Find out if your 80% means 80%.
A calibration journal for your predictions. Log a falsifiable claim with a confidence percentage, resolve it hit or miss later, and let pure math tell you whether your confidence actually tracks reality — superforecaster-style calibration, in your pocket.
The honest-measurement story
Everyone thinks they're well calibrated. Almost no one checks. When you say you're 80% sure, the only way to know what that number is worth is to gather every 80% call you've ever made and see how often they came true. If it's around 80%, your confidence is honest. If it's 55%, it isn't — and that's useful to know.
Reticle keeps the record that makes the check possible. Every statement is sealed the moment you log it, so you can't quietly move the goalposts once you see how things turned out. Nothing here interprets, softens, or spins the number. Your calibration is just arithmetic on your own resolved calls — and arithmetic doesn't flatter you.
How an entry works
- Capture in about five seconds. A claim, a confidence slider, a resolve-by date, and an optional tag. That's the whole form.
- The slider knows a trick. It reads 1–99 but stores a canonical 50–99. Drag below 50 and it flips to logging the negation — “this will NOT happen — 65% sure.” Being 30% on X is the same information as 70% on not-X, so it files it that way.
- Resolve from the inbox. Due, open, and resolved views, with honest one-tap exits. Mark it hit, miss, or void when the outcome is in.
- The record stays fixed. Statements are immutable — no retroactive fudging. Your calibration curve is built from what you actually said, when you said it.
Features
Five-second capture
Claim, confidence, resolve-by, optional tag — then get on with your day.
Canonical confidence
Sub-50 drags flip to the negation, so every call lands in one honest bucket.
Immutable statements
Once logged, a prediction can't be edited. A prediction you can rewrite isn't one.
Voids, counted
Unresolvable calls are excluded from scoring but still tallied — so sloppy questions can't hide.
Streaks reward logging
The streak counts predictions made, not hits — you can't sandbag easy calls to protect a number.
No confetti on a hit
Celebrating hits trains overclaiming. Reticle stays quiet either way.
Resolve inbox
Due, open, and resolved, with one-tap exits and a nudge for forgotten pendings.
Imports what you have
Bring your history in from Fatebook, PredictionBook, or CSV.
How the score is made — all pure math
- FREE FOREVER
Hit rate — hits ÷ scored. The plain fraction of calls that came true. - AT ≥10 SCORED
Brier score — the mean of (your probability − outcome)², taking confidence as a 0–1 probability and outcome as 0 or 1. 0.25 is coin-flipping; lower is better. Shown once you have enough resolved to mean something. - AT ≥20 SCORED
Calibration curve — decade buckets from the 50s to the 90s: what you said versus what actually happened, with the gap shown for each bucket. - THE VERDICT
Plain language — names your widest honest gap: “Most overconfident in the 70s: you say 70–79%, but hit 45% there.” Or simply well calibrated, if every gap is small.
No AI, on purpose
No AI here on purpose — your calibration curve is your data, not a model's guess about your data. A calibration score is trustworthy precisely because it's un-arguable arithmetic on your own record, with no model in the loop anywhere. There's nothing to argue with, and nothing to talk into a better number.
Privacy
Reticle runs entirely on your device — no accounts, no analytics, no servers. Optional Pro iCloud sync moves your record through your own private iCloud, and nowhere else.