Methodology overview
How the Tilt Score is built, what it measures, what it doesn’t, and how the methodology evolves.
What the Tilt Score measures
The Tilt Score is a 0-to-100 read on the seller’s flexibility for a given property at a given asking price. Higher means more buyer leverage — more room for the data to support an offer below ask. Lower means the asking price is well-supported and your negotiation room lives in terms (close date, inspection contingencies, repair credits, inclusions) rather than price.
It is not a valuation. It is not a prediction of closing price. It does not tell anyone what to pay or whether to buy. It tells a buyer what their starting position should look like given what the public records say about the seller and the market.
What we look at
The score is built from public-records and licensed-data signals across several categories: how the property has transacted historically, how the asking price relates to independent valuations and their trends, signals about owner motivation (tenure patterns, ownership type, mailing-address geography, equity position, distress indicators), neighborhood-level market context, and hazard exposure that affects long-run cost.
These signals are combined through a proprietary methodology — the asset PriceTilt is built around. We don’t disclose the numeric weights, the formula structure, or the score-to-offer-range mapping. We do show the inputs, the directional contribution of each category, and the specific public-records facts driving each direction. Same posture as Carfax, FICO, and other proprietary scores.
What the Tilt Score does NOT measure
School ratings and general crime statistics are not score inputs. Both are available on every consumer real-estate site already. The PriceTilt chatbot will surface these via licensed data sources when you ask, but they don’t drive your Tilt Score — paying ATTOM for data you can see on Redfin would inflate cost without changing the answer.
Race, income demographics, and protected-class neighborhood composition are excluded by design. Federal Fair Housing law prohibits using these signals in real-estate decisions or services that facilitate them. PriceTilt has explicit guardrails at the methodology layer, the prompt layer, and the AWS Bedrock guardrail layer preventing these signals from entering scores or chat responses. We measure property-level and seller-level signals — never who lives where.
Future closing price is not predicted. The score advises on a starting position the data supports. What a property actually closes at depends on competing buyers, negotiation dynamics, and timing — none of which we model.
Methodology versioning
Every score is timestamped with the methodology version that produced it. Past scores are immutable — when the methodology changes, we don’t silently restate history. Revisions get a version bump with a documented rationale.
This matters because PriceTilt is being built around a defensible, version-controlled IP asset. The audit trail is part of what makes the methodology credible to an acquirer’s diligence team and useful to buyers comparing scores over time.
What’s coming
Calibration against actual sale outcomes. ATTOM updates deed records 30 to 90 days after a closing, which means PriceTilt can passively observe predicted-versus-actual for every property it scored that eventually transacted. That signal accrues as a side effect of normal usage — no manual data entry required.
Additional seller-rating signals as data sources mature: multi-property owner patterns, rental-balance indicators, short-term-rental density. Whether each becomes a methodology input or stays as chatbot context is decided at the methodology layer, not the marketing layer.