# Keep zero outcomes out of undefined percentage-error claims

Illustrative planning brief; no automatic product import.

Synthetic planning case: A toy duration evaluation has A:forecast two recorded minutes, observed zero; B:forecast three, observed four. A’s zero can reflect the recording convention, whose physical meaning is unresolved.

## Decision

A’s absolute percentage error is undefined because its observed denominator is zero. B’s is 25%; report A’s two-minute absolute error separately rather than inventing an overall finite percentage average.

## Owned work

- Verify observed-value semantics
  - Owner role: Evaluation owner
  - Acceptance evidence: Zero’s recording convention and unit are retained without inferring instantaneous physical work.
- Keep undefined scores visible
  - Owner role: Reviewer
  - Acceptance evidence: A is not replaced with zero percent or silently dropped from all-case coverage.
- Choose a justified comparison
  - Owner role: Planning lead
  - Acceptance evidence: Report compatible unscaled errors or an explicitly justified different metric, with its changed meaning declared.

## Workflow

1. Verify observed-value semantics. Check: Zero’s recording convention and unit are retained without inferring instantaneous physical work.
2. Keep undefined scores visible. Check: A is not replaced with zero percent or silently dropped from all-case coverage.
3. Choose a justified comparison. Check: Report compatible unscaled errors or an explicitly justified different metric, with its changed meaning declared.

## Judgment

Recorded zero does not establish actual effort or duration semantics. No future forecast performance is inferred.


## Filled manual planning note

A’s absolute percentage error is undefined because its observed denominator is zero. B’s is 25%; report A’s two-minute absolute error separately rather than inventing an overall finite percentage average. Absolute percentage error uses 100×|actual−forecast|/|actual|. B gives 100×1/4=25%; A requires 2/0 and is undefined. A’s unscaled absolute error remains two minutes. Recorded zero does not establish actual effort or duration semantics. No future forecast performance is inferred.


## Workflow questions

### Does a zero observed value mean zero prediction error?

No. A predicts two while zero is recorded, so its absolute difference is two.

### Can I divide by a tiny substitute value?

Not silently. That introduces an arbitrary scoring rule and can radically change the result.

## 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 | Forecast recorded minutes | Observed recorded minutes | Absolute error | Absolute percentage error
--- | --- | --- | --- | ---
A | 2 | 0 | 2 min | Undefined: zero denominator
B | 3 | 4 | 1 min | 25%

### Reasoning

Absolute percentage error uses 100×|actual−forecast|/|actual|. B gives 100×1/4=25%; A requires 2/0 and is undefined. A’s unscaled absolute error remains two minutes.

### Bounded result

A’s absolute percentage error is undefined because its observed denominator is zero. B’s is 25%; report A’s two-minute absolute error separately rather than inventing an overall finite percentage average.

### Distinct decision

This checks an observed-zero denominator in percentage error, distinct from a zero training-reference scale.

### Limits

Recorded zero does not establish actual effort or duration semantics. No future forecast performance is inferred.

### Definitions and method context

- Forecasting: Principles and Practice — accuracy — https://otexts.com/fpp3/accuracy.html — Genuine held-out forecast and point-error evaluation context. Binary scoring examples use their own explicit toy definitions; no real task model is validated.