Median $3.28m, range $3.18–3.42m. Each links to the agency-direct listing — not a portal. 3 more comps hidden.
Not in our data: strata isn't relevant (freestanding), but Bronte's flood/overland-flow overlay we don't hold. Confirm the certificate with the selling agent before auction.
Does it fit my brief though?
Fit vs your brief · scoreFit()
★
✓
3-bed house — matches 🔗 listing field
✓
Walk to transit — 650m to Bondi Junction buses, 8 min 🔗 GTFS stop · computed
✓
School catchment — Bronte Public (in-area), top decile 🔗 NSW catchment + SEIFA SA1
✕
Budget — guide $3.60m is over your $3.5m ceiling, and over fair value 🔗 brief + appraisal
Then find me ones under budget that actually fit
Working… building your shortlist
✓
Read brief: 3-bed · <$3.5m · walk · school
✓
Scanned 41 live agency-direct listings
●
Running appraisals & fit on 9 candidates…
Culling overpriced & off-brief
Ranking by value, not date
Pinned artifacts 3 saved
14 Imperial Ave, Bondi
Fair · at comps
7 Knox St, Clovelly
~5% under
12 Hewlett St, Bronte
8–13% over
⚖Appraise
◫Compare pinned
🏠Explain Bronte
✨What's new
ⓘ
Conversation spine
Conservative / shippable take — variant 1 of 3
Core interaction
A familiar chat thread, but the agent never replies with a wall of text. Every answer materialises as a tool-rendered widget from a fixed kit — appraisal card, comp table, fit checklist, honest-gap. The model picks and fills components; it never authors the price or its markup.
Why it fits a buyer's agent
Numbers you'll bid on must be auditable. KNOW-vs-RATE is made visible: the computed fair-value carries a □ computed tag + confidence badge; the agent's read sits softer above it. Every claim has a provenance link to a comp, a catchment, a GTFS stop. Missing data is a designed object, not faked.
Pinned tray
Any artifact pins to a persistent tray that survives the conversation — the durable shortlist. Tap the star on a card. This is the safe bridge from "chat" to "canvas."
Conservative ↔ inventive
Most shippable of the three: the chat metaphor is understood by everyone, latency is masked by the live reasoning step-trace, and the component kit is small and controllable. Trade-off: you still scroll a linear thread — comparison and spatial arrangement are weaker than the board (variant 3).
Real vs faked
REAL Bronte/Bondi/Clovelly suburbs, comp method, $/m², SEIFA decile, GTFS walk times, agency-direct provenance, confidence-from-comp-count.
MOCK Specific addresses, sold prices & images (picsum) are illustrative; no live backend wired.