PJC Practitioner Guide

Document status: Research Preview v0.6
Purpose: Run a complete prospective judgment-calibration cycle.
Not: a performance-review system, an automated decision maker, or proof of effectiveness.

1. Check that the task fits

Use PJC when:

• a judgment is formed before an outcome is known;
• the outcome will arrive within a practical observation window;
• the outcome source can be specified in advance;
• similar judgments recur, making calibration over time possible.

Do not use it when outcomes are unobservable, criteria can only be invented after the fact, or records will directly punish individuals.

2. Write a settleable judgment

A judgment must be a proposition that a later result can support, contradict, or leave inconclusive.

Weak: “This design should work well.”

Settleable: “By 30 September 2026, the new onboarding flow will raise first-canvas success to at least 45%, measured from production events.”

Include a proposition, observation window, outcome source, and decision threshold.

3. Complete the prospective record

Before the outcome is known, record:

• subject and time;
• proposition;
• confidence;
• principal rationale;
• at least one alternative explanation;
• observation window and outcome source;
• falsification condition;
• stop condition.

Confidence is a subjective probability that the proposition will hold. It is not confidence in personal ability or task completion.

4. Commit and freeze

After the record enters Committed:

• original fields cannot be overwritten;
• new information is append-only;
• every appendix keeps its author and time;
• post-outcome information cannot be presented as original reasoning.

If outcome information had already leaked at commitment time, mark the record Invalidated.

5. Execute and wait

PJC does not prescribe an execution method. Agents may support research, implementation, and checking, but an identifiable human remains the judgment subject.

During the wait, append new evidence, environmental changes, and execution deviations without changing the proposition, confidence, or settlement rule.

6. Settle against the outcome

At the end of the observation window, choose:

Record the actual outcome, evidence, comparison with the threshold, reasons for inconclusiveness or invalidation, and a concrete adjustment for the next cycle.

7. Calibrate over time

One record describes one judgment. After a set of comparable records exists, inspect:

• actual frequency among judgments made at similar confidence;
• rationale types that repeatedly fail;
• task classes that are often inconclusive;
• overconfidence under workload;
• whether agent participation changes judgment quality, not only output speed.

v0.6 specifies no universal score. Do not compress heterogeneous judgments into one number.

8. Common failure modes

Writing after the outcome

A post-outcome record is not prospective. Invalidate it instead of manufacturing a complete history.

Vague outcome criteria

“Users like it” cannot be settled reliably. Pre-specify data, reviewers, or a review protocol.

Equating completion with correctness

An agent completing an action does not settle the proposition.

Keeping only wins

Every committed record remains in history, including failures and inconclusive cases.

Letting an agent own the judgment

An agent may organise evidence, but the human subject confirms the proposition, confidence, and falsification rule.

Turning PJC into punishment

When records determine rewards, people optimise records rather than calibration. PJC defaults to learning and method review.

9. Minimum start

1. Choose one recurring, low-risk judgment type.
2. Collect at least ten records.
3. Avoid elaborate scoring.
4. Review settlement rate and field completion.
5. Repair propositions and outcome sources.
6. Expand only if the workflow remains usable.

The first test is not “did accuracy improve?” but “can the team produce honest, consistently settleable records?”