Skip to content

test(adaptive-quality): guard tier cost-monotonicity invariant#148

Merged
mvalancy merged 1 commit into
devfrom
test/adaptive-quality-monotonic-invariant
Jun 20, 2026
Merged

test(adaptive-quality): guard tier cost-monotonicity invariant#148
mvalancy merged 1 commit into
devfrom
test/adaptive-quality-monotonic-invariant

Conversation

@mvalancy

Copy link
Copy Markdown
Member

Adds a unit guard so a richer quality tier can never render less work than a cheaper one — across attachmentPreviewSize, labelLodZoom, and the four effect flags. The existing suite only guarded maxInitialNodes.

Motivated by a dashboard FPS anomaly (LOW dragFps < HIGH dragFps in a stale scale-sweep sample): even though the current BASE_PROFILES are correct, nothing locked the invariant against future config drift.

Pure test addition — no production code touched. Full adaptiveQuality suite: 31/31 green.

🤖 Generated with Claude Code

A richer quality tier must never render less work than a cheaper one
(smaller attachment preview, higher label-LOD threshold, or a disabled
effect a lower tier enables). The existing suite only guarded
maxInitialNodes; this locks the same invariant across attachmentPreviewSize,
labelLodZoom, and the four effect flags — the config-regression class that
would let a 'lower' tier paradoxically do MORE work.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@mvalancy mvalancy merged commit 332b4da into dev Jun 20, 2026
10 of 11 checks passed
@mvalancy mvalancy deleted the test/adaptive-quality-monotonic-invariant branch June 20, 2026 11:16
@github-actions

Copy link
Copy Markdown

🧪 Comprehensive Test Suite

  • Unit suites (Node 18.x & 20.x) — core, web, server, mcp-server: ✅ passed
  • Installer & deploy config: ✅ passed

Full-stack smoke gate runs in the CI workflow.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant