From public records to acquisition priorities.

Public property information arrives through separate systems with different responsibilities. Parcel services describe geometry, land use and building presence. Appraisal records provide assessed values, ownership, mailing information and sales history. Tax records add unpaid bills and collection events. Code-compliance records contribute case histories. The pipeline joins these sources through a normalised parcel identifier, giving the same property a consistent identity across successive enrichment stages.

A SQLite cache stores source payloads alongside retrieval timestamps and hashes. This keeps collection separate from scoring: the scoring stage can run from existing records without contacting the upstream services again. The sequence is explicit—refresh parcel data, refresh appraisal data, apply eligibility filters, enrich taxes and compliance, then score. Derived CSV outputs carry the eligible universe and its enriched, ranked forms into subsequent analysis.

Eligibility and priority are distinct decisions. Hard filters define the acquisition universe, including property characteristics, the client's value range and data requirements. Configurable rules then calculate component scores and combine them through weighted composites. This makes the screening policy explicit: a property can qualify for consideration while ranking differently on asset conditions, ownership signals and physical site characteristics.

Each score travels with machine-readable reason codes and readable explanations. The composite provides a sort order; the dimensions explain its composition. These are screening indicators. A high ownership score does not establish an owner's intention to sell, and a site-potential score does not establish development permission.

Method diagram

Close

Feature engineering has consequences at the level of a single property. For lot and building metrics, the implementation selects appraisal information when that source is available and uses parcel-service data when it is absent. It records which source supplied the inputs. The site calculation uses actual building area divided by lot area—the floor-area ratio—and exports the exact areas used, allowing the dashboard to explain a classification in physical terms.

Tax features also depend on time. The scorer evaluates unpaid bills against an explicit as-of date and distinguishes the current collectible year from older debt. Collection-event handling separates active and terminal states and carries supporting dates, statuses and source labels into the output. Missing required appraisal, tax or value information can trigger a configurable exclusion, with exclusion reasons counted in the scoring summary. These details make omissions and changing records part of the research process.

A recurring acquisition workflow needs durable execution state as well as data. A persistent SQLite job queue records scheduled work; a uniqueness constraint prevents duplicate queued or running jobs for the same job and date. Workers claim jobs transactionally, while the pipeline records phase progress, cursors, counts and property-level errors. Compatible interrupted runs can resume from those checkpoints.

The cache uses write-ahead logging to support concurrent reading during updates. Output files are written to temporary paths and replaced atomically, limiting the chance that a reader encounters a partly written export. Collection, scoring and publication therefore have clear boundaries. An analyst can inspect an existing scored snapshot while the refresh machinery works through the next run.

The Streamlit dashboard turns a scored file into a review surface. A sortable property list drives a detail pane showing the composite score, component scores, recorded market value and score reasons. Deeper sections expose property attributes, legal descriptions, ownership, tax information and metadata. The selected snapshot remains visible through the source selector, preserving the relationship between a review and its underlying export.

Coordinates are transformed into the latitude and longitude used by Google Maps and Street View. Dedicated actions open those views and the official property-appraisal portal for the selected parcel. The analyst can move from a numerical priority to the street setting, then to the public record, without reconstructing the property search. This is the practical deliverable: an acquisition research system that carries the team's criteria through collection and scoring into an inspectable shortlist.