Chaos laboratory

Chaos testing is deterministic. A campaign is generated from a seed, then replayed as an explicit operation list.

Run a campaign

CHAOS_SEED=3737844653 CHAOS_STEPS=100 npm run test:chaos

The default laboratory combines browser-state faults such as:

• DOM insertion, removal, and reordering;
• form and focus changes;
• scroll and pointer movement;
• visibility and attribute changes;
• Shadow DOM changes;
• frame attachment and detachment;
• virtualized list transitions.

Failure record

Every failing campaign records:

• seed and step count;
• exact generated plan;
• browser project;
• checkpoint hashes;
• replay command;
• Playwright trace and screenshot;
• minimized failing history when reduction succeeds.

This record converts “the browser was chaotic” into a reproducible test case.

Minimization

The reducer removes ranges of operations and replays the candidate plan. It keeps a smaller plan only when the same invariant still fails.

Important invariants include:

• same plan and initial state produce the same semantic checkpoints;
• all frame parents and owner paths resolve;
• all node parents and references remain closed;
• capture does not mutate the inspected page;
• configured budgets and explicit truncation agree;
• TQ1 round trips preserve the snapshot exactly.

Operational use

Keep the exact seed in regression tests after fixing a defect. Add a smaller hand-authored conformance case when the minimized history reveals a stable browser primitive worth testing directly.