Cost and coverage

Diagnostics answer “what should I inspect?” The cost and coverage commands answer two related questions: “why might this render be expensive?” and “what could the analyzer not prove?”

Symbolic render cost

qual cost scenes/demo.py

The report groups evidence by Scene:

profiles: production (cairo, 1920x1080, 60 fps)

scene scenes.demo.TrackerDemo (scenes/demo.py)
  plays:
    scenes/demo.py:13:9 play duration 8 s -> frames ~480
  hot contexts:
    scenes/demo.py:9:31 entry always_redraw; factors frames
  per-frame constructions:
    scenes/demo.py:9:39 MathTex construction x at least ~480 invocations
  resource-key growth:
    scenes/demo.py:9:39 MathTex distinct cache keys: at least ~480

Qual tracks dimensions such as frames, family members, points, curves, pixels, and distinct resource keys. It does not turn unknown values into a fabricated wall-clock estimate. An unresolved duration remains per-frame or unknown.

Analysis coverage

A clean lint run can mean either “no findings” or “some relevant behavior was not statically resolvable.” Inspect that boundary explicitly:

qual coverage .
qual coverage . --format json
qual check . --analysis-summary

Coverage reports include parse success, unresolved imports and calls, unknown play durations, unknown animation targets, and helper-summary fallbacks. The counts describe computed analysis facts, not statistical estimates.

Cache behavior

Normal checks keep a disposable SQLite cache in:

.qual-cache/cache-v2.sqlite3

Add .qual-cache/ to the analyzed project's ignore file. The cache is never required for correctness; corruption or I/O failure falls back to analysis. Use --no-cache to disable all cache reads, writes, and directory creation.