🏛️ Auction Property Hunter

Run the AI evaluation and geocode enrichment for a whole import, and read the resulting summary. Analysis runs through the AI job queue.

Query a county/state ArcGIS parcel service by APN to fill geometry, address, owner, etc. — no paid key. Give the layer query URL (…/FeatureServer/0 or …/MapServer/0) and the APN field name.

Gated auction platforms need a real login (and, for crawl4me headless crawling, the operator's browser-extension consent). Track which accounts we need vs. have here. No secrets are stored — just status + a note on who holds the login.

Keys are stored server-side and never exposed publicly. Reload the page after saving for map tabs to update. Leave a field blank to clear its key.

Server-side keys for owner / deed / sale-history & (soon) lien lookups by APN. Never sent to the browser. Paid enrichment runs only through the job queue when an admin triggers it — it is not part of the free auto-enrich. Precedence when multiple are set: Regrid → ReportAll → ATTOM.

Keys we issue for our own parcel API at /api/v2 — point→parcel, address, APN, owner portfolio, area, field query and typeahead, served from our crawled data instead of a paid provider. Tokens are workspace-scoped and stored hashed: the value is shown once, on creation. Browse the API · field schema

Live view of the crawl4me GovEase job — the county queue, capture progress, and any bad (page-shell) captures that need re-running.

Every crawl4me capture is saved before processing, traced through the AI column-mapper to the parcels it produced, then gzipped so the bulk leaves the live disk (kept for audit). Follow any single capture end to end, and watch which counties are healthy.

The grade is deterministic math — value ÷ cost measured against the county norm, adjusted for signals and data confidence. No AI. Letters come from percentile cutoffs on the 0–100 score, so a grade means "this good relative to everything we hold", with absolute guards that stop a percentile from flattering a bad deal.

Only genuinely actionable problems: crawls producing nothing, counties that stopped arriving past their own cadence, sale dates that passed with no ingest, or thin data where we hold real volume. Counties we simply haven't started are tracked as coverage backlog, not alerts.

One row per registered source. Alerts = a feed that was working and broke (labelled with the cause + fix, not a bare ✗). Backlog = researched-but-unwired coverage. Population/join % show a source can be reachable yet useless. Drill in for the raw body + lineage.

One cell per source per day — thousands of crawls at a glance. Green = all captures produced rows, amber = some skipped, red = failures. Click a cell to inspect exactly those captures.

How well we are gathering each county: freshness against that county's own sale cadence, crawl reliability, and how complete the data is once it lands.

Every county's tax-sale source: third-party platforms (GovEase, etc.) vs. counties that publish directly, and how far along we are ingesting each. Filter to find gaps and direct-publishing counties to build next.

Cross-county auction aggregators — the crawl4me targets. Where sale listings live nationally/regionally, whether they're open or WAF/CAPTCHA-gated (need a headless crawl), and our crawl status. Add sources here as we find them.

What background jobs & automations (imports, the Kansas tax-sale monitor, enrichment, errors) have been doing. Filter by message type.

The homepage map reads a cached JSON snapshot (public/demo-map.json) — a dot per active scored lot plus one featured lot. Rebuild it from live data when you want the marketing map to catch up.

The sample cards under the hero. Leave blank to auto-pick the best scored lots, or pin specific parcel IDs (comma-separated) to curate exactly what shows.