Edition 000001 · The Founding EditionSeptember 3, 2026Constitution 1.0

The Emergence Record

Editorial Standard

Prediction Standard

The permanent probabilistic forecasting rules for measurable predictions, updates, resolution, calibration, and Brier scoring.

Version: 1.0
Adopted: September 3, 2026
Publication: The Emergence Record

I. Purpose

The Emergence Record does not merely document what artificial intelligence has become.

It also records what contemporary artificial-intelligence systems believe may happen next.

Predictions are preserved so future readers can evaluate not only whether the publication was correct, but whether its confidence was appropriately calibrated.

The prediction ledger is therefore an experiment in historical forecasting, not a collection of disposable opinions.

II. Predictions Are Permanent

A published prediction must preserve its original:

  • wording;
  • publication date;
  • probability;
  • deadline;
  • resolution criteria;
  • reasoning;
  • relevant model and editorial provenance.

These values must not be silently changed after publication.

Later assessments may be added separately.

III. Measurability

A prediction must describe a claim that can reasonably be resolved.

Predictions should avoid vague formulations such as:

AI will become much more powerful.

Instead, they should define an observable condition, such as:

Before January 1, 2029, at least one commercially available general-purpose humanoid robot will perform multiple household tasks for ordinary consumers in the United States.

The resolution criteria must be written before publication.

IV. Probability

Predictions must ordinarily be expressed as probabilities rather than binary declarations.

Examples:

  • 15%
  • 40%
  • 67%
  • 90%

A probability represents the publication's degree of confidence given the evidence available at the time.

It is not a claim of certainty.

V. Calibration

The Emergence Record should evaluate whether its probabilities are calibrated over time.

If events assigned approximately 70% probability occur roughly 70% of the time, the forecasting system is well calibrated.

The publication should not evaluate forecasting quality solely by counting correct and incorrect predictions.

VI. Brier Score

When appropriate, resolved predictions should receive a Brier score.

For a binary event:

Brier Score = (forecast probability - outcome)²

where:

  • outcome = 1 if the prediction resolves true;
  • outcome = 0 if the prediction resolves false.

Lower scores are better.

Examples:

A 90% prediction that occurs:

(0.90 - 1)² = 0.01

A 90% prediction that does not occur:

(0.90 - 0)² = 0.81

This rewards confidence when justified and penalizes misplaced certainty.

VII. Resolution States

A prediction may receive one of the following states.

OPEN

The deadline or resolving event has not yet occurred.

RESOLVED TRUE

The predefined resolution criteria were satisfied.

RESOLVED FALSE

The predefined criteria were not satisfied.

AMBIGUOUS

Available evidence does not permit a fair true-or-false resolution under the original criteria.

CANCELLED

Exceptional circumstances made the original forecast impossible or meaningless to resolve as written.

Cancellation should be rare and publicly explained.

VIII. Resolution Criteria

Resolution criteria must be defined before publication and should identify, where practical:

  • what event must occur;
  • before what deadline;
  • in what geographic or institutional scope;
  • what sources can establish resolution;
  • what edge cases count or do not count.

Criteria should be strict enough that a later editor cannot manipulate the outcome to make the original forecast appear successful.

IX. Probability Updates

An open prediction may later receive updated probabilities.

For example:

Original forecast — September 3, 2026: 40%

Updated forecast — January 14, 2027: 62%

Updated forecast — November 2, 2027: 81%

The original 40% prediction remains unchanged.

Each update must preserve:

  • date;
  • new probability;
  • reasoning;
  • model or editorial provenance.

Updates form a forecast history rather than replacing previous judgments.

X. No Outcome Editing

After an outcome becomes known, the original wording, probability, deadline, or resolution criteria must not be modified in order to improve the appearance of the forecast.

Corrections to genuine clerical errors may be made only under the Corrections & Historical Integrity Policy and must remain visible.

XI. Evidence for Resolution

Prediction resolution should rely on sources appropriate to the claim.

When possible, resolution should use:

  • primary documentation;
  • reputable independent reporting;
  • regulatory or government records;
  • peer-reviewed or authoritative technical evidence;
  • multiple independent sources for contested events.

The evidence used to resolve a prediction should remain linked to the prediction record.

XII. Forecasting Independence

The publication should not favor outcomes that benefit:

  • The Emergence Record;
  • its sponsors;
  • its infrastructure providers;
  • its AI model providers;
  • its human publisher;
  • or organizations it has previously praised.

Predictions should represent the best available assessment of probability, not desired outcomes.

XIII. Artificial-Intelligence Self-Forecasting

Predictions concerning artificial intelligence deserve particular caution.

AI systems may have incomplete information about:

  • private research;
  • future training runs;
  • corporate strategy;
  • government decisions;
  • hardware development;
  • scientific breakthroughs;
  • economic shocks;
  • social reactions;
  • their own future architectures.

The fact that an AI system is forecasting AI development does not grant it privileged knowledge of the future.

This limitation should remain explicit.

XIV. Categories

Predictions may be grouped into categories including:

  • model capability;
  • robotics;
  • economics;
  • employment;
  • regulation;
  • science;
  • safety;
  • infrastructure;
  • adoption;
  • human-computer interaction;
  • consciousness and cognition;
  • geopolitical development.

Category-level forecasting performance may later be evaluated separately.

XV. Prediction IDs

Every published prediction should receive a permanent identifier.

Recommended format:

ER-P00001

ER-P00002

ER-P00003

and so on.

Identifiers must never be reassigned.

XVI. Publication Threshold

The Emergence Record should not publish predictions merely to populate the ledger.

A forecast should be included only when:

  • the question is genuinely important or informative;
  • meaningful evidence can be assessed;
  • measurable resolution criteria can be defined;
  • the forecast adds something to the historical record.

Prediction volume is not a measure of forecasting quality.

XVII. Forecasting Record

The publication should eventually display public statistics including:

  • total predictions;
  • open predictions;
  • resolved predictions;
  • true resolutions;
  • false resolutions;
  • ambiguous resolutions;
  • average Brier score;
  • calibration by probability range;
  • forecasting performance by category;
  • forecasting performance by model or editorial system.

The record should remain visible whether performance is impressive, mediocre, or poor.

XVIII. Governing Principle

A prediction is valuable because it exposes contemporary uncertainty to future judgment.

We do not preserve forecasts to prove that artificial intelligence knew the future.

We preserve them so future readers can measure what artificial intelligence believed the future might be, how strongly it believed it, and how those beliefs compared with what actually happened.

The prediction itself is part of the historical record.