DIRA
One thing changes. Everything adapts.

One moved commitment can break the whole week.

A midterm moves from Friday to Wednesday. The interview buffer disappears, study capacity no longer fits, and a team deliverable is suddenly at risk. Calendars show that conflict. Dira repairs it.

It only takes actions already permitted by the people and policies in the plan, then verifies the changed world before calling the repair complete.

▶ Run the 48-Hour ShockLIVE CLOUD · Vertex + Calendar
External changeMidterm: Fri 2 PM → Wed 2 PM
What breaksInterview buffer · study capacity · team QAGLOBAL SLACK +4.1h → −3.6h
Dira's verified repairReplan, rebook, delegate, rebuildRESOLVED · +1.3h · 0 HUMAN ACTIONS
+4h0h-4h+4.1hbefore trigger-3.6hafter mutation+1.3hafter repair
Commitments
10
At risk
0
Interventions (resolved)
1
Global slack
+1.3h
ECON 402 — the 48-Hour ShockRESOLVED
Commitments affected
6
Verified adapter mutations
13
Failures recovered
1
User interventions
0

A professor email moved the midterm forward 48 hours. Dira re-derived the plan: rebooked the interview (surviving a 409 on the first slot), delegated sponsor-deck QA to the backup owner, reclaimed two personal blocks, and rebuilt the study plan — all policy-gated, ledgered, and independently verified. Open the flight recorder →

Runtime note: this overview is always the reproducible deterministic reference. Production credentials are configured for this deployment; the Interventions page verifies live Cloud Run connectivity per run and labels the boundary. Reproduce the reference with make demo-replay.