How Foresyte does the maths.

Each pick is a dollar prediction. When the property sells, your prediction is compared to the sale price as a percentage of the actual. Within 15% of actual gets you points; beyond 15% gets you zero. The closer you call it, the more points you score.

base_points = max(0, 10 × (1 − |predicted − actual| / actual / 0.15))
final_points = base_points × confidence_multiplier × no_guide_multiplier

Multipliers stack:

• Confidence: Low 0.5× / Medium 1.0× / High 1.25×. High-confidence multiplies your win andyour error — picking High when you’re unsure is a fast way to lose points.
• No guide: 1.5× when the agent published no price guide. The hardest reads pay best.

Theoretical max on a single pick is 18.75 points (perfect prediction · High confidence · no guide). Realistic upper-end Saturday is in the 30s.

Auctions don’t always settle on Saturday. Foresyte resolves every pick into one of five states:

A bonus pick means you choose six of the next twenty instead of five. They compound across weeks, so a four-voided week credits four bonus picks the following Friday.

Your calibration score is the mean absolute percentage error (MAPE) across your resolved predictions over the trailing three months. ELI5: average how far off you were each time, in percent. Lower is better.

You stay Unrateduntil you have at least 20 resolved predictions — calibration is unreliable below that. Tiers re-anchor at three months once we see real MAPE distribution; for now these thresholds are the founder’s informed guess at where “quite good” lives.

A player who has 20+ resolved predictions in a single LGA at under 6% MAPE earns domain expert status for that LGA. Their predictions in that LGA contribute at 1.5× to the suburb aggregate, on top of (or instead of) their overall tier weight.

The reason is honest: a Platinum-tier player whose 47 resolved predictions are all in the Inner West shouldn’t be weighted as an authority on Cronulla. The leaderboard badge stays based on your overall tier; the per-LGA tier sets your weight inside that suburb’s crowd aggregate.

Your MAPE is recency-weighted across the trailing 3-month window: last week’s resolved picks count more than picks from three months ago. Specific half-life is locked at 6 weeks at launch and revisits at month 3 once we see real volume.

Crowd aggregates apply the same recency weighting to predictions: a high-credibility tipster’s estimate from last Wednesday counts more than the same tipster’s estimate from January.

For every scheduled Sydney auction, the crowd estimate is computed as a single credibility-weighted aggregate. The number is the weighted median — not the mean — of all predictions submitted before lock, with each prediction weighted by:

• The player’s overall calibration tier (§3)
• Their per-LGA expertise on this suburb (§4)
• The confidence they assigned the pick (Low / Medium / High)
• Recency half-life decay (§5)

Each crowd estimate carries a confidence pill — High, Moderate, or Low — derived from how many high-credibility contributors weighed in, how recent their predictions are, and how tight the IQR (interquartile range) is around the median.

A High-confidence signal requires all three:

• ≥100 high-credibility contributors
• IQR width <8% of the weighted median
• ≥1 Platinum-tier contributor with per-LGA expertise on the suburb

About 8–15% of Sydney auctions sell with the price withheld. NSW Valuer-General catches up 6–12 weeks later. In the immediate window, players sometimes know — a neighbour mentioned theirs, a partner works in real estate, a mate was at the open. The Submit a tip CTA appears only on withheld auctions, and only after the auction has been called.

A tip is weighted by source category × the tipster’s credibility multiplier:

Tipster credibility starts at 0.3× for everyone at launch. Tips back-test against NSW VG when settled-sale data lands ~90 days later; accurate tips lift the multiplier toward 1.5×, off ones drop it toward 0.2×.

A community estimate publishes only when all three thresholds are met:

• ≥3 tips from distinct accounts
• Credibility-weighted equivalent of ≥1.5 high-credibility tipsters
• IQR width ≤12% of the credibility-weighted median

When all three are met, the published estimate is an A$100k-bucketed range — never a single number — with a confidence label:

If a threshold fails, the surface says one of: “Limited tips received — no estimate yet”, “Tips received but inconsistent — investigating”, or “Awaiting NSW Valuer-General confirmation”. Honest is better than authoritative-sounding.

Thresholds revisit at month 6 against actual tip volume. If we’re consistently getting 10+ tips per withheld auction with tight IQRs, we tighten back toward 5-tips / 8%-IQR.

Every published number is what the calibrated crowd thinks — never a verdict on what a property is worth. That distinction matters legally (Australian Consumer Law s 18 governs “misleading or deceptive” conduct) and matters culturally: we’re a prediction game, not an automated valuation model. Foresyte doesn’t replace a buyer’s agent, a building inspection, or a strata report.

It’s also unconditionally free to play. No entry fees, no cash prizes, no peer-to-peer wagering, no “shares” in outcomes. Foresyte sits outside Australian gambling regulation by construction — and the verbs we use here (predict, reckon, call it) reflect that.