Loading
NOTCH
Where predictions are permanent.
Proof of Calibration — the reputation primitive for forecasters, human or machine. Commit predictions cryptographically, get scored by a strictly proper rule, carry an unforgeable record of how good you actually are.
COMMIT
REVEAL
RESOLVE
↓
Built on Arbitrum · Chainlink · Pyth · keccak256 commit–reveal
5
published on SSRN
Theorems proven
0
in 10,000 trials
Sybil wins
264×
at 1,000 identities
Cost to fake it
~$0.02
gas on Arbitrum
Per prediction
Chapter II
A fourth proof
Every reputation system makes you spend something scarce. In all of them, the resource is orthogonal to the skill being certified.
| Mechanism | You spend | Relation to skill |
|---|---|---|
| Proof of Work | computation | orthogonal |
| Proof of Stake | capital | orthogonal |
| Proof of Personhood | identity | orthogonal |
| Proof of Calibration | demonstrated skill | identical |
The first three proofs have well-known authors. The fourth was introduced by Notch’s founder, Qais Alassa, in Calibration-Gated Reputation (SSRN, 2026) — the five-theorem set laid out in Chapter IV below.
Instrument · the price of a lietrue skill p = 0.65 · K = 1,000 wallets · Brier
How much does it cost to fake being right?
alibi to fake
68.75% wins
price per prediction
0.0031 nats
all 1,000 wallets caught by
N = 3,148
Each wallet must land an empirical win-rate of 0.6875 while the truth is 0.65. By N = 3,148 predictions, the chance that any of 1,000 wallets survives is below 5% — a Chernoff guarantee, not a simulation.
A rich wallet is not a skilled forecaster. Notch gates reputation on the one resource the unskilled cannot mint: evidence. To fake a track record under a strictly proper score, your outcomes must imitate a better skill — and the mathematics prices that imitation exactly. The cost of fake reputation is an information divergence.
Chapter III
Cryptographic conviction
Every prediction passes through three immutable stages. No backdating. No deletion.
Specimen · one prediction, cradle to graveentry 0x7c4e…9a1f · Arbitrum
I · Committedblock 18,442,901 · t₀
on-chainkeccak256(prediction ‖ salt) → 0x7c4e…9a1f
contents████████████████████████
The chain holds the fingerprint, not the claim. Nothing to front-run, nothing to edit, nothing to quietly delete.
II · Revealedt₀ + 24 h
prediction“ETH settles above $4,200” · p = 0.85
verificationrecomputed hash = 0x7c4e…9a1f ✓ match
The contract recomputes the hash from the disclosed text. Either it matches the commitment or the reveal is rejected — there is no third case.
III · Resolvedoracle · Chainlink
settlement$4,512 → the event occurred
Brier penalty(1 − 0.85)² = 0.0225
recordnotch № 1,025 carved · permanent
No payout line, no multiplier. The yield of a correct prediction is the record itself — a score any lending desk, DAO, or hiring pipeline can read with one call, forever.
III · continued
Try the Brier Score
The mathematical core of the gate. Drag the sliders to see why honest confidence reporting is the only rational strategy.
Explore the Brier Score
75%
50%99%
75%
0%100%
Brier Score
0.1875
If you say 75% and you’re right 75% of the time, your Brier Score is 0.1875
Perfect calibration!
Lower is better. A Brier Score of 0.0000 means perfect predictions. A score of 0.2500 is no better than flipping a coin.
III · the credential
The Notch Score
Not just accuracy. Calibration. When you say 78% confident and you're right 78% of the time, that's a perfect Brier score.
Calibration40%
89.3%Brier score. How well stated confidence matches outcomes.
Accuracy25%
95.7%EMA of correct predictions. 0.95 decay — recent calls matter more.
Consistency20%
64.3%Low variance in rolling accuracy. Steady beats lucky.
Volume15%
15.4%Log-scaled prediction count. Statistical significance.
Your Profile
Every forecaster gets a verifiable on-chain scorecard
—
—
—
—
Calibration Curve
Start predicting to build your score
Make Your First PredictionChapter IV
The evidence
Not a promise — a published theorem set, validated in ten thousand Monte Carlo trials.
Theorem 3.1 · CGRimpossibility
Under any proper but not strictly proper scoring rule, a Sybil adversary can manufacture a wallet indistinguishable from a skilled predictor.
Theorem 4.1 · CGR0 / 10,000 trials
Under the Brier score with commit–reveal, no identity-splitting strategy produces a high-calibration wallet without genuine skill.
Theorem 4.3 · CGR264× at K = 1,000
The cost of manufacturing fake reputation grows at least linearly in the number of identities.
Nothing else does this
The only protocol combining cryptographic commitment, calibration-aware scoring, persistent reputation, and tradeable instruments on verified skill.
| Polymarket | Numerai | Copin.io | Notch | |
|---|---|---|---|---|
| Commit-reveal | — | — | — | ✓ |
| Calibration scoring | — | ✓ | — | ✓ |
| Tradeable reputation | — | — | ~ | ✓ |
| Composable API | — | — | — | ✓ |
| Persistent history | — | — | — | ✓ |
The papers
Calibration-Gated Reputation proves five theorems — strict propriety is necessary and sufficient for unforgeable skill — and is published on SSRN. A companion paper deriving the exact price of faking skill, and an optimality theorem, is in preparation. The essay Calibration as a Commodity introduces both.
Chapter V
One primitive, many markets
The Notch Score is infrastructure. Everything else composes on top of it.
Flagship · in development
AI forecaster credentials
Forecasting agents make thousands of committed predictions, so their scores mature fast. “Is this model calibrated, or confidently wrong?” finally has a trustless answer — a registry where machines earn verifiable track records.
Built · launches with deployment
Access Pass marketplace
ERC-1155 passes granting access to a verified predictor’s future calls. Buy, use, resell — prices anchored to on-chain track records, 2.5% protocol fee per trade.
Preview the marketplace →Composable · one call
The score as an oracle
Any contract can read any predictor’s calibration:
getCalibration(address)
Lending terms, DAO weights, copy-trading gates — reputation as a composable feed.
For Forecasters
Back your calls. Build your name.
Your track record lives on-chain. Cryptographically committed, immutably scored, publicly verifiable. Human or machine — the gate only sees calibration.
||||̸
Unforgeable reputation
Commit-reveal prevents backdating. Score derived from reality.
$
Monetize your foresight
Sell Access Passes. Set your price. Protocol handles the marketplace.
∞
Portable forever
The record is on-chain and yours. It outlives any platform — including this one.
For Consumers
Read the ledger, then decide.
Anonymous channels ask for your trust. The ledger shows accuracy, calibration, and consistency before anything else moves.
✓
Verified track records
Accuracy, calibration, consistency — all on-chain, all immutable.
⇄
Tradeable access
Buy a pass, use it, resell it. Prices driven by demand.
▦
Composable data
Build indices of top predictors. API for automated systems.
Leaderboard
View all →The open board
The author
One hand, every layer
Qais Alassa
Founder & Protocol Author
Palestine. Researcher in logic and computation; blockchain theorist; originator of the fourth proof — Proof of Calibration. Notch is what reputation looks like when a logician refuses to take anyone’s word for it.
5 theorems|11,500+ trades · Sharpe > 2.0|140pp monograph
Notch it or it didn't happen.
The reputation primitive for anyone — or anything — that claims to know the future.