# E24 — Reproduction Complexity (essential vs accidental)

Complexity only contributes to defensibility when it is **essential** — i.e. required for a real
governance/security property, not stylistic. This is an engineering analysis, not a legal or valuation
determination.

## Dimensions (measured from the repository)

| Dimension | Measure | Kind |
|---|---|---|
| State dimensions | E21: 14 domain-separated roots; E24: 12 | ESSENTIAL |
| Transition semantic fields | E21: 34 bound fields → one transition hash | ESSENTIAL |
| Invariant corpus | I13–I26 (E20), I27–I50 (E21), I51–I72 (E22), I73–I96 (E23), I97–I108 (E24) | ESSENTIAL |
| Cryptographic dependencies | Ed25519 certs, RFC-6962-style Merkle, HMAC transport, domain separation | ESSENTIAL |
| Distributed-system requirements | N=4/Q=3 TCP transport, divergence detection, quarantine, recovery | ESSENTIAL (but same-host only) |
| Adversarial corpora | 16 Byzantine state attacks, 7 recovery attacks, 14 chaos scenarios, TCP fault matrix | ESSENTIAL |
| Evidence infrastructure | signed bundle + manifest + zero-import verifier per evolution | ESSENTIAL |
| Cross-product execution | 5 in-process product executors; 11 real compositions; 20/20 conformance | ESSENTIAL (partial) |
| Product count | 20 products | PARTLY ACCIDENTAL (count itself is not the moat) |
| Total lines of code | large | ACCIDENTAL (explicitly not a moat) |

## What makes reproduction genuinely hard
1. **Agreement across layers.** The state root E21 commits must be the one E22 binds, the one E23
   conserves authority across, and the one the independent verifier recomputes. A re-implementation
   that gets any of these subtly different fails the clean-room verifier.
2. **Negative knowledge.** The invariant/attack corpora encode what was tried and refused. Reproducing
   the code without the adversarial corpus does not reproduce the assurance.
3. **Live promotion discipline.** The shadow→enforce gate, with documented NOT_PROMOTED status, is an
   operational pattern that is easy to claim and hard to execute honestly.

## What is NOT a moat
- Number of products, files, or lines.
- Reachability (HTTP 200) of endpoints.
- Multi-process same-host transport (present, but not independent failure domains).
- Claims of enforcement that are not deployed.

## Honest bottom line
The essential complexity is real and evidenced; the accidental complexity is not counted. The strongest
single barrier is the **clean-room verification of a composed state→authority→execution→evidence
proof** — because it forces every layer to agree, and it is checkable by a party that does not trust
the runtime.
