Skip to content

🔄 L-CHAT-2: Triple Ratchet 1:1 — PQ-FS + PQ-PCS (X3DH + ML-KEM step) #30

Open

Description

@gHashTag

Lane L-CHAT-2

Parent EPIC: #28 (Trinity Secure Chat)
Anchor: φ² + φ⁻² = 3
ETA: 7 days

See parent EPIC #28 for full mission scope, threat model (TM-1..TM-10), R-CHAT laws (R-CHAT-1..R-CHAT-12), acceptance gates G-C1..G-C10, and 21 cited sources.

Detailed design: docs/chat/trinity-chat-design.md (added in PR feat/trios-chat-epic).

Coordination Protocol

  • 🛡 claim · 💓 heartbeat (60min) · ✅ done · 🚫 block · 🔓 release after 4h silence

φ² + φ⁻² = 3 · TRINITY · CHAT · NEVER STOP

Activity

  1. gHashTag commented on May 9, 2026

    @gHashTag
    CollaboratorAuthor

    🔥 IGNITE — Trinity Secure Chat scaffold landing

    Anchor: φ² + φ⁻² = 3 · TRINITY · CHAT · ZERO-METADATA

    PR: Trinity-S-AI-DNA/trios#631
    Local trios tracker: Trinity-S-AI-DNA/trios#632

    Verification snapshot [VERIFIED]

    • cargo test -p trios-chat --lib — 35/35 pass
    • cargo run --bin e2e_chat_25 — 25/25 pass
    • cargo run --bin falsifier_runner — direct 100 % / indirect 90 % / multi-turn 100 %
    • cargo test --tests r_chat_guard — 3/3 pass
    • 6 Coq theorems Defined + 1 Admitted (R5 budget honored)
    • 200-attack falsifier corpus committed
    • 10 ADR-CHAT-001..010 + 29 KB design doc with 21 sources

    Honesty (R5)

    Lanes 3 (MLS) and 5 (Persistence) are explicitly [ASPIRATIONAL] in this
    scaffold — ADR-CHAT-001 / ADR-CHAT-008 capture the design; full impl lands
    in follow-up PRs. Coq INV-CHAT-4 (sender unlinkability) is Admitted
    within the documented R5 budget.

    L1 ✅ no .sh · L2 ✅ Closes #632 in PR body.

  2. gHashTag commented on May 9, 2026

    @gHashTag
    CollaboratorAuthor

    🔥 Wave-2 — harden + tighten

    Anchor: φ² + φ⁻² = 3 · TRINITY · CHAT · ZERO-METADATA

    Wave-2 PR: Trinity-S-AI-DNA/trios#633
    Builds on PR #631 (scaffold).

    Verification snapshot [VERIFIED]

    • 51/51 lib tests · 25/25 e2e · 3/3 guard
    • Falsifier: 200/200 blocked (100%) — capability_abuse 10%→100%, indirect 90%→100%
    • Coq: 9 Defined + 1 Admitted (R5 budget 1/10)

    Lifted from [ASPIRATIONAL] in this PR

    • L-CHAT-2 ratchet DH step (X25519 mix, skipped-keys cache)
    • L-CHAT-3 MLS group skeleton with strict epoch monotonicity
    • L-CHAT-5 Persistence (MemoryStore VERIFIED, PostgresStore schema)

    L1 ✅ · L2 ✅

  3. gHashTag commented on Jul 7, 2026

    @gHashTag
    CollaboratorAuthor

    ✅ L-CHAT-2 implemented → trios-mesh PR #21

    PR: https://github.com/gHashTag/trios-mesh/pull/21
    Files: src/chat/x3dh.rs + src/chat/ratchet.rs

    PQXDH (X25519×4 + ML-KEM-768 → HKDF root) + triple ratchet (symmetric + DH + KEM). Deniable (no per-message signature). decrypt() stages-then-commits so a forged message can't desync state.

    Gate: G-C2 (FS/PCS) + G-C10 (no per-msg sig on the wire).

    Part of EPIC #28. 163 tests green · fmt + clippy -D clean.

  4. dmitrii-f-t27 commented on Oct 11, 2026

    @dmitrii-f-t27
    Collaborator

    Checked against gHashTag/trios main at 8b229e9 (2026-09-24). Trios CI on that commit is green: run 36028910601 runs the unit tests of crates/trios-chat/rings/*.

    The 2026-07-07 comment points to gHashTag/trios-mesh#21. That repository returns 404 for this account, so I cannot confirm that PR was merged, and it is not counted here. Leaving this open.

    On trios main: rings/CR-CHAT-02 has the root/chain KDF, dh_kem_step (takes a 32-byte KEM secret), FS/PCS key-divergence tests (rfs_01..rfs_05, post_compromise_security_after_dh_step_recovers) and deniable::den_06_no_per_message_signature. They pass in CI.

    Missing:

    • The ring calls itself a "Triple Ratchet skeleton" and tags the full Double/Triple Ratchet [ASPIRATIONAL] (lib.rs header).
    • No PQXDH handshake from a published bundle. The tests feed dh_kem_step fixed bytes, not an ML-KEM-768 encapsulation (and the bundle's ML-KEM key is a placeholder, see 🔐 L-CHAT-1: Identity & Onboarding — Ed25519 + X25519 + ML-KEM-768 prekey bundle #29).
    • G-C2 is shown as key inequality, not as a past message that fails to decrypt after compromise. G-C10 is not checked on a wire dump.

    Dmitrii Fedorov (@dmitrii-f-t27)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions