Appearance
Outcomes and Strategy
Strategy helps the advisor test what may limit a flow of work. Outcomes tests whether a change worked. Every claim cites Company Repo files, and code sets every evidence word.
Where each part lives
| Part | Where in the app |
|---|---|
| Possible constraints | Opportunities › Possible constraints |
| Theory of Constraints exhibit | Opens from a possible constraint, under Opportunities |
| RACI exhibit | Work Map › RACI |
| Summary and SWOT / TOWS | Strategy › Summary and Strategy › Frameworks |
| Measures | Outcomes › Measures |
| Changes being tested, and the weekly review | Outcomes › Results |
Old links to the earlier Strategy tabs still open the new places.
Possible constraints
A possible constraint is a hypothesis about what may limit a flow. It stays a hypothesis until a test supports it. DutyGraph never calls it a bottleneck or a root cause.
Signals: code, no AI
Rules in code (R1 to R11) read the Repo and find signals for each flow. A signal points to Repo files. It is not a finding. Signals cost nothing and need no AI.
Find possible constraints: one AI call, after consent
- Select a flow. Read its signals: the rule that fired, the exact text and the files.
- Select Find possible constraints. Read the consent note. One click starts one AI call for this flow, through the company's approved AI connection (Claude first, else OpenAI).
- Read the result. At most 3 possible constraints show, and at most 1 is marked Leading. Each shows:
- why DutyGraph suggests it: the file, its evidence word and the exact quote;
- what points the other way, and other explanations;
- what would show it is wrong;
- the test: the measure, the baseline (or Not found) and the window;
- questions to named people. Send as ask drafts an ask. Nothing is sent until you send it.
- Select Accept… and choose the accountable owner. Or select Reject… and give a reason.
Each possible constraint has a key, C1, C2 and so on, one per flow. The same key shows on every exhibit, in Route and on Your team. When you accept, you can edit the statement and the test measure.
DutyGraph checks the AI text and your edits the same way. It refuses words such as "bottleneck" or "root cause", and any number that no quote states. When a cited file changes, the search reads Out of date and you must Find again before you accept. A rejected possible constraint is remembered for that flow, so the next search does not offer it again without a new signal.
Frameworks
Three frameworks run on the Repo as consulting diagrams:
| Framework | What it shows |
|---|---|
| RACI | Task Cards as rows, people and roles as columns. R and A come from each Task Card. A cell marked AI is a Consulted or Informed entry the AI added. |
| Theory of Constraints | The flow left to right. The step of the leading possible constraint has a thick border. Below it, the 5 focusing steps, each with a prediction and a stop condition. |
| SWOT / TOWS | Strengths, weaknesses, opportunities and threats, then the TOWS matrix that pairs them. |
Each framework shows Ready, Runs with gaps or Not ready, with what is missing and who to ask. Select Run and confirm the consent note. One run is one AI call. Select a box to see its evidence word, sources, quotes and who to ask. Accept marks it Reviewed by advisor. That is your review, not the client's confirmation.
The evidence words
Code sets each word. The AI does not.
| Word | Meaning |
|---|---|
| Client confirmed | A person at the company confirmed the value in DutyGraph. |
| Reported | A source states the value. |
| Inferred | Research suggests it. Nobody confirmed it. |
| Assumption | Nothing in the Repo supports it. |
| Disputed | Sources or people disagree. |
| Not found | DutyGraph has no usable value. The box names who to ask. |
Summary and Present
Strategy › Summary gives the answer first. Code builds every line from the possible constraints and the exhibits. It lists at most 3 decisions for the client, each with a person and a source file.
Present (or the P key, on a computer) shows 7 slides: the Summary, the leading possible constraint against its rival, Theory of Constraints, RACI, SWOT, TOWS and the decisions. The slides read the same data as the pages.
DutyGraph Agent
On the Strategy pages, DutyGraph Agent runs in Ask mode, docked on the right. It uses the company's approved AI. It reads the Strategy part of the Repo: the company, initiatives, findings, flows, handoffs, measures, departments and Task Cards. Each answer names the files it uses. In Ask mode it answers questions. It does not change a record. Its two buttons open Find possible constraints and a framework Run.
Limits on AI use
- Constraint searches and framework runs: 30 per company in 24 hours, together.
- DutyGraph Agent on Strategy: 60 questions per company in 24 hours.
- DutyGraph logs the tokens, the model and the estimated cost of each AI call, including a call that fails.
- The sample companies make no AI call. Their output says Proposed by code · sample or Sample output · no AI call.
Outcomes: measures and pilots
A change is tested against a measure. Code, not the advisor, sets the result word.
- Define the measure in Outcomes › Measures, or from a route card with Start the baseline…. A missing value stays Not found. Zero means a measured zero.
- Collect the baseline. Select Record observation for each reading. The baseline needs at least 10 runs over at least 14 days.
- Record the prediction before the pilot: low, high, unit and direction, plus a guardrail measure, a window, a stop rule and a review date. Route writes these when you accept a change.
- Run the pilot and record its readings.
- Read the result. Code sets Supported, Not supported or Inconclusive. A person can lower a result to Inconclusive, never raise it.
A supported change reads "Supported by 1 pilot", then the pilot dates, then "not proven". DutyGraph never says "proven".
Outcomes › Results lists each change being tested, with its prediction, what was observed on its measure, the latest outcome review and the next review date.
The weekly review
Outcomes › Results also prepares the weekly review:
- Ask for updates before the meeting drafts four questions for each person from their current Task Cards and measures. Code drafts them, with no AI. You can edit them and the due date.
- The update goes out through the normal request flow. Sending email needs Resend. Saving a draft sends nothing.
- A participant can type or record an answer. A recording is transcribed with OpenAI on demand, and the person checks the text before sending.
Scheduled invitations, importing a meeting transcript automatically and an AI-drafted meeting agenda are not built.
INFO
Status. Constraint signals, Find possible constraints, Accept and Reject, the RACI, Theory of Constraints and SWOT / TOWS exhibits, the Summary, Present, DutyGraph Agent, measures, the proof rule and the weekly update requests are Built. The sample companies show authored output with no AI call. Business Model Canvas, McKinsey 7-S, Industry Map, Five Forces and PESTLE on the Repo, scheduled weekly invitations, automatic transcript import and an AI meeting agenda are Planned.
Source: gitbook/product/opportunities/outcomes-and-strategy.md in the DutyGraph repository.