Eastern Subs board
| Guide | $3.30m | $3.45m | $3.60m |
| Comp fair □ | $3.48m | $3.44m | $3.28m |
| vs fair | 5% under | at fair | 8–13% over |
| $/m² | $7.6k | $9.1k | $8.7k |
| Land | 434m² | 372m² | 412m² |
| Walk | 7m | 9m | 8m |
| Decile | 10 | 9 | 10 |
| Confid. | High 7 | Med 4 | High 6 |
| Read | Better | thin comps | over |
Comparison board
Core interaction
A spatial, draggable board — no chat thread at all. Listings, the compare matrix (centrepiece), comp-provenance, and the agent's "working" node all live as cards on an infinite canvas you arrange in 2-D. The agent annotates items in place with pinned notes ("Best buy", "guide optimistic"), wired to the cards they describe. A command bar + chips drive it. Try dragging a card.
Why it fits a buyer's agent
Buying is a comparison + spatial-memory task. A board lets a couple lay out the contenders, see the comp-fair-value verdict on each, and read the agent's in-context judgement — KNOW (computed □ cells, provenance node) vs RATE (the purple agent annotations) physically separated. The matrix is the durable, shareable artifact.
Conservative ↔ inventive
Furthest from chat — highest ceiling, highest risk. Powerful for power-users arranging a shortlist; the open canvas re-introduces some of the "blank space, what do I do" discoverability problem (mitigated by pre-populating it and the command chips). Hardest to build on Lynx (drag, zoom, wiring). Pair with v1/v2 as the default entry.
Real vs faked
REAL Clovelly/Bondi/Bronte, comp-fair-value method, $/m², SEIFA decile, walk times, agency-direct provenance, confidence-from-comps. Drag is genuinely functional.
MOCK Addresses, prices, images (picsum); zoom + wiring are illustrative; no backend.