# Plan a rolling-origin task forecast evaluation

Illustrative planning brief; no automatic product import.

Synthetic planning case: A toy one-step forecast plan uses observations through day three to predict day four, then through day four to predict day five, and through day five to predict day six.

## Decision

Each outcome stays beyond its own fitting cutoff while the one-step horizon remains fixed. Keep all three origins rather than letting a later fitted rule retrospectively see earlier held-out outcomes.

## Owned work

- Freeze each origin boundary
  - Owner role: Evaluation planner
  - Acceptance evidence: Every prediction has its own allowed information cutoff.
- Match the evaluation horizon
  - Owner role: Reviewer
  - Acceptance evidence: All three target records describe one-step predictions.
- Retain genuine origin forecasts
  - Owner role: Recorder
  - Acceptance evidence: Later fits do not overwrite the saved earlier predictions or their input boundaries.

## Workflow

1. Freeze each origin boundary. Check: Every prediction has its own allowed information cutoff.
2. Match the evaluation horizon. Check: All three target records describe one-step predictions.
3. Retain genuine origin forecasts. Check: Later fits do not overwrite the saved earlier predictions or their input boundaries.

## Judgment

Actual features, fitting procedure and outcome timing must respect the cutoff. No native forecasting workflow is asserted.


## Filled manual planning note

Each outcome stays beyond its own fitting cutoff while the one-step horizon remains fixed. Keep all three origins rather than letting a later fitted rule retrospectively see earlier held-out outcomes. There are three prospective origin/target pairs. At each origin, the target day is excluded from fitting; earlier observed targets can become allowed history at a later origin. Actual features, fitting procedure and outcome timing must respect the cutoff. No native forecasting workflow is asserted.


## Workflow questions

### Can day-four observations be used at the day-four origin?

Yes once observed and allowed; they could not be used in the earlier day-three prediction of day four.

### Does three-origin coverage establish future accuracy?

No. This plan defines evaluation boundaries, not a validated performance 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

Forecast origin | Allowed observations | Held-out target | Horizon
--- | --- | --- | ---
Day 3 | Days 1–3 | Day 4 | One step
Day 4 | Days 1–4 | Day 5 | One step
Day 5 | Days 1–5 | Day 6 | One step

### Reasoning

There are three prospective origin/target pairs. At each origin, the target day is excluded from fitting; earlier observed targets can become allowed history at a later origin.

### Bounded result

Each outcome stays beyond its own fitting cutoff while the one-step horizon remains fixed. Keep all three origins rather than letting a later fitted rule retrospectively see earlier held-out outcomes.

### Distinct decision

The artifact plans repeated prospective cutoffs at a fixed horizon, rather than a single held-out split or hindsight replacement.

### Limits

Actual features, fitting procedure and outcome timing must respect the cutoff. No native forecasting workflow is asserted.

### Definitions and method context

- Forecasting: Principles and Practice — time series cross-validation — https://otexts.com/fpp3/tscv.html — Rolling-origin evaluation context only. The explicit origins, observations and errors here are invented original examples, not validated model results.