Result Interpretation
The Decision Console always returns one of three answer shapes, depending on the question type you picked. This page walks through real examples of each, using actual results from this app, so you know exactly what the numbers mean.
1. Yes / No (noul)
You get back a single number between 0 and 1 — the probability the answer is “yes.” There is no separate confidence score for Yes/No: since there are only two possible outcomes, that one number already tells you everything. Read it as a percentage.
Situation: “54-year-old male, sudden chest tightness and shortness of breath, pain radiating down the left arm, started 20 minutes ago.”
Question: “Does this patient need immediate emergency care?”
How to read it: 0.95 is close to 1, so this is a confident YES — the model is 95% sure emergency care is needed.
Situation: “The app crashes every time I try to upload a photo.”
Question: “Is this a bug report?”
How to read it: 0.99 is as close to certain as it gets — treat this as a confirmed YES.
Situation: “The food was great but we waited 45 minutes for our table even though we had a reservation.”
Question: “Is this customer satisfied overall?”
How to read it: 0.44 is close to 0.5 — that's not a weak YES or NO, it's the model telling you the signal is genuinely mixed (good food, bad wait). Don't round this to a clean answer.
2. Pick one option (choice)
You get a choice (the single best-matching option) plus a probabilities breakdown across every option you listed. The bars always add up to 100%. A single dominant bar means high confidence; several bars with meaningful weight means the model saw real overlap between options.
Situation: “My running shoes arrived in the wrong size. Can I swap them for a size 10?”
Question: “Which team should handle this?”
Best match: returns
How to read it: 100% on one option with 0% everywhere else is as confident as this gets — an unambiguous case.
Situation: “I've had a mild sore throat and a runny nose for two days, no fever.”
Question: “What level of care does this patient need?”
Best match: self_care
How to read it: self_care clearly wins, but notice urgent_care and emergency aren't exactly zero — the model is confident, just not absolutely certain. That small residual is normal and still counts as a strong result.
3. Rate on a scale (score)
You get a score (a weighted-average position across your levels — e.g. a 3-level scale runs 0 to 2), a confidence value, and probabilities per level. The score is a summary; the probability bars show you why — always check both together.
Situation: “Patient twisted their ankle during a run. It's swollen and painful to stand on, but they can still wiggle their toes and there's no visible deformity.”
Question: “How severe is this injury?”
Score: 0.97 (out of 2)
How to read it: Score 0.97 sits right next to level 1 (Moderate) — and the 90% bar on Moderate confirms it directly. High confidence, clear verdict: Moderate.
Situation: “The export button crashes the settings page in Safari. It works in Chrome, but a few of our customers only use Safari.”
Question: “How severe is the reported issue?”
Score: 1.43 (out of 2) · Confidence: 0.35
How to read it: Score 1.43 lands between levels 1 and 2 because the probabilities really are split (57% / 43%) — and confidence is only 0.35, the model's own signal that this is a genuine borderline call, not a clean one. Compare this to the injury example above: same scale shape, very different certainty.