Consensus v1.2 — Cloud Deployment (Render + MongoDB)

Live in the cloud, 24/7

Consensus is now deployed and running around the clock on Render, backed by a durable MongoDB Atlas ledger — no more ephemeral storage, decisions persist across every restart.

What changed

• Hosting: migrated to a Render web service (Socket Mode), running 24/7. Kept awake by an in-app self-ping plus a scheduled wake-up backstop.
• Storage: ledger moved to MongoDB Atlas for durable persistence. The node:sqlite/JSON backend remains the local-dev fallback.
• Reliability: single-instance cut-over (no event duplication); health endpoint for liveness.

Quality

• 144/144 tests passing (incl. live MongoDB backend-parity suite)
• Judge eval on the hosted Cerebras GLM-4.7 model: P=0.966 / R=1.000, classifier 11/11, audit 2/2 — AI prompts unchanged, receipts still valid
• 0 type errors, lint clean

Verified live in Slack

Ambient capture, cross-channel contradiction alerts, permission-redacted provenance, and the consistency audit all confirmed working on the deployed MongoDB build.

Consensus v1.1 — Governance & Lifecycle

Adds the decision governance layer on top of the v1.0 consistency guardian.

New in v1.1

• Decision lifecycle — 7 states (proposed / active / confirmed / exception / expired / rejected / superseded); only enforceable states trigger alerts.
• Authority model — trusted-channel gating (untrusted channels capture as proposed, not enforced) and authorized-owner Confirm/Reject, with a strict production mode (fail-closed defaults).
• Injection hardening — NFKC normalization + 3 new obfuscated-delimiter eval cases; 9/9 injections defeated across all three model stacks (GLM-4.7, gemma, Claude).
• Abuse blunting — per-user dismissal memory, per-author daily capture cap, rate + queue + audit metering.
• Live Real-Time Search in the hosted cloud brain (public-channel-only, fail-open).

Quality

138 tests passing · 0 type errors · CI green · judge eval P/R at/near 1.000 on all stacks.

Built for the Slack Agent Builder Challenge 2026, Track 1.

Consensus v1.0 — Slack Agent Builder Challenge submission

Ambient consistency layer for Slack. Submitted to Track 1, judged build. Eval: GLM-4.7 55/55 P=1.000 R=1.000.