# CAIN CAUSALITY — Competitive Analysis, Hostile Review & Risk / Market Analysis

## Research landscape (fresh, architectural; no paper copied)
Distributed causality (Lamport/vector clocks, happens-before), provenance (W3C PROV), distributed tracing/event
sourcing (what/how), causal-inference libraries (population effects via SCMs/do-calculus), and recent long-horizon
agent harnesses / agentic world models (trajectory evaluation, collusion detection). None binds **evidence-gated,
per-outcome, multi-actor causal accountability with counterfactual necessity and non-collapsing responsibility** into a
runtime certificate. That is the gap CAUSALITY targets.

## Hostile competitive matrix

| System / field | Already does | CAUSALITY difference |
|---|---|---|
| OpenTelemetry / distributed tracing | request flow, spans | traces ≠ causal necessity; CAUSALITY proves supported edges and contribution |
| W3C PROV | provenance graph | PROV has no counterfactual necessity or responsibility categories |
| Event sourcing | durable local log | internal log, not independent-outcome causality across systems |
| SIEM / SOAR | alert + response | alerts, not machine-verified accountability |
| Blockchain provenance | immutable log | immutability ≠ causal inference |
| Causal inference libraries | population causal effect | not per-outcome actor accountability across delegated agents |
| LLM judge | plausible explanation | plausibility ≠ evidenced causality |
| IRONCLAD | single-effect stage chain | CAUSALITY is multi-event, multi-actor, typed |
| REALITY | external reality reconciliation | CAUSALITY explains WHY, not only WHAT |
| EPICENTER | mission completion verdict | CAUSALITY attributes contribution and responsibility |

**If the answer were "CAIN is just distributed tracing," the project would have failed.** It is not: CAUSALITY adds
evidence-gated causal edges, counterfactual necessity with real controls, six non-collapsing responsibility categories,
laundering/collusion detection, and a clean-room-verifiable accountability certificate — plus a REAL two-system
experiment. This is architectural differentiation, not a rename.

## Hostile investor review (answered honestly)
- *Why isn't this OpenTelemetry/prov* → those record flow/provenance; they neither test necessity nor separate
  authority from causal/operational/observational responsibility.
- *Why isn't this event sourcing / a log* → a log is not causality; CAUSALITY refuses timestamp order as causation.
- *Two agents collude* → `MultiAgentCausalAttributionEngine` flags CAUSAL_COLLUSION, correlated actors, shared roots.
- *The observer lies* → evidence independence + REQUIRED independent roots; a self-certified observer is not independent.
- *Clocks disagree* → ordering is vector-clock based; concurrent events stay CONCURRENT.
- *External state changes independently* → concurrent-actor detection; the change is not attributed to CAIN.
- *Causality cannot be determined* → the state stays CAUSALITY_UNKNOWN / EVIDENCE_INSUFFICIENT; no false blame.
- *What prevents false blame* → contribution requires evidence; contradictions and unresolved causes block verification.
- *What would make it merely engineering* → if it were only tracing; it is not, per the matrix.

## Risk / market analysis (TECHNICAL EVIDENCE vs LEGAL/COMMERCIAL INTERPRETATION)
- **Technical evidence capability:** reproducible causal certificates (ordered edges, contribution, necessity,
  responsibility, clean-room verification).
- **Legal/commercial interpretation:** NOT provided. CAUSALITY makes **no** legal, insurance, liability or contractual
  determination. It produces evidence a human/insurer/court may choose to interpret.
- Candidate applications (evidence only): autonomous-system incident reconstruction, machine-to-machine transaction
  accountability, vendor/agent accountability evidence, autonomous audit. Requires third-party validation before any
  commercial claim.

## Honest limits
Two real external systems demonstrated; architectural comparison only; no third-party validation; no legal claim;
physical SIMULATED; pre-revenue.
