The notebook is set up and two registers exist. Before the prompts are revised, this hour goes through each document type and records what you ask of it, what you make from it, and where it goes wrong. The session is recorded; the prompts are revised afterwards from the transcript.
The same three questions, every document type
What do you ask of it?The real questions, in your own words. Write the words down exactly: the tool finds documents by the words in them.
What do you produce from it, and where does the time go?Every output is a candidate draft. Every slow step is where a prompt pays.
Where does it go wrong today?Wrong figure, stale version, missed action, mismatched term. Each one becomes a check the prompt is told to run.
How the hour runs
0:00 to 0:03
The format
Six minutes per document type, the same three questions each time, in the order on the timeline. Rows are document types, columns are the questions. The transcript fills the grid afterwards.
Say it the way you would say it to a colleague.The tool finds documents by the words in them, so the exact wording matters more than a tidy version of it.
Small and frequent counts.A ten-minute lookup done every week matters more than a rare hour-long task.
Say where it went wrong.A correction after the fact is where a check belongs in the prompt.
Status reports
0:03 to 0:09
The weekly artefact, read and written by everyone in the room.
What the notebook does with it today
Distils each new report into the current-state fact base: what exists, what is in flight, what is committed, what has stopped, every figure and date with its source.
Carries the previous fact base forward and marks conflicts between old and new rather than picking the later one.
Can draft the narrative sections of the next report from the registers, with gaps marked instead of filled. Not yet tried on the real notebook.
To test, not to lead with
Where the report's figures actually come from each week, and whether the notebook ever sees them. Whether anyone reviews a draft against last month's before it goes out.
What do you ask of a status report?Yours, and the ones you receive. When you open one, what are you looking for?
What do you produce, and where does the time go?Walk through the hours of one reporting cycle. Which parts are assembly, which parts are judgement?
Where does it go wrong?The last time a status report had to be corrected after it went out, what was wrong and how was it found?
Governance minutes and forum papers
0:09 to 0:15
The record of decisions, and the papers that ask for them.
What the notebook does with it today
Distils minutes into the decision log and the actions-and-risks register. Closed items are kept in a closed table, never deleted.
Can prepare a one-page brief for a named meeting from the registers: who is in the room, prior decisions, open actions, figures likely to be asked. Not yet tried on the real notebook.
To test, not to lead with
Whether meeting transcripts exist and could feed the registers directly. Whether forum papers have a house shape the notebook has never seen. Who tracks whether an action asked of someone came back.
What do you ask of the minutes and the papers?"What did we decide about X" or "who owns Y", or something else entirely?
What do you produce, and where does the time go?Writing the minutes, chasing actions, preparing a paper for the forum: which of these eats the week?
Where does it go wrong?A decision recorded differently by two people, an action nobody chased, a paper rewritten because the format was wrong.
Templates and workshop packs
0:15 to 0:21
Deliberately kept out of the notebook. They are forms, not records, and their vocabulary competes with the real documents. But they hold the shape every output is expected to take.
What the notebook does with it today
Nothing. Drafts come out in the tool's shape, not the engagement's.
To test, not to lead with
Whether a one-page description of each output's shape (section order, length, house phrasing) would let the drafts land closer to usable. That is a new kind of file, and only worth making if the room says drafts are currently reshaped by hand.
Which outputs have a fixed shape here?Status report, forum paper, deliverable sections, anything the client expects to look a particular way.
How much time goes into reshaping?When something is drafted, how much of the work after is making it look right rather than making it correct?
Where does it go wrong?An output sent in the wrong shape, an old template used, a section the client always asks to be added.
The cost model and the dashboard
0:21 to 0:27
The numbers. The dashboard cannot be a notebook source at all. The workbook can be uploaded, but nobody has yet confirmed the notebook reads it correctly.
What the notebook does with it today
Nothing specific. Figures reach the notebook only when a document states them, and every figure bound for a stakeholder is checked by a person against the workbook regardless.
To test, not to lead with
Whether anyone would trust a number the notebook quoted. Which numbers get asked for repeatedly and could be carried by a document that states them with their period and unit. Whether a workbook test (upload, ask three stated values, check them) is worth running.
What do you ask of the numbers?Which figures get requested week after week, and by whom?
What do you produce, and where does the time go?Pulling a figure, explaining a variance, walking an allocation from cost pool to service: which takes the time?
Where does it go wrong?A figure quoted from the wrong period, two sources with different totals, a number that changed after it was sent.
The business case
0:27 to 0:33
The funded baseline. The most-asked-about document on any engagement: what did we assume, and what was approved.
What the notebook does with it today
Distilled into a baseline register: assumptions and approved figures, original and adjusted values labelled separately. Done on the real notebook.
Answers grounded questions against it, with a confidence grade and the source named.
To test, not to lead with
Whether anyone compares the current position to the approved baseline, and how. Whether benefits are tracked against it, which would justify a benefits register.
What do you ask of the business case?Try it live: ask the baseline register a real question from last week and see whether the answer would have saved you a search.
What do you produce from it?Variance to baseline, benefits tracking, the "as approved" line in a paper: which of these exist, and who makes them?
Where does it go wrong?An original figure quoted where the adjusted one applied, an assumption everyone forgot was made.
Phase deliverables
0:33 to 0:39
Cost structure, service model, change and stakeholder strategy, and the rest. The current phase's finals are in; superseded phases stay out.
What the notebook does with it today
Drafts against the cost taxonomy page, with every ambiguity flagged as a decision for a person.
Rewrites a passage for a finance executive reader, claims unchanged, unsupported claims flagged.
To test, not to lead with
Whether anyone reviews a draft deliverable for consistency with the others (terms, figures, what phase three said). Whether the executive summary is the slowest part. Whether earlier phases get re-read because nobody remembers what they concluded.
What do you ask of the deliverables?Current phase and earlier ones. When do you open a superseded deliverable, and what for?
What do you produce, and where does the time go?First draft, review, executive summary, reconciling with earlier phases: rank them by hours.
Where does it go wrong?A term used two ways across two deliverables, a figure that contradicts the business case, a section the client sent back.
The notebook's own pages
0:39 to 0:45
Six craft pages on IT financial management, the registers, the index and the handoff. The files we wrote rather than the engagement's.
What the notebook does with it today
The craft pages ground answers on taxonomy, service costing, showback and chargeback, benchmarks and public-sector budget mechanics.
The handoff carries a session to the next person. The index says what exists and when it was last touched.
To test, not to lead with
Which craft page has never been useful, and which page is missing. Whether the handoff has been read by anyone other than the person who wrote it. Whether a log of every question asked, answered or not, would tell us what to build next.
What have you asked the notebook so far?The questions it answered well, the ones it could not, and the ones you did not bother asking because you assumed it would not know.
What do you wish it had a page on?A concept you keep explaining to the client, a method you keep looking up.
Where does it go wrong?A confident wrong answer, an answer with no source, a source that was out of date.
What you know that the documents do not say
0:45 to 0:51
A notebook holding only the engagement's documents can only reflect those documents back. Competently, and with no point of view. This is the material that gives it one.
What is the real question behind this engagement, as distinct from the scope's wording?
History, previous attempts, what people are nervous about, the data that exists and the data that does not, where the organisational tension sits. None of that is in a document, and all of it changes what a good answer looks like.
Anything commercially candid stays out of a shared notebook: pricing positions, resourcing, views of individuals.
What should run without you
0:51 to 0:57
Everything the notebook cannot do by itself is done by hand today: the weekly distillation, saving registers back in, re-uploading the handoff. Some of that could become mechanism later. This question tells us which.
If one of the manual steps ran on its own the moment a new document arrived, which one would you choose, and what would you need to see before you trusted it?
The answer is not a commitment. It is the evidence for whether a more automated tool is worth asking for, and what it would have to prove first.
After the session
0:57 to 1:00
The transcript fills the grid. The prompts are revised from it.
The recording is transcribed and each cell of the grid is filled from what was said.
Each existing prompt is checked against its row. Where the wording differs from yours, the prompt changes. Where a failure was named, a check is added.
New prompts only where a row asked for one.
New file types only if a row calls for one: shape descriptions for outputs, a log of questions asked, a benefits register, meeting transcripts as short-stay sources.
The revised library comes back with a change note per prompt.
What stays with a person
Judgement, every stakeholder-bound figure, every deletion from the notebook, and the weekly pass itself.