelf. The system that produces these documents had to be built: formats locked so a script, not a mood, decides the layout; pre-print gates that stop a build cold when a single number is unsourced; self-tests that compare every output against a known-good page before anything ships; procedures tuned until the same inputs produce the same document every time. That is tooling, testing, and project work — done once, carefully, so every deliverable after it is fast, repeatable, and checkable. A prompt gets you an answer. A build gets you an instrument.

And the part almost nobody says out loud: AI is in this work — collection, drafts, cross-checks — because it is genuinely good at that. Then every line it touches is checked against a source before it moves, under a written procedure that says what gets gathered, in what order, and verified how. The difference was never the tool. It is that the tool runs inside a system — records, procedures, fixed measures — and a person stands behind the output.

The risk, plainly: producing a convincing document now costs nothing, which means a convincing document is no longer evidence of anything. The companies that get hurt in the next few years will not be the ones that skipped AI. They will be the ones that made decisions from unverified output — because it looked like this page.

Constructed case

Workflow Teardown

The workflow teardown

An $86,400 quote goes silent, and nothing anywhere counts it. The trace, hour by hour, across four systems — what broke, what it cost, and what could not be known.

$86,400quoted

42 daysof invisible silence

4 systemsno owner

Read the workflow teardown Download PDF

Constructed case

Account Dossier

The account dossier

Two documents about one company: a decision brief that says what to do, and the evidence record that earns the right to say it — every fact carrying its source, its tier, its date and how far to trust it, with an expiry date on the cover and a register of what could not be found.

25atoms of evidence

12sources, tier-ranked

6gaps recorded

Read the account dossier Download PDF

Why constructed cases: nothing here has met a client yet, and a fake report presented as a real one would be caught by exactly the reader it matters to. When engagements produce real reports, anonymised versions replace these — with the client's written permission, and not otherwise.

Describe your situation