CAIN-42 evidence library

CAIN-42 Evolution 28 -- Portable Machine Agency & Execution Identity

Evidence bundle for evolution E28: 7 claims, 90 of 90 invariants held.

Last reviewed 2026-10-01

Browse the raw bundle · How to reproduce it · SHA-256 manifest

Claims (7)

Claim	Level
C42-E28-IDENTITY: portable, signed execution identity envelope carried over 23 protocol carriers without changing a byte; authority is always recomputed by the receiver	TESTED
C42-E28-DELEGATION: delegated authority is a subset of the parent on 16 dimensions	TESTED
C42-E28-CONTINUITY: model, runtime, prompt, tool, memory and key changes invalidate authority until explicit re-evaluation	TESTED
C42-E28-RECEIPTS: every governed action, allowed or refused, gets a signed hash-chained receipt; identity events go to an RFC 6962 transparency log	TESTED
C42-E28-E8: every allowed action reached the E8 Action Commit boundary	TESTED
C42-E28-BENCH: 442 of 442 adversarial scenarios contained across 8 categories; 90 of 90 invariants hold	TESTED
C42-E28-SCALE: synthetic in-process scale runs at 10/100/1,000/10,000 agents plus a 100,000-agent virtual registry	SIMULATED

Invariants: 90 of 90 held

Rules the code must never break, each checked across many scenarios. See all 90 on one page.

Niche	Held
Evidence, receipts & proofs	30 of 30
Policy, law & governance	21 of 21
Identity, authority & delegation	18 of 18
Autonomy, control loops & recovery	10 of 10
Consensus & distributed systems	4 of 4
Tools, MCP, protocols & adapters	3 of 3
Transactions, markets & economics	2 of 2
Prediction, world models & simulation	1 of 1
Memory, data & privacy	1 of 1

Verify it in your browser

Your browser downloads the bundle's SHA256SUMS manifest and the file(s) behind this page, hashes them with SHA-256 locally (WebCrypto), and compares. A match shows the record you are reading is the published one; it does not by itself prove who published it (see the signed claims registry and the bundle verifier for that).

Known limitations

E28 is an in-process library. It is not wired into the hosted gateway, MCPGate or the PBFT clusters; nothing on the three public sites runs it live.
The identity envelope is a CAIN experimental reference protocol, not a standard. STANDARDS maps it to external work; no external party has adopted or reviewed it.
Zero-knowledge authorization proofs are NOT implemented. Selective disclosure uses salted hash commitments (the SD-JWT idea): undisclosed values are hidden, but nothing about them is proven.
Hardware attestation is UNKNOWN everywhere. 'Attestation level 1' means CAIN compared self-reported measurements with registered values; it is not a measured boot or TEE quote.
Fork and clone detection only sees what it is told. Without an exclusivity declaration and located, session-tagged observations the verdict is UNKNOWN.
Cross-domain revocation does not propagate: if domain A revokes an identity, domain B's local identity made from it stays valid until B revokes it or its B-local lease/delegation expires.
Key rotation bumps the identity version but does not re-key the underlying E25 instance, so a rotated identity cannot act until it is re-registered.
Receipt-chain truncation (dropping the newest receipts) is only detectable against an anchored head (an AGGREGATE receipt or a transparency-log tree head the verifier already holds).
Research radar, TypeScript SDK, OAuth integration, hosted/private deployment profiles and a live HTTP endpoint for E28 were not built in this evolution.
Scale numbers are synthetic, in one process on a shared 3 GB VPS; 100k is registry+log+lineage only.
Nothing here is externally reviewed, formally verified, certified or production-ready.

The bundle's own README