ADR-0008: No live dashboard in MVP

• Status: accepted
• Date: 2026-05-27
• Reaffirmed by: ADR-0019 — the chat agent's coach mode is the formal "what's happening?" surface that replaces a streaming dashboard. The no-dashboard call stands; coach mode covers the gap with grounded, on-demand answers.

Context

A "live trading dashboard" (streaming equity curve, position deltas, fills appearing in real-time) is the natural expectation for a trading product. It's also genuinely expensive to build well:

• WebSocket layer from server to browser
• Per-deployment subscription management
• Concurrency / re-render budget on the client
• Visual design + charting library investment
• Testing under various network conditions

It is also not the critical path to validating the product. The product hinges on whether Skills can be authored, simulated, and deployed safely — not on whether the deployer can watch ticks roll in real-time.

Decision

MVP ships without a live dashboard. The deployment detail page is read-on-demand:

• Current agent_state (single row read)
• Recent decision_snapshots (paginated, manual refresh)
• Recent agent_logs (paginated, manual refresh)
• Recent agent_commands + ack status
• Chat tab for asking the agent about current state

If the deployer wants a fresh state read, they click Refresh. If they want to know "what's happening right now?", they ask the chat agent.

Alternatives considered

Alt A — Full live dashboard with Supabase Realtime subscriptions

• Subscribe to agent_state rows, push deltas to browser
• Build effort: weeks of design + impl + polish
• Not picked: not on the MVP critical path. Defer to Phase 4+.

Alt B — Polling at fixed interval (e.g., every 5s)

• Cheap to build, looks "live"
• Wastes bandwidth + DB load when nobody's looking
• Not picked: not worth the half-measure. If we're going to do live, do it right with subscriptions.

Alt C — Server-Sent Events on demand (open when page open)

• Like Supabase Realtime but built ourselves
• Reasonable middle ground
• Not picked: still on the wrong side of "is this MVP-critical?"

Consequences

Positive

• Weeks saved on MVP build
• Lower hosting cost in MVP (no persistent UI connections)
• Chat agent fills the gap for "what's happening now?" — answering with grounded data, not just a chart
• Forces us to validate the core value (skills + sim + live) without leaning on dashboard sizzle

Negative / trade-offs

• The deployment detail page feels static compared to other trading tools
• Users have to remember to refresh, or click suggested chat prompts
• Some users will ask for it on day one — we'll point them at the roadmap

Things we'll need to revisit

• If user feedback consistently demands real-time visualization → Phase 4 priority
• If we add notifications ("liquidation imminent", "daily loss halt triggered") via email/Discord/Telegram, that may obviate most of the demand for a live dashboard

References

• docs/architecture/live-runtime.md — state writes
• docs/architecture/chat-agent.md — fills the "what's happening?" gap
• docs/product/roadmap.md — dashboard punted to post-MVP