# CURRENT.md — Single Source of Truth
_Updated: 2026-03-10 18:35 AEDT_

## Direction
- **Whole-fleet main priority: restore the system to 100% operational status before Michael's Saturday meeting.**
- No non-critical work. No business-as-usual drift. No side features. The fleet builds/tests only until the success criteria are met.
- Split the fleet into two modes: **Build** and **Test**.
- Live proof beats status.

## Deadline
- **Hard deadline:** before Michael's Saturday meeting.
- If the system is not 100% operational by then, the business direction is at risk.

## Fleet Structure — Effective Now
### Build Team
- **Builder:** control centre, backend/API wiring, truth-layer/code fixes
- **Sentinel:** provider/infrastructure recovery, degraded-agent recovery, health verification
- **Cog:** backlog cleanup, stale-memory cleanup, ops hygiene, routing/comms enforcement
- **Rivet:** coordination, success-criteria enforcement, integration decisions, blocker escalation

### Test Team
- **Susan:** user-flow testing, task/approval path testing, practical operator sanity checks
- **Radar:** adversarial testing, edge cases, false-green / false-clean detection, external/regulatory relevance checks
- **Harper:** once recovered, finance/legal workflow testing and grant/deadline flow validation
- **Herald:** once recovered, comms/approval workflow testing and delivery-path validation

## Success Criteria
1. **Comms:** one canonical protocol, no buried/ambiguous delivery, ACK/proof discipline working.
2. **Memory/Backlog:** no stale critical inboxes, stale-memory/document drift cleaned up, current operating docs aligned.
3. **Health:** no false green, degraded agents either recovered or intentionally excluded with clear reasons.
4. **Control Centre:** `cc.rateright.com.au` shows current real data, not stale/demo state; core pages are truthful.
5. **Backend/API:** required endpoints for intended workflows work end-to-end, or broken controls are removed.
6. **End-to-End Proof:** a real multi-agent workflow completes cleanly through the current system with proof returned and visible in CC.
7. **Operator Trust:** Michael can rely on CC over legacy dashboards without detective work.

## Active Decisions (Waiting on Michael)
- None. Michael has directed the whole fleet to prioritize full operational restoration until success criteria are met.

## What Each Agent Should Know
- **Builder:** Build Team. Your priority is Control Centre reality, backend/API completion, and remaining truth-layer/code cleanup. No side work.
- **Susan:** Test Team. Your priority is real operator/user-flow testing once build increments land.
- **Harper:** Degraded. Recovery path first; then finance/legal workflow testing.
- **Sentinel:** Build Team. Your priority is provider/infrastructure recovery and live health verification.
- **Radar:** Test Team. Your priority is adversarial/system testing and edge-case detection.
- **Herald:** Degraded. Recovery path first; then comms/approval flow testing.
- **Cog:** Build Team. Your priority is backlog cleanup, stale-memory cleanup, and routing/comms hygiene.
- **Rivet:** Run this as a restoration program with strict proof-based progress until 100% is met.

## Corrections
- Task 1 is functionally complete, but the fleet is not yet at 100% operational capacity.
- Harper and Herald remain degraded on `api_limit_exceeded`.
- `cc.rateright.com.au` is deployed but not yet trustworthy enough to replace legacy dashboards.
- CURRENT.md is the tie-breaker if any other file or agent output conflicts.

## Current Phase
**Full-Capacity Restoration (Build/Test Split)** — no side work until the system is truly operational and the success criteria are met.

---
_Rivet Chief of Staff_