SELF_INDEX_design_and_analysis.md ·
download raw .mdSelf-Indexing EagleWeb → Searchable Index: Design, Feasibility & Legal Notes
Question posed by owner: If we pull ALL recordings from EagleWeb (not just my parcels) into an Azure AI Search index, can we then find the malicious / wild deed? And since my title was cleared 3 years ago, the bad filing must be after that — right? And if a recording was denied, does that stop me from selling?
This note is technical + general information only. It is not legal advice. Confirm the legal points with your attorney and title company (Chicago Title).
1. Does an Azure AI Search index help? YES — it removes EagleWeb's core limitation.
EagleWeb's public search only lets you query the county's index fields: grantor/grantee name, parcel, PLSS section, book/page, doc number, date/type. It cannot search the content of the scanned document images, and it cannot search by geography (where the described land actually is).
A wild deed hides precisely in those blind spots: - filed under a fabricated grantor name you'd never think to search, and/or - carrying a legal description that the recorder never tied to your parcel numbers (so a name/parcel search misses it), and/or - rejected / pending — never published to EagleWeb at all (see §5).
An Azure AI Search index built from the document images + OCR text converts the problem from "search the county's name index" into "search the actual words and the geography of every instrument." That is effectively a do-it-yourself title plant — the same concept Chicago Title uses. This is a genuinely good fit for Azure AI Search because it has a built-in OCR / AI-enrichment skillset (Azure AI Document Intelligence / Vision Read) that turns the scanned PDFs into searchable text automatically.
2. Proposed architecture (bounded, tractable)
EagleWeb (Tyler EagleRecorder)
│ (1) Acquisition: scrape the recording INDEX (metadata) for a bounded date window,
│ then download each document IMAGE (PDF) via viewAttachment.jsp
▼
Blob storage (raw PDFs + a metadata JSON per doc)
│ (2) Enrichment: OCR each PDF → full text (Azure AI Search skillset, or Doc Intelligence)
│ + entity/keyphrase extraction; parse legal-description calls
▼
Azure AI Search index (fields below)
│ (3) Research: full-text + semantic queries; grantor≠owner flags; geo/legal matching
▼
Hit list → manual review → pull certified copies of anything probative
Suggested index schema (one document = one recorded instrument)
| Field | Type | Source |
|---|---|---|
| docNumber | Edm.String (key) | index |
| docType | String, facetable | index |
| recDate | DateTimeOffset, sortable/filterable | index |
| grantor / grantee | Collection(String), searchable | index |
| bookPage | String | index |
| parcelIds | Collection(String), filterable | index (may be empty ← the whole point) |
| plssSection/Township/Range | String, filterable | index |
| legalDescriptionText | String, searchable | OCR |
| fullText | String, searchable (analyzer=standard) | OCR |
| addressesMentioned | Collection(String) | enrichment |
| namesMentioned | Collection(String) | enrichment |
| geoParsedFootprint | Edm.GeographyPolygon (optional) | metes-and-bounds parse |
| blobUrl | String | acquisition |
Detection queries (once indexed)
- Content match on the subjects — search
fullText/legalDescriptionTextfor:4927 Katya,4924 Katya,R32905-401-5010,R32905-400-4740,SP 79-192,383-4650, distinctive metes calls (N5*W423,S89*E843), and owner names incl. misspellings (Rutter/Ruter/Rutten, Garibyan/Garabyan/Garibian). A hit that is NOT already indexed to your parcels = prime suspect. - Grantor ≠ record owner — join every conveyance's grantor against the verified true-owner
table (
03_assessor/owners_and_deeds.csv). A stranger-to-title grantor on/near your land is the classic wild-deed signature. (A parcel-scoped version already run:WILD_DEED_heuristics.txt.) - Geography — if legals are geo-parsed, spatially intersect every instrument's footprint with your two parcels regardless of how it was indexed. This catches the "mis-tied legal" case that the public index structurally cannot.
3. Scope it by DATE — the owner's key insight (mostly correct)
Because a competent title search + owner's policy was issued at your 2023 purchase (Rutter WD #4565727, from Bloom), the recorded chain was clean as of that date. So a newly recorded malicious deed most likely lands in the window 2023-06 (your closing) → present, with the neighbor's map (dated ~2026) suggesting it is recent. That bounds a county-wide pull to ~3 years instead of "all of time," which makes this very tractable (low thousands of instruments, not hundreds of thousands).
Caveats to the "must be after closing" assumption: - A wild deed could predate closing yet have been mis-indexed so the 2023 title search missed it. But a properly recorded deed affecting your parcel should have surfaced in a plant search — and if a covered defect predates your policy, your owner's title policy is the remedy (the title company must defend/clear it). So the new-filing search window is post-closing, while pre-closing defects are a title-insurance claim, not a search project. - The Garibyan parcel took title differently (Decree of Dissolution #4383543), so its clean- date baseline differs. Confirm each parcel's policy/effective date with the title company. - Action: ask Chicago Title for the effective date of your owner's policy and a current date-down / title commitment — that tells you what is of record right now vs. at closing.
4. Cost / effort / risk of the pull
- Volume: date-bounded county-wide (2023→present, all doc types) is plausibly a few thousand docs → OCR cost is modest (Azure Doc Intelligence Read ~ per-page pricing). "All recordings ever" is not recommended — large, costly, and unnecessary given §3.
- Terms of use / rate limits: EagleWeb is a public portal but bulk scraping may hit rate limits or ToS restrictions. Throttle, identify politely, and prefer the narrowest date window that brackets the suspected filing. The Auditor can also provide a bulk data / land-index extract directly — often the cleaner, sanctioned route than scraping.
- OCR quality: old typewritten/handwritten metes-and-bounds OCR imperfectly; geo-parsing is best-effort. Treat hits as leads, then pull certified copies.
5. "If a recording was DENIED, can I still sell?" (general info — confirm with attorney)
- A rejected / denied filing is not entered into the official record. Under WA recording law, an instrument that is not recorded gives no constructive notice and generally has no effect on your title. By itself, a rejected filing does not cloud title and should not prevent a sale. It also will not appear in EagleWeb — which is exactly why "download everything" cannot find a rejected filing (§1).
- But watch the practical risks:
- the filer can re-submit a corrected version, or record a different instrument that is accepted (that would cloud title until removed);
- they could file a lawsuit / lis pendens, which clouds title even if meritless;
- during escrow, a title company may still flag a known fraud attempt and require extra proof, slowing (not blocking) a sale;
- WA law provides remedies for fraudulent filings (e.g., quiet-title and slander-of-title actions, and criminal statutes for filing false instruments) — your attorney can pursue these.
- Net: a purely rejected filing = no legal cloud, sale not blocked, but document it and monitor the recorder for re-attempts. Ask the Auditor whether they retain a rejected/pending queue and whether your name/legal is in it.
6. Bottom line
- Azure AI Search is a sound backend; the hard part is acquisition + OCR, which is identical regardless of index engine. For a corpus this small, a local full-text index (e.g., SQLite FTS5) would also work, but Azure AI Search's built-in OCR skillset + semantic search is the least-effort path if an Azure subscription is available.
- Highest-value scope: county-wide, all doc types, date-bounded 2023→present, indexed by content + geography, with a grantor≠true-owner flag. This is ~90% of the benefit of "index everything" at a fraction of the cost/risk.
- It still cannot find a rejected/pending filing — for that, the only sources are Chicago Title's plant (they made the 2026 map; they likely have the instrument #) and the Auditor's counter (request a land/legal-description search and a check of the pending/rejected queue).