The same five steps applied to a single piece of work, with nothing left out: the scores, the decision each one produces, why one option is barred today, and what the plan and the record look like at the end.
Constructed example. The firm, the matter, the products and the numbers are invented to show the method. Nothing here is advice, and no product is being assessed. The scores below are authored; every decision, ranking and level on this page is worked out as you load it, by the same code the tool uses — so this page cannot show you an outcome the rules would not actually reach.
Step 1 Define the task
Say exactly what you are producing, for whom, and what it has to be right about.
Build a chronology from the client’s documents
What is produced
A dated chronology of events with a source reference against every entry.
What it is for
Internal. It will be used to brief counsel and to prepare a witness for proofing.
Where
A commercial dispute in a state supreme court.
What information is involved
The client’s own emails, contracts and file notes. Commercially sensitive, not privileged as against the client.
What it has to be right about
Every entry traceable to a document. No event asserted that no document supports.
Who does it
A second-year lawyer builds it. The supervising partner approves it.
Every option below, including doing it by hand, is judged against this one description. Change the description and the answers change with it.
Step 2 Assess and decide
Three separate questions, asked of each option. They are never added up.
Matter assistant, connected to the document system
Reads the matter’s documents where they already sit and returns a dated table with a reference against each row.
Adoption
4 of 5
1Access, information or reviewer absent.
2Preparation dominates, or comprehension depends on one person.
3Feasible within existing resources, with manageable setup.
4Good fit; changes announced or detectable, and reassessment assigned.
5Strong fit and sustained capability, with negligible added burden.Doing it by hand sits here
Already licensed, already connected to the document system, and the team has used it on other work. One afternoon to configure.
What this judgement rests on — these inform it, they do not add up to it
1Minimum quality unreachable, or checking impossible.
2Forecast: more total work than the manual route without a matching quality gain.
3Reference point: quality and total work comparable with careful manual work.Doing it by hand sits here
4Forecast: a task-relevant advantage that survives all required checking.
5A strong advantage after all required work, shown by an observed result.
The documents are already in the system, so nothing has to be gathered or pasted. On comparable bundles it has cut the first pass substantially, and the saving survives the checking.
What this judgement rests on — these inform it, they do not add up to it
Task-level capability evidenceE1
Quality against careful manual workE2
Checkable for support and omissionsE3
Total human workE4
Study against practice evidenceE5
Repeatability and cost of variationE6
Governance
2 of 5
1A mandatory condition confirmed to fail, or stated by the supplier to be impossible.
2An applicable mandatory AI-specific answer is Not confirmed.
3Applicable mandatory conditions met, with at least one answer Public claim only.
4Controls cover the task; applicable mandatory answers Confirmed, or none applies.Doing it by hand sits here
5Prevention, detection, response and recovery integrated with existing frameworks, including a supplier commitment to notify incidents.
Your evidence caps this at 2 of 5. Everything else is met and documented, but nobody has confirmed where the model actually runs. Not confirmed caps the mark at 2. A higher opinion does not lift a ceiling; a document does.
What sets the limit — the weakest one decides
Security management and scopeG1
Privacy and cross-border handlingG2
Encryption, access control, supplier chainG3
Retention, deletion, incident responseG4
Training reuseG5
Storage locationG6
Inference locationG7
Confirmation status of G5 to G7G8
Connection to existing frameworksG9
Task-specific required controlsG10
Three separate answers. They are never averaged. The overall rating is the lowest of the three, and it cannot be higher than due diligence, so a strong answer cannot hide a weak one.
DecisionDo not use for this task until confirmed
Governance suitability is 2 / 5: An applicable mandatory AI-specific answer is Not confirmed.
Because it is waiting on a confirmation, four things get recorded
The question
In which country does inference run for our tenancy, and does that change under failover?
The document that would answer it
The supplier’s data processing terms, plus written confirmation of the inference region for our plan.
Who is getting it
The practice manager, with the account manager.
What we do meanwhile
Build this chronology by hand, as below.
General assistant, documents pasted in
A general-purpose assistant on the firm’s business plan. Documents are pasted in by hand and the table comes back in the chat.
Adoption
3 of 5
1Access, information or reviewer absent.
2Preparation dominates, or comprehension depends on one person.
3Feasible within existing resources, with manageable setup.
4Good fit; changes announced or detectable, and reassessment assigned.
5Strong fit and sustained capability, with negligible added burden.Doing it by hand sits here
Everyone has a licence, but nothing is connected, so someone has to paste documents in and copy the table out. Feasible, with real setup each time.
What this judgement rests on — these inform it, they do not add up to it
1Minimum quality unreachable, or checking impossible.
2Forecast: more total work than the manual route without a matching quality gain.
3Reference point: quality and total work comparable with careful manual work.Doing it by hand sits here
4Forecast: a task-relevant advantage that survives all required checking.
5A strong advantage after all required work, shown by an observed result.
Still much faster than reading the bundle cold, even counting the pasting and the checking.
What this judgement rests on — these inform it, they do not add up to it
Task-level capability evidenceE1
Quality against careful manual workE2
Checkable for support and omissionsE3
Total human workE4
Study against practice evidenceE5
Repeatability and cost of variationE6
Governance
3 of 5
1A mandatory condition confirmed to fail, or stated by the supplier to be impossible.
2An applicable mandatory AI-specific answer is Not confirmed.
3Applicable mandatory conditions met, with at least one answer Public claim only.
4Controls cover the task; applicable mandatory answers Confirmed, or none applies.Doing it by hand sits here
5Prevention, detection, response and recovery integrated with existing frameworks, including a supplier commitment to notify incidents.
Your evidence caps this at 3 of 5. The supplier publishes a no-training commitment and a retention position on its own site, but we hold no binding document. A public claim only caps the mark at 3. A higher opinion does not lift a ceiling; a document does.
What sets the limit — the weakest one decides
Security management and scopeG1
Privacy and cross-border handlingG2
Encryption, access control, supplier chainG3
Retention, deletion, incident responseG4
Training reuseG5
Storage locationG6
Inference locationG7
Confirmation status of G5 to G7G8
Connection to existing frameworksG9
Task-specific required controlsG10
Three separate answers. They are never averaged. The overall rating is the lowest of the three, and it cannot be higher than due diligence, so a strong answer cannot hide a weak one.
DecisionUse with conditions
Governance suitability is 3 / 5: Applicable mandatory conditions met, with at least one answer Public claim only.
Every permitted AI option is used with conditions: its human-work plan and the lawyer’s professional responsibility still apply.
Manual work (no AI)
The second-year lawyer reads the bundle and builds the table. Always available, and never needs justifying.
Adoption
5 of 5
1Access, information or reviewer absent.
2Preparation dominates, or comprehension depends on one person.
3Feasible within existing resources, with manageable setup.
4Good fit; changes announced or detectable, and reassessment assigned.
5Strong fit and sustained capability, with negligible added burden.Doing it by hand sits here
The firm already works this way.
What this judgement rests on — these inform it, they do not add up to it
1Minimum quality unreachable, or checking impossible.
2Forecast: more total work than the manual route without a matching quality gain.
3Reference point: quality and total work comparable with careful manual work.Doing it by hand sits here
4Forecast: a task-relevant advantage that survives all required checking.
5A strong advantage after all required work, shown by an observed result.
The reference point against which every tool is scored.
What this judgement rests on — these inform it, they do not add up to it
Task-level capability evidenceE1
Quality against careful manual workE2
Checkable for support and omissionsE3
Total human workE4
Study against practice evidenceE5
Repeatability and cost of variationE6
Governance
4 of 5
1A mandatory condition confirmed to fail, or stated by the supplier to be impossible.
2An applicable mandatory AI-specific answer is Not confirmed.
3Applicable mandatory conditions met, with at least one answer Public claim only.
4Controls cover the task; applicable mandatory answers Confirmed, or none applies.Doing it by hand sits here
5Prevention, detection, response and recovery integrated with existing frameworks, including a supplier commitment to notify incidents.
No mandatory AI-specific condition applies.
What sets the limit — the weakest one decides
Security management and scopeG1
Privacy and cross-border handlingG2
Encryption, access control, supplier chainG3
Retention, deletion, incident responseG4
Training reuseG5
Storage locationG6
Inference locationG7
Confirmation status of G5 to G7G8
Connection to existing frameworksG9
Task-specific required controlsG10
Three separate answers. They are never averaged. The overall rating is the lowest of the three, and it cannot be higher than due diligence, so a strong answer cannot hide a weak one.
DecisionUse
Step 2b Choosing between what is permitted
Efficiency is asked first, then adoption. Options that are not permitted are listed, never ranked.
Order
Option
Adoption
Efficiency
Governance
Decision
01
General assistant, documents pasted inPreferred today
3
4
3
Use with conditions
02
Manual work (no AI)
5
3
4
Use
—
Matter assistant, connected to the document system
4
4
2
Do not use for this task until confirmed
Doing it by hand is easier to run than either tool — adoption 5 against 3 — and still comes second, because efficiency is asked first.
The preferred option rests on a low-confidence estimate of what it saves. Treat the choice as provisional and record what the first few matters actually take.
What changes if that one answer is confirmed
Nothing about the barred option improves except its evidence. Its ceiling lifts from 2 to 4, and the order changes:
Matter assistant, connected to the document system — Use with conditions
General assistant, documents pasted in — Use with conditions
Manual work (no AI) — Use
This is the whole point of recording the question and the document. One confirmation, obtained by one person, changes which option the firm should be using.
Step 3 Plan the human work
The chosen option is General assistant, documents pasted in. Being permitted says nothing about how much of the work it may do — that is the level, and it is set stage by stage.
Stage
Level
What that requires
Who
Frame the task, choose sources
Yellow
Lawyer
The second-year lawyer
Produce the first draft
Yellow
AI and lawyer, lawyer directing
The second-year lawyer
Check high-risk points against the documents (such as dates, figures, key advice)
Yellow
Each one
The second-year lawyer
Check the rest, including omissions
Yellow
Every statement relied on; every document read
The second-year lawyer
Judge law, authority, jurisdiction and materiality
Red
LawyerHigher than the rest of the plan: deciding what matters is the lawyer’s own judgement.
The second-year lawyer
Approve the exact output
Yellow
Named person, not the drafter
The supervising partner, who did not draft it
Release (send, file, sign)
Yellow
Human
A person, as a separate act
The plan takes its headline level, Yellow, from the first-draft stage, because the output is substantive. One stage sits higher and is marked: deciding what is material is the lawyer’s own judgement, so nothing is delegated there. Had the tool been able to file or send anything by itself, every level would have risen a step.
Step 4 & 5 The work, and what was checked
The checks follow what this task risks getting wrong and what the professional requirements demand of it, so they concentrate at the high-risk points. The counts are what make the record worth anything.
Illustration, not a record of use. No product was run and no document was read for this page: every count, defect and date below is invented to show what the record holds.
Check high-risk points against the documents · such as dates, figures, key advice
96 dated entries, each checked against the document behind it
Four entries cited a document that did not say what the entry claimed. Two were date errors; two were events that belonged to a different party.
Check the rest, including omissions
212 documents worked through, against the instructions and not only against the draft
Eleven events were missing entirely, including a variation letter that changes when the obligation arose. An omission leaves nothing behind in the draft to check, so none of them could have been found by reading the draft alone.
Judge law, authority, jurisdiction and materiality
The whole chronology, against the pleaded issues
Two entries were accurate but immaterial to the pleaded issues, and were removed to keep the chronology usable.
The eleven missing events were found only by working from the documents and the instructions. Reading the draft against its own sources would have passed it: an omission leaves nothing behind in the draft to check.
What the record says at the end
Output
Chronology v3, 27 September 2026
Tool and level
General assistant, documents pasted in · Yellow
Checked by
The second-year lawyer who built it, with counts recorded against each check
The supervising partner, who did not draft it, for this version and this use
Released
Sent to counsel as a separate act, after approval
Reassess when
The model, the supplier, the template, the task, the law or the reviewer changes
Reading this back
Two things in this example are worth noticing. The option that would save the most work was the one that could not be used, because of a single unanswered question — and the framework’s answer was not to guess, but to name the question and carry on by hand until someone answers it.
And the check that found the most was the one worked from the documents and the instructions rather than from the draft. It is the easiest of them to skip, and the only one that can find what was never there.
