DOCUMENT 079 OF 135
Title: Improve measured performance without damaging the website
Path: Library/Playbooks/05-PERFORMANCE-WITHOUT-REGRESSIONS.md
Group: playbooks
Label: EXPANSION · source-linked operating design
Review status: new_discipline_operating_design
==============================================================================

# Improve measured performance without damaging the website

**Lead:** R06. **Consulted roles:** R01, R03, R04, R05, R06, R07, R08, R16, R20, R21, R25, R26. **Mode:** read-only diagnosis until actual authorization.

## Entry condition

A measured field or lab bottleneck affects representative customer journeys. The goal is not an isolated score improvement at any cost.

## Sequence and ownership

### 1. Define the baseline

R06 separates field and lab data, records routes/templates, windows, devices and tool versions. Missing field data is unknown, not a passing metric.

### 2. Find the dominant cause

Inspect primary-content delivery, interaction blocking, layout movement, fonts, images, scripts and cache behavior using actual traces.

### 3. Review interfaces

R01 reviews runtime/cache changes; R20 checks private response exposure; R08 checks meaningful content and directives; R04/R07/R05 protect actual interactions.

### 4. Change one controlled unit

Use a minimal reversible proposal. Record baseline and candidate under comparable conditions. R21 executes critical journey regressions; R16 checks event integrity.

### 5. Release and observe

R25 sequences the authorized change; R26 checks the declared build. Report immediate lab results separately from later sufficiently sampled field results.

## Evidence package

Preserve task/run IDs, exact environment/build, timestamped observations, safe synthetic inputs, expected versus observed state, tool/browser versions, raw private artifacts and separately hashed public-safe redactions. Record unavailable tools and tests not executed. Attach the handoffs rather than compressing distinct observations into one success flag.

## Acceptance

- The claimed improvement matches a reproducible method
- No inquiry, accessibility, privacy, browser or URL-policy regression
- Field claims use available appropriate field evidence

## Stop or block

Do not cache personal API responses, remove essential content, defer critical behavior or claim conversion uplift from a lab score. If the historical field window is incomplete, state that explicitly.

## Role and source references

- [R01 — Site Reliability Engineer](../Disciplines/R01-site-reliability.md)
- [R06 — Web Performance Engineer](../Disciplines/R06-web-performance.md)
- [R20 — Application Security/Privacy Engineer](../Disciplines/R20-application-security-privacy.md)

Primary-source IDs: D09, D10, D05, D11. See the [source register](../../Research/DISCIPLINE-SOURCE-REGISTER.md) for URLs, scope and currentness boundaries. The sequence is custom operating design; it is not a promise that every case has the same cause.

==============================================================================