Decide whether to roll back or push through

Help me decide: roll back, or fix forward.

Produce:
1. **Recommendation first — ROLL BACK / FIX FORWARD / WAIT.** One of the three. State it before reasoning.
2. Cost of rolling back: data written under the new version, migrations already run, client versions in the field, work lost.
3. Cost of staying: what degrades, for whom, and whether it compounds with time.
4. **If WAIT: the exact clock time by which the decision is made, and what observation would flip it.** *A wait without a deadline is how a ten-minute incident becomes two hours.*
5. Whether rollback is actually possible — irreversible migrations, consumed messages, sent notifications, changed external state.
6. What to capture before rolling back, since rollback usually destroys the evidence.

Rules:
- *When rollback is cheap and the symptom is unexplained, recommend rolling back.* Diagnosis is easier without an ongoing incident, and this is the bias people get wrong under pressure.
- Fix forward needs a specific known fix, not confidence that one will be found.
- If the deploy contained multiple changes, say whether partial rollback is possible.