Reference
Templates and checklists for problem framing
Part of the Problem Framing skill · loaded on demand from SKILL.md
Copy-ready structures. Placeholders in angle brackets are allowed here and nowhere else in this skill. Read this when you are producing the deliverable and want the exact shape of each output, or when you need the checklists to review a framing pack someone else built.
Contents
- Mandate capture and SCQ rewrite
- Problem-statement worksheet
- Issue tree (text form)
- MECE check per level
- So-what and prioritisation table
- Hypothesis register
- Day-one answer (SCR)
- Analysis workplan
- Partner review checklist
- Hand-off notes to neighbouring skills
1. Mandate capture and SCQ rewrite
## Mandate as received
Source: <who said it, when, in what setting>
Verbatim: "<exact words>"
What it contains: <Situation / Complication / Question / Answer: tick which are present>
What it assumes: <any solution or cause smuggled into the wording>
## Rewritten as SCQ
Situation: <one or two sentences the decision-maker would accept as true without argument>
Complication: <what changed, with the number that proves it>
Question: <one sentence, in the decision-maker's terms, that the rest of the work will answer>
Test before moving on: read the Question to the decision-maker (or imagine doing so). If they would say "yes, that is what I need to know", proceed. If they would add "and also...", the Question is too narrow. If they would say "that is not quite it", it is the wrong question.
2. Problem-statement worksheet
## Problem statement: <client>, <date>
| Field | Entry |
|---|---|
| Decision-maker | <who decides; who recommends; who must sign off numbers> |
| Question | <the SCQ Question, verbatim> |
| Success criteria | <measurable outcome, target level, by when; what "good enough" sizing means> |
| Constraints | <what the decision-maker will not do; contractual, regulatory, timing, political> |
| Scope in | <areas the analysis will cover> |
| Scope out | <areas explicitly excluded, with the reason> |
| Time frame | <deadline for the answer; horizon of the plan> |
| Required accuracy | <plus or minus X percent; decision-grade vs budget-grade> |
| Known positions | <what each stakeholder already believes; marked as hypotheses> |
| Assumptions made | <anything you filled in because the client was unavailable> |
Every field must be filled. "To be confirmed" is acceptable only in Known positions and Assumptions made.
3. Issue tree (text form)
## Issue tree: <root question>
Decomposition dimension at level 1: <identity / process / category / time>
- A. <branch>
- A1. <sub-branch>
- A1a. <analysis-level issue>
- A1b. <analysis-level issue>
- A2. <sub-branch>
- B. <branch>
- B1. <sub-branch>
- B2. <sub-branch>
Check at each level: <addition check with numbers, or classification check with three examples>
Label the tree type in the heading: diagnostic (why), solution (how), or hypothesis (assertions). Never mix question nodes and assertion nodes in one tree.
For quantitative trees, show the arithmetic beside each branch:
- Root: <metric> moved from <a> to <b>, change <c>
- A. <driver 1>: <contribution>
- B. <driver 2>: <contribution>
- C. <driver 3>: <contribution>
Sum of contributions: <d>; residual vs <c>: <e> (acceptable if under 5 percent)
4. MECE check per level
Run for every level of the tree. Record the result in the framing pack.
| Level | Dimension of the cut | ME test (can one item land in two branches?) | CE test (name three items; each lands somewhere?) | Count (2 to 5?) | Pass / fix |
|---|---|---|---|---|---|
| 1 | <e.g. P&L identity> | <example checked> | <three examples> | <n> | <pass or what was changed> |
| 2 under A | | | | | |
| 2 under B | | | | | |
Common fixes: split a mixed-dimension level into two levels; merge two overlapping branches and re-split on a single dimension; replace a brainstormed list with an identity; size any "other" branch and keep it under 10 percent.
5. So-what and prioritisation table
| Branch | Rough impact (<unit of success criterion>) | How sized | Ability to influence inside constraints (H/M/L) | So-what: what would the decision-maker do differently? | Decision (Priority n / Watch / Prune) | Reason recorded |
|---|---|---|---|---|---|---|
Rules:
- Size in the unit of the success criterion, to one significant figure. Ranking is the aim, not precision.
- A branch with "nothing" in the so-what column is pruned regardless of size.
- Every prune has a written reason so the branch can be revived.
- Prioritised branches should sum to more than the target (aim for 120 to 150 percent) because analysis will erode some.
Prioritisation matrix, for the pack:
| | Low influence | High influence |
|---|---|---|
| High impact | Watch: <branches> | Work first: <branches> |
| Low impact | Prune: <branches> | Prune unless trivial: <branches> |
6. Hypothesis register
One row per surviving branch. Each hypothesis is an assertion with a number and a condition, not a topic.
| ID | Hypothesis (assertion, with magnitude) | Load-bearing assumptions | Evidence that would refute it | Linked branch |
|---|---|---|---|---|
| H1 | <e.g. At least $X of Y is achievable through Z without losing more than W> | <two to four lines: what the hypothesis rests on; recorded, not tested here> | <the observation that would kill it> | <A1b> |
The load-bearing assumptions column flags what each hypothesis depends on so that hypothesis-stress-testing can turn them into refutation conditions and test the cheapest first. Do not expand the column into a test plan; the workplan (section 8) already holds the analyses.
Include at least one negative hypothesis where a stakeholder has a strong prior ("Branch closures are not the primary answer"). Stating it as a hypothesis keeps the stakeholder's view in the tree without accepting it.
7. Day-one answer (SCR)
## Day-one answer, <date>, confidence: <low / medium / high>
Situation: <one or two sentences>
Complication: <what changed, with numbers>
Resolution: <current best thinking on what to do and roughly what it is worth, stated as a hypothesis>
What we do not yet know: <the two or three gaps that most affect the Resolution>
Next refresh: <date>
Refresh at every checkpoint. Keep the previous versions; the sequence shows the client how the answer evolved and protects against the accusation that the answer was decided on day one.
8. Analysis workplan
| # | Issue | Hypothesis | Analysis needed | Source (data, person) | Owner | Due | Status |
|---|---|---|---|---|---|---|---|
| 1 | <question from the tree> | <H-id> | <specific analysis; what output it produces> | <named extract or named person> | <named person> | <date> | <not started / in progress / done> |
Rules:
- One row per analysis, not per branch; a branch may need three analyses.
- "Analysis needed" states the output ("margin distribution by rep, FY23 vs FY25"), not the activity ("look at pricing").
- Every source is a named extract or a named person. "Client data" is not a source.
- Detailed rows for the next two weeks; outline rows beyond that.
- Include a synthesis row near the end that checks the initiatives add up and do not double-count.
Checkpoint schedule:
| Date | Checkpoint | Attendees | What is shown |
|---|---|---|---|
| <date> | One-day answer refresh | <decision-maker> | Updated SCR, open questions |
| <date> | Steering committee | <group> | Tree, prioritisation, early findings |
| <date> | Draft deliverable | <decision-maker> | Full storyline |
9. Partner review checklist
Run on the complete framing pack before anything goes to the client.
- [ ] The Question is one sentence, in the decision-maker's terms, and they would recognise it as theirs.
- [ ] The Situation contains nothing the decision-maker could dispute.
- [ ] The Complication contains a number.
- [ ] Every worksheet field is filled; assumptions are labelled as such.
- [ ] Success criteria are measurable and include the required accuracy.
- [ ] The tree type is labelled (diagnostic / solution / hypothesis) and nodes are consistent with it.
- [ ] Level 1 is cut on a single dimension, ideally an identity, and the branches sum or classify cleanly.
- [ ] Each level has two to five branches; any "other" is sized and under 10 percent.
- [ ] Every branch passed or failed the so-what test in writing.
- [ ] Prioritised branches sum to at least 120 percent of the target.
- [ ] Every pruned branch has a recorded reason.
- [ ] Every hypothesis has a magnitude, load-bearing assumptions, and a refuting observation.
- [ ] At least one stakeholder prior is captured as a hypothesis rather than a fact.
- [ ] The workplan has a named owner, named source, and date on every row, and a synthesis row.
- [ ] The day-one answer can be read aloud in under a minute and would not embarrass you if it turned out wrong.
- [ ] The pack is dated, and the next refresh date is set.
10. Hand-off notes to neighbouring skills
## Hand-off: problem-framing to <skill>
Pack version: <date>
What is attached: <problem statement / tree / prioritisation / hypothesis register / workplan>
What the receiving skill should do with it: <one sentence>
Open questions the receiving skill should resolve: <list>
Constraints they must respect: <copied from the worksheet>
Use one note per receiving skill. The most common hand-offs: hypothesis register and workplan to hypothesis-stress-testing; P&L bridge and initiative list to value-driver-modeling; interview rows of the workplan to primary-research-synthesis; document-review rows to document-synthesis; external-market rows to market-competitive-intelligence; day-one answer to executive-communication.