# Inspect false positives and false negatives at a decision threshold

Illustrative planning brief; no automatic product import.

Synthetic planning case: Four cases have supplied ready probabilities 0.9,0.7,0.4,0.2 and observed ready labels 1,0,1,0. The declared rule predicts ready when probability is at least 0.5.

## Decision

The rule has one true-ready, one false-ready, one missed-ready and one true-not-ready case. Keep the two error types separate rather than reporting only 50% accuracy.

## Owned work

- Freeze the threshold and positive meaning
  - Owner role: Evaluation owner
  - Acceptance evidence: Positive means ready; equality at 0.5 counts as ready.
- Retain each case classification
  - Owner role: Reviewer
  - Acceptance evidence: A/B/C/D map to all four named cells under the declared rule.
- Review error consequences separately
  - Owner role: Decision owner
  - Acceptance evidence: An early commitment and an unnecessary wait get distinct actions rather than an undifferentiated error count.

## Workflow

1. Freeze the threshold and positive meaning. Check: Positive means ready; equality at 0.5 counts as ready.
2. Retain each case classification. Check: A/B/C/D map to all four named cells under the declared rule.
3. Review error consequences separately. Check: An early commitment and an unnecessary wait get distinct actions rather than an undifferentiated error count.

## Judgment

No real classifier, calibrated readiness forecast or automatic TeamBoostAI decision is asserted.


## Filled manual planning note

The rule has one true-ready, one false-ready, one missed-ready and one true-not-ready case. Keep the two error types separate rather than reporting only 50% accuracy. At threshold 0.5, A and B predict ready while C and D predict not ready. Only A and D match observations: 2/4=50% accuracy. False-ready B and missed-ready C have different planning consequences. No real classifier, calibrated readiness forecast or automatic TeamBoostAI decision is asserted.


## Working artifact

- False-ready case B / 0.7 predicts ready; observed not ready / Inspect premature commitment consequence
- Missed-ready case C / 0.4 predicts not ready; observed ready / Inspect unnecessary wait consequence
- Correct cases A and D / Match under supplied rule / Do not erase the two different errors


## Workflow questions

### Does this threshold have optimal performance?

No. It is a supplied rule; choosing a threshold requires an objective and appropriate evaluation.

### Can these counts be treated as probabilities?

They describe four invented cases, not future error rates.

## Product connection

Use the owned checks and downloaded brief to discuss this planning decision alongside your TeamBoostAI tasks. Confirm available fields, roles and account features separately. The example is manual; it does not calculate live analytics, create work or run an experiment in the product.

Confirm account availability before adopting this manual outline.

## Original worked case

Synthetic records, manual planning only. No account import or live analytics.

### Inspect the invented case records

Case | Ready probability | Observed ready | Predicted ready | Category
--- | --- | --- | --- | ---
A | 0.9 | 1 | 1 | True positive
B | 0.7 | 0 | 1 | False positive
C | 0.4 | 1 | 0 | False negative
D | 0.2 | 0 | 0 | True negative

### Reasoning

At threshold 0.5, A and B predict ready while C and D predict not ready. Only A and D match observations: 2/4=50% accuracy. False-ready B and missed-ready C have different planning consequences.

### Bounded result

The rule has one true-ready, one false-ready, one missed-ready and one true-not-ready case. Keep the two error types separate rather than reporting only 50% accuracy.

### Distinct decision

This maps a fixed threshold to four case-level outcome cells, distinct from probabilistic scoring or majority-class accuracy.

### Limits

No real classifier, calibrated readiness forecast or automatic TeamBoostAI decision is asserted.

### Definitions and method context

- Original worked-case definitions — Definitions, policy choices, records and calculations are authored for this worksheet. No external standard, statistical validation or live product measurement is claimed.