FORGE -- Source of Truth
Master tracking document — 2026-04-01 — Generated at
Forge v0.3.0 | Sprint state: BUILD | 12 services online | 10 memory providers
1 SYSTEM HEALTH — Live Service Status
Core Infrastructure
Memory Router:3200OK
API Bridge v0.3.0:3100OK
OpenClaw Gateway:18789LIVE
OpenViking v0.2.15:1933OK
Hermes Track Managers
Hermes RMT:18800OK
Hermes Identity:18801OK
Hermes Lottery:18802OK
Hermes x402:18803OK
Hermes Orchestrator:18810OK
LaunchAgents
com.forge.api-bridgePID 9277
com.forge.memory-routerPID 75082
com.forge.hermes-rmtPID 1034
com.forge.hermes-identityPID 1072
com.forge.hermes-lotteryPID 1065
com.forge.hermes-x402PID 1062
com.forge.hermes-orchestratorPID 1039
com.forge.infra-servicesPID 1035
com.forge.mlx-serverPID 1055
com.forge.paperclipPID 1067
com.forge.mirosharkEXIT 127
com.forge.x-bookmarks-digestIDLE
2 MEMORY INFRASTRUCTURE — 10 Providers
| Provider | Status | Entries / Size | Transport | Notes |
|---|
| CXDB |
ACTIVE |
51,000 entries / 57,000 edges |
SQLite + FTS5 |
Primary context store, graph + attestations |
| hermes-memory |
ACTIVE |
17 entries |
HTTP :18810 |
Cross-track synthesis, CTO reporting |
| hermes-state |
CONNECTED |
Read-only |
SQLite (forge-state.db) |
Track state, experiment results |
| OpenViking v0.2.15 |
ACTIVE |
Dynamic |
HTTP :1933 |
Long-term associative memory |
| claude-memory |
ACTIVE |
Dynamic |
MCP |
Claude native project memory |
| sprint-state |
ACTIVE |
Current sprint |
JSON file watch |
sprint-checkpoint.json |
| git-history |
ACTIVE |
Full log |
Git CLI |
Commit history, diffs |
| working-memory |
ACTIVE |
3 streams |
Markdown files |
Sprint, research, git work streams |
| session-logs |
ACTIVE |
Append-only |
JSONL |
events.jsonl audit log |
| rmt-reputation |
ACTIVE |
Track-scoped |
SQLite |
RMT reputation model state |
| Mem0 / Qdrant |
NOT DEPLOYED |
— |
Docker (planned) |
Phase 1a incomplete — deprioritized for Hermes |
3 RESEARCH TRACKS — 4 Active
RMT Reputation
0.9397
STOPPED
| Composite Score | 0.9397 |
| Sybil Detection | 93.7% |
| Iterations | 38+ converged |
| Status | Research loops stopped, needs real data |
Identity Attestation
0.4230
STOPPED
| Composite Score | 0.4230 |
| Architecture | 6-dim orthogonal |
| Params Locked | 11 changed from default |
| Status | ADRs complete, awaiting calibration |
x402 Trust Channels
RESEARCH
STOPPED
| Phase | Research / competitive analysis |
| Analysis | Strategic analysis complete 2026-03-29 |
| Config | loop-config.yaml ready |
| Status | Awaiting research loop activation |
Lottery
3 ITER
STOPPED
| Iterations | 3 completed |
| Registry | Parameter registry complete |
| Config | loop-config.yaml updated |
| Status | Awaiting research loop restart |
4 PHASE COMPLETION
Phase 1: Infrastructure Sprint 14/15
- ✓ 1a: Activate OpenViking memory provider
- ✓ 1a: Connect Hermes state.db as Memory Router provider
- ○ 1a: Deploy Mem0 + Qdrant stack DEPRIORITIZED
- ✓ 1b: Replace research.yaml with tracks.yaml
- ✓ 1b: Build forge-state.sqlite with Data Access Layer
- ✓ 1b: Convert dispatch-review to async PR-based
- ✓ 1b: Fix 47 ARCH findings
- ✓ 1c: Build eval-matrix.ts (5-dimensional)
- ✓ 1c: Build real-data loader for research calibration
- ✓ 1d: Verify and enforce state machine sequential flow
- ✓ 1e: Build HermesSupervisorHandler (event-driven)
- ✓ 1f: Visualization infrastructure (visual-explainer + CF deploy)
- ✓ Enforcement hooks via .claude/settings.json
- ✓ MiroFish/MiroShark rename + simulation scaffold
- ✓ P0 security + P1 type error fixes from Codex review
Phase 2: Research Activation 4 PENDING
- ○ Restart session (activate enforcement hooks)
- ○ Propose RMT hypotheses with visual-explainer
- ○ Run RMT research loop with real data + eval matrix
- ○ Fix Grok's 6 architecture review items
Phase 3: Product Ship 2 PENDING
- ○ Ship Sybil Oracle (first product)
- ○ Configure multi-chain indexer for real data
Phase 4: Scale 6 PENDING
- ○ Mainnet deployment
- ○ Multi-chain indexer production
- ○ Identity attestation product launch
- ○ x402 trust channel integration
- ○ Lottery system production
- ○ Cross-track synthesis dashboard
5 TODAY'S COMMITS — 12 Total
| Hash | Type | Description |
|---|
| 8de3584 | feat | enforce visual-explainer v0.6.3 + mandatory visualization rules |
| 54b6a71 | docs | Phase 2 handoff doc + phase roadmap visualization |
| 77d4f6a | docs | Phase 1 infrastructure sprint retrospective |
| df19b44 | fix | P0 security + P1 type errors from Codex code review |
| ff04305 | feat | add --track scoping to dispatch-review |
| 230f571 | feat | ARCH fixes + HermesSupervisorHandler + real-data loader |
| e155411 | feat | activate enforcement hooks via .claude/settings.json |
| a940803 | feat | forge-state.sqlite + hermes-state provider + eval-matrix 5-dimensional |
| 3abf643 | feat | tracks.yaml + OpenViking activation + new-track.ts update |
| c0ecfa8 | refactor | rename MiroFish → MiroShark throughout codebase |
| b5a72cf | fix | wire Swarma state transitions + BUILD_NEEDED + CONVERGE to Memory Router event bus |
| d494681 | feat | full-day strategic research + infrastructure wiring (2026-04-01) |
6 CODE REVIEW STATUS
Codex (GPT-5.4)
CONDITIONAL FAIL → FIXED
- ✓ P0: SQL injection in cxdb-provider (FIXED)
- ✓ P0: Missing auth on /health endpoints (FIXED)
- ✓ P1: Type assertion errors in bridge (FIXED)
- ✓ P1: Unhandled promise rejections (FIXED)
Grok (CTO)
REQUEST_CHANGES
- ○ Hermes state persistence across restarts
- ○ Circuit breaker config externalization
- ○ MiroShark error recovery paths
- ○ Cost tracker budget alerting thresholds
- ○ OpenViking memory TTL configuration
- ○ Cross-track event ordering guarantees
Internal (Claude)
PASS
- ✓ State machine engine — PASS
- ✓ Memory Router providers — PASS
- ✓ API Bridge routing — PASS
- ✓ Eval framework — PASS
- ✓ Research harness — PASS
7 COMPETITIVE POSITION
| Platform | Agents | Architecture | Status | Our Advantage |
|---|
| Forge (Us) |
0 mainnet |
Decentralized, 4-model consensus, graph-based |
INFRA READY |
— |
| Helixa |
132K indexed |
Centralized, capped at 50 active |
LIVE |
Decentralization, no agent cap, sybil resistance |
| ERC-8004 |
130K+ |
Registry standard, no reputation |
LIVE |
Reputation layer, trust scoring, identity attestation |
| Virtuals Protocol |
18K |
Token-bonded, revenue collapsed |
DECLINING |
Sustainable model, not token-dependent |
8 NEXT ACTIONS
P0 — Immediate
Restart session to activate enforcement hooks (.claude/settings.json now configured with pre-commit review guard + session-start boot)
P0 — Research
Propose RMT research hypotheses for user approval using visual-explainer v0.6.3 for presentation
P1 — Validation
Run RMT research loop with real Shyft/mainnet data + full 5-dimensional eval matrix
P1 — Architecture
Address Grok's 6 REQUEST_CHANGES items (state persistence, circuit breakers, error recovery, budget alerts, TTL config, event ordering)
P2 — Product
Ship Sybil Oracle as first product after research validates RMT composite above 0.95 threshold
P2 — Infrastructure
Fix MiroShark LaunchAgent (currently EXIT 127) — likely missing Python path
9 KEY DECISIONS LOG — 4-Model Convergence
DEC-001: Hermes over Mem0/Qdrant for track memory
Rationale: Hermes provides per-track SQLite persistence with HTTP API, eliminating Docker dependency. Mem0/Qdrant deprioritized. Consensus: Claude + Codex + Gemini + Grok (4/4).
DEC-002: OpenViking as long-term associative memory
Rationale: Native integration, no infrastructure overhead, v0.2.15 stable. Replaces planned vector DB for associative recall. Consensus: 4/4.
DEC-003: 5-dimensional eval matrix over single-metric
Rationale: Composite scoring across correctness, robustness, efficiency, security, and architecture prevents metric gaming. Drift detection catches regression. Consensus: 4/4.
DEC-004: Research before product (validate, then ship)
Rationale: Running research loops with real data before shipping Sybil Oracle ensures product claims are defensible. Competitive analysis shows no urgency to rush. Consensus: 4/4.
DEC-005: MiroFish → MiroShark rename
Rationale: Brand differentiation from upstream MiroFish project. Apple Silicon fork with adversarial simulation focus warranted distinct identity. Consensus: 3/4 (Grok abstained).
DEC-006: Enforcement hooks over advisory warnings
Rationale: Pre-commit review guards and session-start boot diagnostics prevent drift. Advisory warnings were routinely ignored. Consensus: 4/4.
10 DASHBOARDS & KEY FILES
Dashboards (Cloudflare Pages)
Key File Paths
.forge/machines/sprint.yaml — Sprint state machine.forge/machines/research.yaml — Research state machine.forge/state/sprint-checkpoint.json — Current sprint state.forge/events.jsonl — Audit log (append-only)infrastructure/config/hermes.yaml — Hermes configurationinfrastructure/config/models.yaml — Model dispatch configresearch/config/research.yaml — Research track registryforge-core/src/cli/forge.ts — CLI entry pointinfrastructure/packages/memory-router/config/router.yaml — Router configinfrastructure/packages/api-bridge/src/model-dispatch.ts — Model routing