# AI change log

Record each change, the problem it addresses and the checks performed.

Revision: 2026-10-13

## When to use it

Whenever instructions, models, sources, rules or integrations change.

## Expected output

Versions with rationale, impact, checks and a release decision.

## 1. Describe the change

Record the proposal before applying it. Avoid vague labels such as “optimisation”: readers should understand what changes in everyday work.

### Identifier, author and version

Record date, author, starting version and a reference to the proposed configuration.

________________

### Problem and evidence

Link the case, defect or request motivating the change; distinguish the symptom from its suspected explanation.

________________

### Components and expected effects

List prompts, sources, fields and systems involved; specify behaviours that must remain unchanged.

________________

> Illustrative example, replace with your own: Illustrative example: revise date extraction to distinguish issue date from deadline after an ambiguous case. Other document fields must retain their existing meanings.

## 2. Check the impact

Repeat relevant checks and inspect connected behaviour for regressions. Fixing one case alone does not demonstrate an overall improvement.

### Reproduction and regression cases

List cases reproducing the defect, ordinary cases and cases where the system should request review.

________________

### Before and after results

Record correctness, critical failures, review requests and observed limits using identifiable configurations.

________________

### Approver and conditions

Specify who assesses operational impact and which defects or conditions prevent release.

________________

> Illustrative example, replace with your own: Illustrative example: the new instruction resolves the ambiguous date but omits an explicit deadline elsewhere. Release stays paused until the new failure is understood.

## 3. Track release and recovery

Keep the connection between versions and processed cases. For changed sources and data, describe how previous context can be reconstructed when permitted and necessary.

### Activation and affected cases

Record environment, time, included groups and how cases processed by the new version are identified.

________________

### Reversion and recovery conditions

Describe the previous version, stopping procedure and checks for effects already produced.

________________

### Follow-up and closure

Assign who observes initial results, when closure is decided and where regressions are recorded.

________________

> Illustrative example, replace with your own: Illustrative example: enable the correction for drafts only. If critical errors appear, restore the previous configuration and review cases from the affected window.

## Pitfalls to avoid

- Changing production configuration without being able to identify it.

- Recording code changes while overlooking sources, prompts and rules.

- Closing a change without inspecting its first live outcomes.

Stolen Orbit — https://stolenorbit.com/en/templates/ai-change-log/

