Will my bid win? · 1465 23rd Ave · hot-market edition

Three price models, rolled forward into the market you'll actually bid in → the probability a given bid takes the house · as of 2026-06-09 · value 1465 → · $2.8M cousins → · when to buy → · browser →

The target — 1465 23rd Ave: 4 bd / 4 ba · 2,389 sqft · renovated-luxury · listed $2.30M ($961/sqft, a deliberate under-ask). "Win" means your bid beats the highest competing offer — approximated as P(final price ≤ your bid). The market is rising fast (our comps measure 0.9%–2.1%/month, and IPO money is about to land), so every number here is rolled forward to the expected late-July print. The unadjusted read stays on the page in gray.

Bottom line

Predicted sale price: median $2.89M (80% interval $2.31M–$3.46M). The backward-looking read says $2.69M; the gap is two more months of a market rising 18%/yr.
Price-to-win: coin flip $2.89M · 70% $3.11M · 80% $3.24M · 90% $3.46M · 95% $3.66M.
The twins' May prints ($2.73M, $2.80M) now read below coin-flip — on purpose. Matching the twin is bidding May's price in July.
No bid buys certainty: there's a ~15% chance one must-deploy IPO buyer shows up and breaks every model. Above ~$3.30M you're buying insurance, not value.
Terms are worth ~1–3%. An all-cash, no-contingency $3.24M plays like ~$3.31M financed.

The three models — and why they disagree

Each model prices 1465 from a different slice of the data; the spread between them is the honest uncertainty. All three are shown after the roll-forward — each card notes its "was".

Average-reno · believes $2.54M · 80% interval $1.89M–$3.42M · was $2.45M · n=1,449

The 8-year market model applied to 1465's specs. "Renovated" is a single category to it, so it prices 1465 like an average remodel — the skeptic's view, blind to the designer finish. Both twins printed $2.73M+ against its $2.45M belief, so it carries only 20% of the ensemble. Roll-forward: +3.6% (2.5 months stale).

Nicest-house · believes $2.90M · 80% interval $2.50M–$3.30M · was $2.75M · n=14

1465's block-level comps: today's-$ $/sqft vs finish score, read at 1465's top-of-set finish of 95 — $1215 ± 124/sqft. The believer in the finish premium and the most confident model: it writes off bids under $2.70M, then rewards each extra $100K fastest. Roll-forward: +5.3% (3.7 months stale).

Bidding-war · believes $2.92M · 80% interval $2.42M–$3.52M · was $2.75M · n=37

Ignores "value" and asks what auctions do: final $/sqft vs list $/sqft (plus avenue and size) across a year of renovated sales, read at 1465's $961/sqft ask. Its key finding: a low list barely drags the close (β = 0.31 — the twin listed $2.00M, closed $2.80M). It needed the biggest fix: its training window averages to a market ~5 months stale. Roll-forward: +9.7% (every sale re-priced to 2026-07-25).

The hot-market correction

All three models learned from auctions in a cooler market. Two kinds of evidence say it's gone. First, the auctions: median over-ask has roughly doubled in five quarters.

Sunset single-family closings by quarter
Quarter	Sales	Median over-ask	Renovated	Reno med $/sqft
2025 Q1	21	19%	5	$1061
2025 Q2	38	15%	14	$1066
2025 Q3	22	27%	8	$1142
2025 Q4	26	27%	3	$1128
2026 Q1	26	31%	13	$1182
2026 Q2	29	39%	10	$1163

Half of that climb is strategy, not price: asks rose only 0.8%/mo over the trailing year while closes rose 1.6%/mo. The other half — deeper bait-pricing — inflates over-ask without moving prices (that's β = 0.31 at work).

Second, the price level. Regressing ln($/sqft) on sale date (controlling for size and avenue) answers differently depending on the sample and window — the spread is real:

Measured market drift — four reads of the same question
Sample	n	Drift /mo	s.e.	Annualized
Renovated, trailing 12m	37	2.1%	±0.7%	29%
All single-family, trailing 12m	110	1.6%	±0.5%	22%
Renovated, trailing 18m	53	1.5%	±0.4%	19%
All single-family, trailing 18m	164	0.9%	±0.3%	12%

The shorter window is hotter in both samples — the drift is accelerating. We average the four reads, keep their spread as uncertainty, shrink toward the citywide 16–18%-YoY prior, and use 1.4%/month ± 0.6% everywhere. One sigma spans both the skeptic's read of that table and the accelerationist's.

On top of the drift, two more changes:

Weights 20%/40%/40%. Average-reno was falsified by both twins before they sold; it stays as the "premium evaporates" hedge at reduced weight.
An IPO-buyer scenario. 15% chance the room contains one must-deploy buyer who pushes the clearing price +7%. Barely moves the median; flattens the top of the curve — 99% certainty doesn't exist at a sane price.

What the correction is worth — bid required at each win probability
Target P(win)	Trailing models	Hot ensemble	Δ
50%	$2.69M	$2.89M	+$204K
70%	$2.90M	$3.11M	+$206K
80%	$3.04M	$3.24M	+$208K
90%	$3.25M	$3.46M	+$206K
95%	$3.46M	$3.66M	+$197K

If the heat stalls, the gray trailing curve is the floor: every threshold relaxes by the Δ column. The bet isn't whether the drift exists — both tables measure it — it's that the next six weeks look like the recent past. The IPO calendar argues hotter, not cooler.

The win curve

Bid on the x-axis, win probability on the y-axis. Hover for a guide line; click a legend chip to select a model and the guide marks the bid where it crosses the hovered win %. Dashed verticals: the twin's $2.80M print and a round $3.00M. The horizontal gap between the gray curve (backward-looking) and the blue one is the price of two more months of this market.

Average-reno runs highest at low bids and flattest (cheap-house belief, most uncertainty). Nicest-house starts low and transitions sharply (confident belief). Bidding-war is now the most demanding — it was the most stale, so it got the biggest correction. The hot ensemble never quite reaches the others' ceiling; that gap is the must-deploy buyer.

The odds, bid by bid

P(your bid wins) — bar length = probability; hot ensemble shaded darker
Bid	Average-reno	Nicest-house	Bidding-war	Hot ensemble
$2.50M	47%	10%	15%	18%
$2.55M	51%	13%	18%	21%
$2.60M	54%	17%	22%	24%
$2.65M	57%	21%	26%	28%
$2.70M	60%	26%	30%	32%
$2.75M	63%	31%	34%	37%
$2.80M = twin’s print	66%	37%	39%	41%
$2.85M	69%	43%	44%	46%
$2.90M	72%	50%	48%	51%
$2.95M	74%	56%	53%	56%
$3.00M = $3.0M mark	76%	63%	58%	61%
$3.05M	79%	69%	62%	65%
$3.10M	81%	74%	66%	69%
$3.15M	82%	79%	70%	73%
$3.20M	84%	83%	74%	77%
$3.25M	86%	87%	77%	80%
$3.30M	87%	90%	80%	83%
$3.35M	88%	92%	83%	86%
$3.40M	90%	94%	85%	88%
$3.45M	91%	96%	87%	90%
$3.50M	92%	97%	89%	91%

Sanity anchors

Where known prices land — "P(win)" reads as the chance that price would have taken the house:

Anchor	Price	Trailing P(win)	Hot P(win)
The list price (a deliberate under-ask)	$2.30M	19%	9%
1501 28th Ave — the other $2.8M twin, sold 2026-05-08	$2.73M	54%	34%
1820 Kirkham — 1465’s truest twin, sold 2026-05-14	$2.80M	61%	41%
1820 Kirkham’s print rolled forward to 2026-07-25 at the measured drift	$2.89M	69%	50%
Finish × location surface model	$2.72M	53%	34%
Cousins-pages verdict center	$2.88M	69%	49%
Cousins-pages verdict top (designer premium)	$2.96M	75%	57%

The trailing column reads the twins' prints as near coin flips — correct for May. The hot column marks them down on purpose: a May price is a stale bid in July. Rolled forward at the drift, Kirkham lands back near the coin flip (50%) — the calibration check passing in the new regime. The cousins verdict center ($2.88M) is now a below-half bid: comp-based "fair value" is what you pay when nobody else shows up.

What this says about strategy

The conversation starts at $2.89M (a coin flip) and gets serious around $3.11M–$3.24M. Each extra $50K buys the most probability in the steep middle (~$2.90M–$3.30M); above ~$3.40M you're mostly paying down the IPO scenario, which never fully pays down. Two things the models can't see, both in your favor:

Terms. All-cash, no contingencies, seller-friendly close: worth ~1–3% of price in a competitive bid. A $3.24M clean offer competes like ~$3.31M financed — read your effective odds one row down the table.
Timing. Every month of waiting raises the curve ~1.4% (~$41K at the median) — before the IPO money even lands. Today's thresholds are the cheapest on offer.

Method & caveats

"Win" is approximated as P(final price ≤ bid): the realized sale price stands in for the highest competing bid had you stayed home. Ignores terms (beyond the rule-of-thumb above) and assumes your bid doesn't change others' behavior.
The drift: ln($/sqft) ~ months-ago + ln(sqft) + avenue, over the 2×2 grid above. The four reads combine as one observation — mean 1.5%, s.e. 0.6% including their ±0.4% spread (they share data, so no √n shrink) — then blend with a 1.3% ± 0.5% citywide prior. The carried ±0.6% keeps the regime spread, so each model's interval widens by (lever × 0.6%)² — the correction adds uncertainty, not just level. If appreciation stops dead, the hot thresholds overshoot by the Δ table; if the 12-month acceleration is the real regime, they undershoot.
The correction stack, separated (all three pieces push the same direction — here's each one's share of the 80% bid): equal-weight trailing $3.04M → re-weighted $3.04M → + drift $3.21M → + IPO layer $3.24M. The drift is most of the move; the weights are nearly free.
Average-reno: the 8-year hedonic refit here; lognormal prediction at 1465's specs (2026 Q2), σ_log = 0.231 after widening. Known bias: "renovated" is one dummy, so a top-1% finish reads as an average remodel.
Nicest-house: 14 judged comps (both twins' doppelgängers), today's-$ $/sqft ~ finish, Student-t at finish 95 — an extrapolation one point past the best comp, so the s.e. widens. Finish scores are LLM-judged from photos; motivated-scoring risk is nonzero.
Bidding-war: ln(sold $/sqft) ~ ln(list $/sqft) + avenue + ln(sqft) on the 37 renovated closings since 2025-06-11, with both list and sold re-priced to 2026-07-25 (re-pricing only the outcome would dilute β). 1465's ask was set in the current market and enters unadjusted. Still assumes the auction format (under-list, multiple offers) behaves as it has.
Ensemble & IPO layer: 20%/40%/40% mixture, then P(win at $X) = 0.85·F(X) + 0.15·F(X/1.07). ε and the uplift are judgment calls — the 80% bid is $3.21M with the layer off, $3.28M at ε = 30%.
Bids below the $2.30M list aren't shown — a seller would re-list rather than take under ask in week one.
Generated by scripts/build-bids.mjs from the same Compass crawl as the other pages; static, re-run after a re-crawl. The pre-correction numbers survive as the gray curve and "Trailing" columns.