Ivyra.
How it works

Private by default · No passwords, just a magic link

A journal that keeps score

A journal for what you think will happen — and how often you're right.

Every entry is scored against what actually happened. Over time, you learn exactly when to trust your own confidence — and where it runs ahead of you.

Your journal

Newest first

July

28 Jul80% · resolves 15/8

We ship the redesign by the 15th

Third week with no assets and I'm starting to think this isn't a delay so much as the plan…

21 Jul0.64

They come back with a better offer by end of month

Said 75%. They didn't come back at all. I keep pricing hope as if it were evidence…

14 Jul0.09

Review time drops below a day within a month of the trunk-based switch

The bottleneck was always batching, and small PRs merge on their own…

Ivyra measures whether your confidence means what you think it does — and shows you exactly where it doesn't.

Know when to trust yourself

Turn “I'm pretty sure” into a number you can check

Ivyra counts how often your predictions actually come true at each confidence level. So “I'm 85% sure” stops being a feeling and becomes a frequency — one you can hold up against reality, call after call.

Your calibration, in one line

When you say 85%, it happens 38% of the time.

UnderconfidentOverconfident

+47 points overconfident — your high-confidence calls land less often than you claim.

The calibration curve

See it across every confidence level

Each dot groups the predictions you made at a similar confidence, then plots that confidence against how often those calls actually came true.

  • The dashed diagonal is perfect calibration — where saying 70% means it happens 70% of the time.
  • Dots below the line mean it happened less often than you said; above means more often.
  • The closer your dots hug the diagonal, the more your confidence can be trusted.

Calibration curve

Sample data

This isn't your data yet — it's what a calibration curve looks like.

Before you commit

Your own track record, at the moment you predict

As you write a new entry, Ivyra finds the calls you've made like it before and shows how they actually turned out — a plain frequency, surfaced before you lock in your confidence. It states what happened and stops there.

Before you save

You've said 75% or higher on 6 calls like this. 2 landed.

The only read that fires before a call, not after it.

The record

It reads back like a journal

Every prediction is a dated entry in your own words: the claim, your confidence, and why you believed it — frozen the moment you save. Months later it's a timeline of your thinking, honest because the score won't let it drift.

Not a blank page — every entry is a prediction, anchored to a claim. The reasoning is optional: a sentence or a paragraph, and it's the part you'll read back.

28 Jul80% · resolves 15 Aug

We ship the redesign by the 15th

Why I think so

Third week with no assets and I'm starting to think this isn't a delay so much as the plan. Still, the team has pulled these in before — I want that to be true, so maybe I'm giving it more weight than I should.

Locks when you save.

Grounded in research

Built on what the research shows works

The loop Ivyra runs — make a call, see what happens, get a real score, adjust — is one of the most studied ways to improve calibration. A few of the findings we built on:

Training moves the needle

In a four-year, government-sponsored forecasting tournament, people who did a short probabilistic-reasoning exercise — under an hour — went on to make about 6–11% more accurate predictions than an untrained group, in every year of the study.

Chang, Chen, Mellers & Tetlock — Judgment and Decision Making (2016)

The feedback loop works in the wild

Weather forecasters are the classic case: their probability forecasts are strikingly well-calibrated — when they say a 70% chance of rain, it rains on about 70% of those days — a benchmark of what steady, scored feedback produces.

Murphy & Winkler — J. Royal Statistical Society, Series C (1977)

But a score alone isn't enough

A 2025 experiment found that showing people their calibration score and an over/under-confidence readout did not, on its own, improve their calibration. A bare number doesn't teach — which is exactly why Ivyra never stops at one.

Martin & Mandel — Futures & Foresight Science (2025)

These are findings about calibration in general, not a promise about your results. We lean on the mechanism the evidence supports — and because a bare score doesn't teach on its own, Ivyra pairs every number with plain-language interpretation.

Start your track record today

It takes about thirty seconds to log your first entry. In a few weeks, you'll know something about yourself most people never do: how often you're actually right.