# Keep a zero baseline out of a scaled-error claim

Illustrative planning brief; no automatic product import.

Synthetic planning case: A toy training series has durations four,four,four days. A one-step last-value reference has zero mean absolute training change. A held-out forecast’s mean absolute error is two days.

## Decision

The two-day error cannot be divided by the zero reference to produce a finite scaled score. Report the absolute error and the undefined relative score instead of inventing a favorable ratio.

## Owned work

- Recover the reference construction
  - Owner role: Evaluation owner
  - Acceptance evidence: The denominator comes from this constant training series.
- Keep the undefined division explicit
  - Owner role: Reviewer
  - Acceptance evidence: The two-day absolute error remains reported without a fabricated finite scaled score.
- Choose a justified comparison
  - Owner role: Planning lead
  - Acceptance evidence: Any alternative scale or metric is declared separately rather than patched with an arbitrary denominator.

## Workflow

1. Recover the reference construction. Check: The denominator comes from this constant training series.
2. Keep the undefined division explicit. Check: The two-day absolute error remains reported without a fabricated finite scaled score.
3. Choose a justified comparison. Check: Any alternative scale or metric is declared separately rather than patched with an arbitrary denominator.

## Judgment

A tiny constant training series cannot validate a model or a replacement scale. Scaled comparison requires a meaningful nonzero reference.


## Filled manual planning note

The two-day error cannot be divided by the zero reference to produce a finite scaled score. Report the absolute error and the undefined relative score instead of inventing a favorable ratio. Reference scale=(|4−4|+|4−4|)/2=0 days. The proposed scaled value 2/0 is undefined; it is not zero error or evidence of an infinitely reliable model. A tiny constant training series cannot validate a model or a replacement scale. Scaled comparison requires a meaningful nonzero reference.


## Workflow questions

### Can I replace zero with one silently?

No. That changes the score’s meaning and comparison basis.

### Does constant training history prove future durations stay constant?

No. The held-out error already illustrates a different outcome.

## 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

Evidence | Values | Derived quantity
--- | --- | ---
Training durations | 4,4,4 days | No observed training change
Naive training errors | 0,0 days | Mean absolute reference error 0
Held-out absolute error | 2 days | Finite unscaled error

### Reasoning

Reference scale=(|4−4|+|4−4|)/2=0 days. The proposed scaled value 2/0 is undefined; it is not zero error or evidence of an infinitely reliable model.

### Bounded result

The two-day error cannot be divided by the zero reference to produce a finite scaled score. Report the absolute error and the undefined relative score instead of inventing a favorable ratio.

### Distinct decision

The artifact checks a zero training-reference denominator, distinct from an observed zero outcome in percentage error.

### Limits

A tiny constant training series cannot validate a model or a replacement scale. Scaled comparison requires a meaningful nonzero reference.

### 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.