# CAIN REALITY — Competitive Analysis, Research Landscape & Hostile Investor Review

## Research landscape (fresh, source-grounded)
Recent work (2024–2026) increasingly treats agent environments as stateful systems whose transitions must be predicted
and reconciled, not merely scored by an LLM: Qwen-AgentWorld / agentic world models, Trace2Env, mission-level embodied
evaluation harnesses, world-model evaluation (physics/world-model benchmarks), OpenShell, CAVA, Microsoft Agent
Governance Toolkit, AgentWall, SafeAgent, Proof-or-Stop, and event-sourcing/consistency work (CRDTs, causal
consistency, distributed tracing). No studied system binds a **cryptographically signed external-state reconciliation +
causal attribution + concurrent-actor defense** into a governed ground-truth verdict. That is the gap REALITY targets.
(Comparisons are architectural; no head-to-head benchmark was run.)

## Competitive matrix

| System / field | Implements | Does not implement | REALITY difference |
|---|---|---|---|
| Observability (OTel, LangSmith) | traces/metrics | signed reconciliation verdict + causal attribution | REALITY turns observation into a governed ground truth |
| Event sourcing / CRDTs | local consistency | external ground truth vs authorized outcome | REALITY reconciles against external reality, not an internal log |
| Digital twins | simulated state | independent external observation | REALITY refuses to treat simulation as real observation |
| Workflow engines (Temporal/Airflow) | durable execution | verified external outcome + authority | REALITY verifies the outcome at the boundary |
| Agent evaluation / LLM judge | plausibility scoring | external observation, non-repudiation | REALITY is runtime and evidence-gated |
| OpenShell / CAVA / AgentWall / SafeAgent / MGT | sandbox/guardrails | external ground-truth certificate | REALITY certifies reality above the guard |
| IRONCLAD (CAIN) | effect verification | independent external observation + causal attribution | REALITY supplies the external observation |
| EPICENTER (CAIN) | mission completion | external reality evidence | REALITY is the evidence source for the mission verdict |

## Hostile investor review (strongest objections, answered)
- **"This is just observability."** Observability emits telemetry; REALITY emits a signed ground-truth verdict with
  causal attribution and a clean-room verifier, and it can refuse a completion.
- **"This is just event sourcing / a state machine."** Event sourcing is local; REALITY reconciles against an external
  system CAIN does not own, with concurrent-actor defense.
- **"This is just an external API checker."** An API checker returns a boolean; REALITY does independent observation,
  field reconciliation, causal attribution, contradiction preservation and evidence independence.
- **"This is IRONCLAD/EPICENTER."** IRONCLAD verifies effects from CAIN's view; EPICENTER verifies mission completion;
  REALITY provides the external reality both depend on for truth.
- **"This is a digital twin."** The opposite: REALITY explicitly refuses simulation as real observation.
- **Weakest point:** only one real external system is demonstrated. Fixing that (a Postgres/K8s/cloud adapter with a
  real reversible canary) is the next milestone.

## Honest limits
ONE real external system demonstrated; physical SIMULATED; no third-party validation; no patent; pre-revenue.
