Behavioral Experiments and Changed Assumptions

E1: Notification and reactive lifetime

Start with L02's snapshot membership. Predict A-removes-B/adds-C and nested-publish traces, then implement immediate suppression as a separate variant. Explain which expectations must change. Model three sources (country, method, address) feeding two derived values (needs_address, can_submit). Show disposal and a late async response. Compare bounded queue policies for view updates and accepted records. Required evidence: both traces, eight input combinations, lifetime diagram and one overflow/stale-result counterexample. A paper model plus executable pure coordinator is sufficient; no reactive framework is required.

E2: History representation and retention

Compare L03 snapshots with a delta-based implementation for append/replace/clear. Use the same version-list oracle. Count retained characters and history entries on a stated document/edit workload; explain sharing and whether the count includes both undo and redo. Inject an invalid edit after undo and a valid no-op edit. Required evidence: counts, semantic agreement, one aliasing counterexample, and a retention policy. Wall-clock timing is optional and must not be presented as a universal complexity result.

E3: Accepted state and notification failure

Compose L04 with L02. Predict a valid transition followed by an observer failure. Keep the accepted record, report callback errors separately and replay without callbacks. Compare with a proposed notify-before-commit order and produce a counterexample showing observers see old state or emit effects for an uncommitted change. Add a fake send that raises after recording its effect and explain the uncertain outcome. Required evidence: state/effect timeline, recovery decision, one deliberately broken variant caught by a test, and excluded durability claims.

Budget 8–12 hours across the three experiments, reusing lab artifacts. Record actual time, disagreement and changes in reasoning. Submit to C09/C10 review.