How well Haleday predicted this profile's days — and whether those predictions were honest. Built from what was forecast vs what you actually logged.
Demo profile — these prediction outcomes are seeded for illustration, not a real person's tracked history. Sign in to see this built from your own results.
better than just guessing your average, over the last 21 predictions
fair · 0 = perfect, 0.25 = a coin flip
21 in the scoring window
Each point is the score recomputed over this profile's predictions up to that day — so you can see whether Haleday is getting better at reading this profile's days, not just today's number.
From -10% to 35.7% — a net +45.7 pts since the score became meaningful. The dashed line is the guess-your-average mark; above it, Haleday does better than simply guessing your average.
When the model forecasts a probability, how often does that actually happen? Closely matching bars mean it's well calibrated; a big gap means it's over- or under-confident in that range.
For forecasts between 40% and 60% (n=10), the model predicted 51% on average and the outcome happened 0% of the time.
For forecasts between 60% and 80% (n=11), the model predicted 70% on average and the outcome happened 64% of the time.
Patterns read deterministically from this profile's seeded logs — thumbs, repeated meals, and training-session effort. No AI guesswork; each line cites its evidence.
What Haleday holds onto about this profile — today, avoided cuisines steer meal ranking, and every principle shapes AI plan explanations when live AI is enabled. A principle graduates once its pattern has shown up 3+ times, at most 7 stand at once, and each fades once its own pattern goes 60 days without reinforcement — all by deterministic rules. No AI model ever writes these.
| # | Predicted | Actual | Hit? |
|---|---|---|---|
| 19 | 45% | 51% | — |
| 20 | 48% | 47% | — |
| 21 | 48% | 54% | — |
| 22 | 51% | 57% | — |
| 23 | 54% | 59% | — |
| 24 | 56% | 65% | — |
| 25 | 61% | 72% | ✓ |
| 26 | 66% | 73% | ✓ |
| 27 | 70% | 83% | ✓ |
| 28 | 76% | 80% | ✓ |
Computed deterministically from logged prediction outcomes (no LLM). The Insight Score is how much the model beats always-guessing-the-average; Brier measures whether its stated probabilities are honest. Demo profile — outcomes are seeded for illustration.
For forecasts between 80% and 100% (n=7), the model predicted 84% on average and the outcome happened 100% of the time.