# Findings AI Skill Measure Area · Findings Block · Decision Map · Oct 23, 2026 (v1.1) --- ## 1. What the Skill Does The Findings skill helps teams call the hunch, then turn the results into a design signal and a next move. It is the last block in the Measure area of Glare's Decision Map. Findings judge **usefulness**: did the design help people do what they came to do? They own **Use** signals, which show whether people can understand the design, find their way, and finish the task. If a test also turns up a Need or Prefer signal, the skill records it and passes it on: Need to Define → Situations, Prefer to Focus → Bets. Every finding opens with a verdict on the hunch. | Verdict | When to use it | |---|---| | Supported | The metric moved the way the hunch predicted, on evidence strong enough for the decision | | Challenged | Results were mixed, or pointed somewhere else | | Refuted | The results showed the hunch was wrong | | No call | The test couldn't answer the hunch; go back to Testing | Then it writes the design signal from four parts: the **behavior** observed, the **metric** that captured it, the **context** that explains it, and the **direction** it points. A metric alone is not a signal. ### The Verdict-First Rule Call the hunch before writing anything else, and weigh the call by the evidence rating in the Test Plan. A Use call needs evidence of use, such as clicks, task success, or errors. A call made only on the leading test is provisional until the lagging check comes in after launch. A single test ends in one of five decision types: | Decision Type | Use it when | |---|---| | Implement | Supported, on evidence strong enough for the size of the change | | Refine Design | The direction is right, but this version isn't strong enough yet | | Test Iteration | The evidence is useful but can't settle the call | | Revisit Later | The idea has potential, but the timing or context is wrong | | Do Not Pursue | Refuted, or not worth more investment | When the team wants to compare options, versions, or segments instead, the finding goes to Focus → Variables. A single test never goes straight to Bets. ## 2. Business Benefit Data is not a signal. Without findings, teams drown in dashboards and argue about what the numbers mean. With them, every result ties back to the hunch it tested and the need and goal behind it. This helps teams: - know what each test actually settled - see who a result holds for, not just the average - connect a result to a user need and a business goal - leave every test with a clear next move ## 3. Skill Output When used correctly, the skill produces a finding. The example below shows how this works for a mobile banking dashboard. | Field | Example Output (Mobile Banking Dashboard) | |---|---| | Hunch | Moving balance above the promo cards will raise first-click success on balance for habitual users | | Verdict | Supported (provisional, leading test only): 79% first-click success on balance; Moderate evidence | | Who It Holds For | Habitual users overall; users who rely on shortcuts took longer to find transfers | | User Need | Basics: Findable | | Business Result | Fewer sessions ending without an action, toward reducing session abandonment | | Design Signal | Behavior: habitual users go straight to their balance. Metric: 79% first-click success. Context: balance now sits above promotions. Direction: keep balance first and find a home for shortcuts. Type: Use | | Next Move | The team wants to compare against the current home screen before committing → Focus → Variables | | Lagging Check | Sessions ending without an action, 60 days after launch | | Failure Mode to Watch | Presenting the 79% as proof the live product improved. It is a leading result until the lagging check comes in. | | Next Step Handoff | → glare-focus-methods (Variables) to set up the comparison, or Implement if no comparison is needed | ## 4. Prompt Strategies The prompts below show different ways to use this skill. Each example uses a mobile banking dashboard update. ### Prompt 1: Verdict Entry: Call the hunch *"Our Helio first-click test on the redesigned mobile banking home screen came back at 79% success on balance, with 100 habitual users. Using the glare-measure-findings skill, call the hunch, say how sure we can be, and write the design signal."* **Why this works:** It opens with the verdict and the evidence strength, so nobody overstates what one test proved. **Best for:** - reading a new test result - writing up a round for the team ### Prompt 2: Split Entry: Find who it holds for *"The redesigned mobile banking home screen tested well overall, but some users took longer to find transfers. Using glare-measure-findings, split the result by audience, say who the design is useful for, and name the next move."* **Why this works:** Averages hide splits. Naming who a result holds for often matters more than the overall number. **Best for:** - mixed or surprising results - products with very different user groups ### Prompt 3: Next-Move Entry: Choose what happens next *"Our hunch about the mobile banking balance was supported, but the team isn't sure whether to build it, refine it, or compare it against the current screen first. Using glare-measure-findings, pick one of the five decision types or route us to Focus, and explain why."* **Why this works:** Every test ends with a named next move, so the result doesn't sit in a deck. **Best for:** - teams stuck after a good result - deciding between building and comparing --- *Glare Framework · glare-measure-findings · Measure Area. Handoffs: glare-measure-questioning (Testing) · glare-measure-hunches · glare-focus-methods (Variables) · glare-define-situations (Need signals) · glare-lead-results (lagging check) · glare-design-signals*