Forge Strategic Roadmap

Phase 1: Infrastructure Sprint

COMPLETED

✓

Forge state machine engine + CLI INFRA

forge-core: transition, validate, verify, boot, evals, dashboard commands
✓

Cross-model review dispatch INFRA

dispatch-review with deliberation protocol, --track scoping, Grok CTO binding
✓

API Bridge (port 3100) INFRA

OpenClaw relay, model dispatch, intent classification, rate limiting, sanitizer
✓

Memory Router (port 3200) — 9 providers INFRA

CXDB, Hermes, OpenViking, Git, Sprint, Forge, Research, Event Bus, Working Memory
✓

CXDB context database — 51K entries INFRA

SQLite + FTS5, graph edges, cross-model attestations, backed up
✓

Credential Broker + ACL enforcement SEC

macOS Keychain integration, per-service ACL, secrets never in env vars
✓

Cost Tracker (port 3500) INFRA

Cross-track budget monitoring, per-model cost aggregation, budget alerts
✓

Hermes track managers (ports 18800-18803) INFRA

RMT, Identity, Lottery, x402 — persistent memory per track
✓

Hermes Orchestrator (port 18810) INFRA

Cross-track synthesis, budget enforcement, CTO reporting
✓

MiroShark simulation engine RESEARCH

Swarm intelligence adversarial topology generation (renamed from MiroFish)
✓

Eval framework — 5D matrix INFRA

Matrix runner, drift detection, ablation, visualization, state compliance suite
✓

Enforcement hooks (settings.json) SEC

pre-commit review guard, session-start boot check, OpenClaw relay enforcement
✓

tracks.yaml + OpenViking activation INFRA

Single config registry, new-track scaffolding, OpenViking memory provider
✓

4-model strategic convergence RESEARCH

Claude + Codex + Gemini + Grok converged on architecture, revenue model, execution plan
⋮

Mem0 + Qdrant semantic search DEFERRED

Deferred to Phase 2 — requires Qdrant deployment + Mem0 vector integration

Phase 1 Retro Findings

Built 18 services in one sprint but research loops were never run with real data
Sprint state stuck in BUILD since March 29 — enforcement hooks now prevent this
MiroFish renamed to MiroShark but some references may linger
4-model review process works but adds ~15 min latency per review cycle
Cost tracking revealed local model inference is 23x cheaper than cloud for research

Phase 2: Research Loops

NEXT UP

1

Restart session with enforcement hooks active INFRA

New session reads settings.json, activates /sprint-dev-process + /visual-explainer. Boot check runs.
2

Propose RMT hypotheses to user RESEARCH

Present hypotheses with budget estimates + /visual-explainer output. User approves before any experiment runs. Use 132K Helixa agent data as baseline.
3

Run RMT research: real data + eval matrix + MiroShark RESEARCH

132K agent graph, PPR convergence, sybil detection accuracy, compound attack resilience. 5D eval matrix scoring.
4

Run Identity research: 6-dimension composite RESEARCH

Biometric, social, behavioral, institutional, network, temporal dimensions. Sybil farm detection with MiroShark scenarios.
5

Run x402 research: trust channel validation RESEARCH

PaymentModule routing, trust scoring, channel capacity optimization. MiroShark payment flood scenarios.
6

Fix Grok's 6 architecture review items SEC

P0 security fixes, type errors, architecture alignment items from Codex code review.
7

Deploy Mem0 + Qdrant for semantic search INFRA

Deferred from Phase 1. Vector storage for CXDB semantic queries. Qdrant container + Mem0 integration.

Phase 3: Ship Products

BLOCKED ON P2

⋯

Ship Sybil Oracle PRODUCT

ERC8004Bridge + ScoreApi + 150K agent onboarding. Free tier: score lookups. Pro: API access + real-time feeds. Requires validated RMT research from P2.
⋯

Ship Identity Portal PRODUCT

Audit 304-commit antilles-v2 branch. VerificationEngine with 6D composite scoring. Requires validated Identity research from P2.
⋯

Ship Trust Channels PRODUCT

PaymentModule + ExternalRoutingAdapter + ShyftTrustHook. x402 payment protocol integration. Requires validated x402 research from P2.

Phase 4: Advanced Capabilities

FUTURE

⋯

Credit product: TrustCreditLine PRODUCT

Under-collateralized credit lines based on reputation scores. FICO-style revenue model.
⋯

Session keys: SessionKeyRegistry PRODUCT

Delegated transaction signing with trust-weighted session permissions.
⋯

VeriScope compliance: EU AI Act (Aug 2026) SEC

AI Act compliance module. Explainability for reputation scores. Audit trail for all scoring decisions.
⋯

ve-staking for Trust Anchors PRODUCT

Vote-escrowed staking for Trust Anchor status. Economic alignment + governance weight.
⋯

Cross-chain deployment INFRA

Deploy reputation contracts across EVM chains. Bridge reputation scores cross-chain.
⋯

Economic flywheel activation PRODUCT

Free Oracle -> Pro API -> Credit -> Staking -> Governance. Self-sustaining revenue loop.

User Workflow at Each Phase

Phase 1 (Done)

Reviewed architecture proposals
Approved 4-model convergence plan
Validated service health checks
Confirmed enforcement hooks

Phase 2 (Next)

See hypothesis proposals with budgets
Approve/reject each hypothesis
View visual experiment results
Approve research conclusions
Review Grok architecture fixes

Phase 3 (Ship)

Approve product feature scope
Review cross-model code reviews
Validate deployment configs
Approve mainnet deployment

Phase 4 (Advanced)

Approve credit risk parameters
Review compliance audit results
Approve cross-chain deployment
Validate economic model