CAIN-42 CAIN Studio
Sign inGet started
Menu
25CAIN-42 Top 25

The CAIN-42 Top 25

The 25 parts of CAIN-42 we think matter most: the most original, the easiest for an outside engineer to check, and the closest to paid work. Ranked best first.

How to read this list: every product page shows a live demo against the real route, the evidence, how to verify it yourself, and its limitations. The "our own check" results were run by us and are not independent. Products marked "being upgraded" are being reworked one at a time and get their full site when the work is done.
  1. 1

    KEYSTONE 13 pages

    Detect, AI-fix, sandbox-prove, sign: a fix you can re-check

    Our own live check, 2026-11-02: bundle verifies; 4/4 current-rule fixes re-run identically

  2. 2

    Quorum-QaaS (4-region PBFT) 14 pages

    Byzantine notarization with offline-verifiable certificates

    Our own live check, 2026-11-02: live; DEGRADED 11-02 (1 of 4 replicas down)

  3. 3

    Offline verifiers (NULLIUS) 14 pages

    Standalone checkers that import no CAIN code

    Our own live check, 2026-11-02: 36/36

  4. 4

    FALSEWORK 13 pages

    Deterministic detector for fake-but-finished code

    Our own live check, 2026-11-02: 44/44; benchmark precision 1.0, blocking recall 0.548

  5. 5

    cain_audit deliverables 12 pages

    One command from engine to signed client deliverable

    Our own live check, 2026-11-02: self-test 6/6

  6. 6

    VeriCoding (Z3) 13 pages

    Real proofs or real counterexamples for Python functions

    Our own live check, 2026-11-02: 14/14

  7. 7

    BROKER 13 pages

    One-shot credentials bound to one exact request

    Our own live check, 2026-11-02: 17/17

  8. 8

    CODEGUARD 13 pages

    Firewall for coding-agent shell commands

    Our own live check, 2026-11-02: blind set D 26/30, 2/40 false alarms; 273 tests

  9. 9

    APPROVAL 13 pages

    N-of-M human approval that becomes one credential

    Our own live check, 2026-11-02: blind set 19/26 held before fixes, 26/26 after; 94 tests

  10. 10

    Black Box flight recorder 14 pages

    Merkle-sealed agent recordings with provable GDPR erase

    Our own live check, 2026-11-02: blind set 16/23 held before fixes, 23/23 after; 64 tests

  11. 11

    Signed decision receipts 13 pages

    Every allow/deny decision signed

    Our own live check, 2026-11-02: blind set 15/18 held before fixes, 18/18 after

  12. 12

    SPEND GOVERNOR 14 pages

    Budget breaker with no overspend under races

    Our own live check, 2026-11-02: blind set 17/20 held before fixes, 20/20 after; 21 tests

  13. 13

    Policy Red-Team 14 pages

    Finds the bypass, writes the patch, proves it

    Our own live check, 2026-11-02: blind set 2/14 held before fixes (all 12 bypasses survived), 14/14 after; 57 tests

  14. 14

    AETHER-33 memory 16 pages

    Governed agent memory that never grants authority

    Our own live check, 2026-11-02: blind set 14/22 held before fixes, 22/22 after; 33/33 invariants

  15. 15

    DriftGuard (MCP pinning) 12 pages

    Catches MCP tool rug-pulls

    Our own live check, 2026-11-02: blind set 8/18 held before fixes, 18/18 after; 23 tests

  16. 16

    KEYSTONE CI 10 pages

    Signed CI verdicts bound to the exact code

    Our own live check, 2026-11-02: blind set 15/27 held before fixes; 27/27 on 1.1.0; 31 tests; 0 verdict changes on 150 real commits

  17. 17

    Fail-closed doctrine 8 pages

    UNKNOWN and ERROR never become ALLOW

    Our own live check, 2026-11-02: route sweep 1,440 malformed requests, 0 yes, 0 crashes; verdict-function blind set 5/18 held before fixes, 18/18 after

  18. 18

    VibeGuard 8 pages

    Security review of AI-written Python

    Our own live check, 2026-11-02: blind set 3/12 held before fixes; 12/12 after; 49 tests

  19. 19

    SurfGuard 8 pages

    Human-approval gate for browser agents

    Our own live check, 2026-11-02: blind set 4/12 held before fixes; 12/12 after; 32 tests

  20. 20

    OpenClaw 8 pages

    Scanner for agent skills and plugins

    Our own live check, 2026-11-02: blind set 2/13 held before fixes; 13/13 after; 85 tests

  21. 21

    AGENT-CI 8 pages

    Agent regression tests with no AI judge

    Our own live check, 2026-11-02: blind set 4/11 held before fixes; 11/11 on 1.1.0; 84 tests

  22. 22

    ACTUARY 7 pages

    Fail-closed fleet risk score with replay proof

    Our own live check, 2026-11-02: blind set 5/13 held before fixes; 13/13 after; 35 tests

  23. 23

    Swarm KEYSTONE 8 pages

    Multi-agent safety checks

    Our own live check, 2026-11-02: 19/19

  24. 24

    Deterministic math gates 8 pages

    Fairness, sample size, cycle checks anyone can recompute

    Our own live check, 2026-11-02: blind set 9/19 held before fixes; 8 gaps fixed, 54 engine tests

  25. 25

    Verify-yourself page 10 pages

    11 live checks from the visitor's own browser

    Our own live check, 2026-11-02: blind set 6/10 held before fixes; page 11/11 live in a browser after

FIRME AI CORPORATION · CAIN-42 is a Byzantine trust fabric. Numbers marked "our own check" were run by us, not by an independent party. Nothing on these pages is a certification.