# CAIN42 — Frontier Research 13 (Prompt 13)

A research-orientation pass over the areas named in Prompt 13. **No claim is made that any specific paper is
implemented or reproduced here.** CAIN mechanisms are implemented independently and tested; the mapping is to
research *areas*. No external URLs are asserted because none were fetched in this batch — treat this as a direction
map, not a literature review.

| Area | Direction | CAIN mechanism | Status |
|---|---|---|---|
| Agentic transactions | ACID-like agent execution, sagas | two-phase effect commit + compensation graph | IMPLEMENTED (in-process) |
| Exactly-once | idempotency keys, dedup | IdempotencyLedger | IMPLEMENTED |
| In-doubt execution | ambiguous outcomes | EffectResolutionGraph (no blind retry) | IMPLEMENTED |
| Effect correctness | authorization ≠ effect | EffectContract + fingerprint | IMPLEMENTED |
| Reversibility | reversibility classes | ReversibilityAuthorityEngine R0–R7 | IMPLEMENTED |
| Evidence independence | correlated sources | EffectResolutionGraph roots | IMPLEMENTED |
| Continuous authorization | TTL, freshness | TemporalEffectValidityEngine | IMPLEMENTED |
| Human oversight | escalation fatigue | HumanOversightBudget | IMPLEMENTED |
| Machine economy | agent payments | EconomicEffectContract | IMPLEMENTED (SIMULATED) |
| Physical AI | sim-to-real | PhysicalEffectAdapter | SIMULATED |
| Machine-verifiable safety cases | effect receipts | VerifiedEffectReceipt + clean-room verifier | IMPLEMENTED |
| Byzantine / distributed | effect agreement across nodes | — | DESIGN (Prompt 14) |

## Missing capabilities this exposes (honest)

- Distributed effect coordination (multi-node effect commit) — not implemented.
- A real framework/MCPGate adapter that binds a downstream tool's actual execution to the effect receipt — not
  implemented (top blocker).
- Real financial and hardware adapters — not implemented.
