INFER / MECHANISM
How INFER works
A protocol for costly probabilistic claims, on-chain belief revision, and measurable judgment.
Section 01
01 / premise
Most information institutions make speech cheap.
That has advantages. It also makes it difficult to distinguish confidence, repetition, status and persuasion from useful information.
INFER tests a narrower mechanism: what changes when stating a probabilistic belief requires a small, explicit economic commitment?
← scroll diagram →
Section 02
02 / lifecycle
The mechanism links a resolvable question, a probability and an economic commitment in one auditable sequence. Select a stage to inspect its function.
← scroll diagram →
forecast / A question must be objectively resolvable under a rule published in advance.
The chain does not decide which forecast is correct. It preserves what was claimed, when it was claimed, and what economic commitment accompanied it. Reality supplies the final comparison.
Section 03
03 / committing a forecast
Illustrative example / question
Will event X resolve YES before September 30?
- Alice's estimate
- 72%
- Forecast cost
- 1 INFER
- Wallet
- 7Gxa...91Df [synthetic]
ALICE / WALLET
│ selects
↓
72%
│ BACK THIS BELIEF
↓
1 INFER COMMITMENT
The commitment does not make Alice's forecast true. It creates a measurable cost for submitting the claim.
One wallet may forecast multiple times. Each revision costs another INFER. Repeated payments do not create repeated voting weight.
Section 04
04 / transaction anatomy
A valid submission combines two instructions inside one Solana transaction: an SPL token transfer and a structured forecast memo.
- signature
- 5JxK...4Pa [synthetic]
- slot
- 327,184,991 [example]
- timestamp
- 23 Aug 2026 21:07 UTC [example]
source
Alice INFER ATA / synthetic
amount
1.000000000 INFER
destination
INFER Treasury ATA / synthetic
INFER|v1|market=MKT001|p=72
STATUS confirmed [illustrative]
latest valid forecast transaction / No live transaction available yet.
← scroll diagram →
INFER validates commitment and claim together.
Section 05
05 / aggregating beliefs
Each wallet contributes exactly one current forecast. Its latest valid forecast replaces its previous forecast for aggregation purposes.
Illustrative dataset
Equal-wallet aggregate
Pmarket = (72 + 55 + 81 + 60) / 4
67%
t1 / 72%
t2 / 68%
t3 / 61% ← CURRENT
Alice 61% + Bob 55% + Charlie 81%
NOT 72 + 68 + 61 + 55 + 81
Historical forecasts remain on-chain. Historical influence does not.
Wallet A
balance / 100,000 INFER
forecast / 70%
contribution / 70% × 1
Wallet B
balance / 5 INFER
forecast / 60%
contribution / 60% × 1
INFER balance does not multiply epistemic weight.
$INFER is a commitment asset. It is not a truth-weighting asset.
Section 06
06 / belief revision
Changing one's mind is not an error. It is information. Every revision changes current derived state while preserving prior claims.
1 / Epistemic history
How a forecaster's probability changed.
2 / Economic history
How often they were willing to pay to revise it.
belief state at time t = latest valid forecast before t
Section 07
07 / resolution
In version 0.1, questions use predefined resolution criteria. Resolved outcomes are confirmed manually in market configuration according to those published criteria. A later design can use deterministic on-chain data or oracle-based resolution.
Illustrative example
- Question
- Will X exceed Y at time T?
- Reference
- Source S
- Observation
- X = 241
- Threshold
- Y = 250
- Result
- NO / 0
Section 08
08 / judgment
BS = (p − o)²
Forecaster A
Forecast 80%
Outcome YES
(0.80 − 1)² = 0.040
Forecaster B
Forecast 55%
Outcome YES
(0.55 − 1)² = 0.203
A was better calibrated on this question. Lower Brier Score is better. A single good prediction does not establish skill; judgment becomes informative across repeated resolved forecasts.
BSi = (1 / N) Σt=1N (pit − ot)²
Accuracy asks
Were you right?
Calibration asks
Did your probabilities correspond to reality?
Section 09
09 / role of $INFER
Why not submit forecasts for free?
They could be. INFER intentionally tests a different institution. The token creates a standardized economic cost for submitting or revising a probabilistic claim.
- ├── forecasting rewards
- ├── information experiments
- ├── treasury
- └── contributors
├──── 1 INFER ────→ TREASURY
└┄┄ 72% belief ┄┄→ AGGREGATE
──── value transfer
┄┄┄ information
What $INFER does not do
The token creates commitment. Reality still determines outcomes.
Section 10
10 / system architecture
The application constructs evidence on Solana, then derives current belief and judgment state without overwriting the event history.
← scroll diagram →
Forecast {
signature
marketId
wallet
probability
amountCommitted
timestamp
slot
}Market {
id
question
closesAt
resolutionRule
status
resolvedOutcome
}CurrentBelief {
marketId
uniqueForecasters
aggregateProbability
totalCommitted
}Forecast and token transfer are raw chain evidence. CurrentBelief is derived state.
The protocol does not overwrite history. It derives current state from immutable events.
Section 11
11 / limits
INFER is an experiment, not a solved information institution. Its weaknesses should remain inspectable.
| PROBLEM | CURRENT MVP RESPONSE |
|---|---|
| Sybil wallets | Not solved |
| Wealth selection | Explicitly measurable |
| Poor question wording | Published resolution rules |
| Thin participation | Show forecaster count |
| Repeated forecasts | Latest / wallet only |
| Fake memo | Verify token transfer |
| Fake payment | Verify mint + treasury |
| Oracle ambiguity | Manual predefined resolution |
| Wallet ≠ person | Explicit limitation |
| Collusion | Not solved |
| Token speculation | Not required for scoring logic |
Sybil resistance
One wallet is not one human. Repeated submissions cannot add weight, but one person can create several wallets. Version 0.1 does not solve this.
Wealth selection
A cost may select for willingness or ability to pay rather than information. This is a hypothesis to measure, not assume away.
Token speculation
Price volatility changes the real forecast cost. A future version could denominate cost against a stable reference.
Section 12
12 / hypotheses
Adding an explicit economic cost does not improve forecast quality.
A small cost reduces low-information submissions and improves aggregate calibration.
Costs mainly select for wealth or conviction, not information.
Belief revision frequency contains information about uncertainty and new evidence.
Public forecasting histories create a useful measure of judgment over repeated questions.
INFER is useful even if some hypotheses fail. A mechanism that can be measured can also be rejected.
The question is empirical.
Section 13
13 / where this could lead
Future mechanism research / not currently implemented
INFER begins with a deliberately narrow primitive: record costly probabilistic claims and measure them.
A future version could introduce market making, information subsidies, conditional estimates and decision markets. Those mechanisms should be added only when the simpler experiment is understood.
Instead of: Will proposal A pass?
E[metric | A passes]
VERSUS
E[metric | A fails]
The first predicts politics. The second estimates consequences.
One-screen summary
INFER in one diagram
← scroll diagram →
Opinion ↓ Cost ↓ Record ↓ Aggregate ↓ Resolve ↓ Measure
That is INFER.
What does INFER claim?
Not that costly forecasts are necessarily better.
Not that token holders know more.
Not that markets automatically discover truth.
Only that probabilistic claims can be made explicit, costly, persistent and measurable.
Whether this produces better information is an empirical question.