Skill Wiki
Perf Budget Rollout
Accumulated execution feedback from teams adopting Core Web Vitals performance budgets via @community/collection-performance-budget.…
$ prime install @community/feedback-perf-budget-rolloutProjection
Always in _index.xml · the agent never has to ask for this.
PerfBudgetRollout [feedback] v1.0.0
Accumulated execution feedback from teams adopting Core Web Vitals performance budgets via @community/collection-performance-budget. Documents the observed friction between budget definition, CI enforcement, and sustained compliance over 3+ months.
Loaded when retrieval picks the atom as adjacent / supporting.
PerfBudgetRollout [feedback] v1.0.0
Accumulated execution feedback from teams adopting Core Web Vitals performance budgets via @community/collection-performance-budget. Documents the observed friction between budget definition, CI enforcement, and sustained compliance over 3+ months.
Target
@community/collection-performance-budget
Observations
- Sample Size: 8
- Period: 2025-Q1 to 2026-Q1
- Success Rate: 0.62
Common Failures:
- Step: Lighthouse CI gating
- Failure: Lighthouse synthetic results vary ±15% run-to-run; team disables gating after first false positive; reverts to no enforcement
- Frequency: 0.62
- Impact: budget exists on paper, no enforcement, regressions ship
- Mitigation: run Lighthouse 5x per PR and use median; require 3 of 5 runs below threshold; allow CI re-run on suspected variance
- Step: RUM data interpretation
- Failure: team alarms on p75 LCP spike; investigation reveals it's a small subset of users on slow connections in remote regions; no actionable fix
- Frequency: 0.5
- Impact: alert fatigue; team learns to dismiss vitals alerts
- Mitigation: segment RUM by device class + connection effective-type + country; alert only on top-50-traffic segments; long-tail goes to a quarterly review, not pager
- Step: @community/anti-pattern-blocking-third-party
- Failure: marketing team adds a new tag (chat widget, A/B test platform, vendor pixel); engineering only learns when LCP regresses
- Frequency: 0.75
- Impact: every quarter LCP creeps up by 100-300ms from third-party sprawl
- Mitigation: third-party additions require an engineering review + budget impact estimate before merge; CSP allowlist gates new domains
- Step: @community/pattern-image-lcp-priority
- Failure: CMS-managed hero images uploaded by content team without explicit dimensions; LCP regresses on every campaign launch
- Frequency: 0.5
- Impact: marketing pages fail LCP gate intermittently
- Mitigation: CMS enforces image dimensions at upload; render layer always emits width/height; CMS plugin auto-converts to AVIF
- Confidence Adjustments:
- @Community/Rule Cls Budget: 0.91
- @Community/Pattern Image Lcp Priority: 0.86
- @Community/Anti Pattern Blocking Third Party: 0.74
- @Community/Principle Perceived Vs Actual: 0.82
Ai Confidence
0.76
Lessons
- CI gating MUST tolerate Lighthouse variance — single-run gates are noise, not signal.
- RUM segmentation is the difference between actionable and aspirational. p75 over all users is too noisy to act on.
- Third-party drift is the single biggest source of long-term performance regression. Process gates beat technical gates.
- CMS-driven content (marketing-uploaded images) needs platform discipline; engineering-set rules without CMS enforcement degrade weekly.
- Quarterly perf review with leadership creates accountability; without it, budgets become dead documents within 6 months.
Loaded when retrieval picks the atom as a focal / direct hit.
PerfBudgetRollout [feedback] v1.0.0
Accumulated execution feedback from teams adopting Core Web Vitals performance budgets via @community/collection-performance-budget. Documents the observed friction between budget definition, CI enforcement, and sustained compliance over 3+ months.
Target
@community/collection-performance-budget
Observations
- Sample Size: 8
- Period: 2025-Q1 to 2026-Q1
- Success Rate: 0.62
Common Failures:
- Step: Lighthouse CI gating
- Failure: Lighthouse synthetic results vary ±15% run-to-run; team disables gating after first false positive; reverts to no enforcement
- Frequency: 0.62
- Impact: budget exists on paper, no enforcement, regressions ship
- Mitigation: run Lighthouse 5x per PR and use median; require 3 of 5 runs below threshold; allow CI re-run on suspected variance
- Step: RUM data interpretation
- Failure: team alarms on p75 LCP spike; investigation reveals it's a small subset of users on slow connections in remote regions; no actionable fix
- Frequency: 0.5
- Impact: alert fatigue; team learns to dismiss vitals alerts
- Mitigation: segment RUM by device class + connection effective-type + country; alert only on top-50-traffic segments; long-tail goes to a quarterly review, not pager
- Step: @community/anti-pattern-blocking-third-party
- Failure: marketing team adds a new tag (chat widget, A/B test platform, vendor pixel); engineering only learns when LCP regresses
- Frequency: 0.75
- Impact: every quarter LCP creeps up by 100-300ms from third-party sprawl
- Mitigation: third-party additions require an engineering review + budget impact estimate before merge; CSP allowlist gates new domains
- Step: @community/pattern-image-lcp-priority
- Failure: CMS-managed hero images uploaded by content team without explicit dimensions; LCP regresses on every campaign launch
- Frequency: 0.5
- Impact: marketing pages fail LCP gate intermittently
- Mitigation: CMS enforces image dimensions at upload; render layer always emits width/height; CMS plugin auto-converts to AVIF
- Confidence Adjustments:
- @Community/Rule Cls Budget: 0.91
- @Community/Pattern Image Lcp Priority: 0.86
- @Community/Anti Pattern Blocking Third Party: 0.74
- @Community/Principle Perceived Vs Actual: 0.82
Ai Confidence
0.76
Lessons
- CI gating MUST tolerate Lighthouse variance — single-run gates are noise, not signal.
- RUM segmentation is the difference between actionable and aspirational. p75 over all users is too noisy to act on.
- Third-party drift is the single biggest source of long-term performance regression. Process gates beat technical gates.
- CMS-driven content (marketing-uploaded images) needs platform discipline; engineering-set rules without CMS enforcement degrade weekly.
- Quarterly perf review with leadership creates accountability; without it, budgets become dead documents within 6 months.
Source
prime-system/examples/frontend-design/primes/compiled/@community/feedback-perf-budget-rollout/atom.yaml