County permit archives in the Mountain West share a common physical origin: paper records produced over decades, eventually scanned as part of a digitization effort, and stored in whatever document management system the county had budget for at the time of scanning. The surface-level story is similar across New Mexico, Arizona, Colorado, and Utah. Below that surface, the variation in archive quality is substantial, and it matters for anything that tries to automatically extract structured data from those archives.
Understanding why this variation exists requires tracing the administrative, demographic, and funding factors that shaped when and how counties digitized. The variation is not random. It follows recognizable patterns, and those patterns predict the extraction challenges you will encounter.
When digitization happened determines scan quality
County permit archive digitization in this region happened across a wide window, from roughly the mid-1990s through the 2010s. Counties that digitized earlier did so with equipment and software that produced different outputs than counties that digitized a decade or two later. Early batch scanning at government archives often targeted 200 DPI or even 150 DPI, which was considered adequate for document preservation at the time. Later projects commonly specified 300 DPI as a minimum.
The practical consequence is that archives scanned at 150 or 200 DPI produce OCR results of noticeably lower quality on anything but large-print text. Small-font form labels, handwritten entries, and dense printed form fields all suffer. The characters are legible to a human eye but present real ambiguity to an OCR engine: a "1" and a "7" may differ by only a few pixels at that resolution, and a handwritten entry that would read clearly at 400 DPI becomes a confidence problem at 150 DPI.
Resolution is not the only variable from this era. Color depth matters too. Documents scanned as 1-bit bitonal (pure black and white, no grays) compress aggressively but eliminate the gray-tone gradients that help OCR engines distinguish faint ink from background noise. Documents scanned as 8-bit grayscale or color retain more information but at much larger file sizes. Counties made these tradeoffs based on their storage budget and bandwidth constraints at the time of scanning, without anticipating extraction use cases that did not exist yet.
Rural versus metro county staffing and form consistency
The Mountain West includes some of the least densely populated counties in the contiguous United States alongside growing urban cores. This demographic spread has a direct effect on how consistently permit forms were completed and how standardized the underlying templates were.
In counties with small planning staffs, forms were often completed by the same few people over many years, creating idiosyncratic but internally consistent patterns. A small rural county might have used the same permit application template from 1982 through 2008, completed by a rotating staff of two or three clerks. That archive, though modest in volume, may have high internal consistency. Extraction schema development is straightforward once you understand the single template in use and its handwriting conventions.
Higher-volume counties updated their templates more frequently, often tied to state-level statutory updates or building code cycle changes. New Mexico, for instance, has updated its adoption of the International Building Code and associated administrative forms on multiple occasions, and each code cycle tends to bring form template revisions. Colorado and Arizona have similar patterns, with some counties on home-rule authority that maintained their own templates independent of state standards. The result is that a high-volume county archive may contain five or six generations of permit form templates across a 30-year backlog, each requiring its own extraction schema.
Document management system fragmentation
There is no dominant document management system across Mountain West counties. The market includes a range of products, some general-purpose DMS platforms used across government functions, some purpose-built land records systems, and some homegrown solutions built in-house by county IT departments at various points. This fragmentation means that when archived documents are accessed for extraction, they come from different systems with different export formats, different metadata schemas, and different file naming conventions.
Some counties store documents as individual single-page TIFF files, one file per scanned page, with a separate database tracking which pages belong to which permit file. Others store multi-page PDFs with all pages of a permit application bundled together. Others export as PDF packages where the bundle structure is encoded in a sidecar XML file that is not preserved when files are downloaded individually. For automated extraction, each of these storage patterns requires a different ingestion strategy before OCR even begins.
Metadata quality at the document level also varies. The most useful documents arrive with at minimum a document type label and a parcel identifier or address associated with the file. Less helpful are archives where the only metadata is a scan date and a sequential file ID. The file name "19940812-00247.pdf" tells you the scan date and the sequence number; it does not tell you the document type or the parcel it belongs to. Those associations exist in a separate lookup table, which may or may not be exported alongside the documents.
APN format variation across counties and over time
Assessor Parcel Numbers in the Mountain West follow county-specific formats with no regional standard. Within a single state, format conventions differ substantially between counties. New Mexico county APNs range from fully numeric hyphenated formats to alphanumeric codes that encode township-range-section information. Arizona counties use a variety of book-map-parcel structures. Colorado's system similarly varies by county assessor jurisdiction.
What makes this particularly challenging for extraction is that format changes over time within a county are common. When a county replatted, annexed, or reorganized its parcel numbering, the same physical parcel may have carried different APN values across different eras of permit records. An older permit from the 1980s might carry an APN that was subsequently reorganized into a different number, with no explicit notation in the permit document itself that the APN changed. Matching that older record to the current parcel database requires either a parcel history table (which some counties maintain and some do not) or address-based matching as a fallback, with its own normalization challenges.
This is not a problem unique to the Mountain West, but the region's history of active land division, rural subdivision activity, and Native land designation boundaries creates some edge cases that are less common in other regions. A parcel in a subdivision recorded in the 1970s near a former agricultural area may carry APN lineage through multiple reorganizations. Automated matching needs to handle these cases gracefully, which means routing ambiguous matches to human review rather than forcing a best-guess link.
What this means for extraction pipeline design
The variation described here leads to a practical design principle: extraction pipelines that work across Mountain West county archives need to be configurable per county, not generic. The document type classification step needs to recognize the specific template generations present in a given county's archive. The APN normalization step needs to know the format rules for that county. The ingestion step needs to handle the storage pattern that county's DMS uses.
Generic extraction that assumes uniform document formats will produce poor results on archives with significant template variation or unusual scan quality profiles. The configuration work is the investment: mapping the template generations, establishing normalization rules for the APN format, and testing on representative samples of each document type. Once that configuration exists, extraction quality is predictable. Without it, results will be inconsistent in ways that are hard to anticipate in advance.
We are not suggesting that extraction is impractical on lower-quality archives. Most of what counties actually need, the core fields of permit number, parcel identifier, document date, and record status, can typically be extracted even from 200 DPI archives with high enough confidence for planning department use. The constraint is not absolute quality thresholds but rather which fields are reliably extractable and which require a higher human review rate. A pilot evaluation on a county's actual documents, rather than an assumed quality level, is the only way to establish this accurately.
State records retention statutes as a floor, not a ceiling
Each Mountain West state has records retention schedules that set minimum retention requirements for permit and land-use records. New Mexico's retention rules, administered through the State Records Center and Archives, specify retention periods for permit records by document type. Arizona's similar framework is administered through the Arizona State Library, Archives and Public Records. Colorado counties operate under the Colorado Local Government Records Act.
These statutes establish that permit records must be retained, but they say nothing about how accessible those records must be. A county complies with retention requirements by storing the documents in any retrievable form, including a filing cabinet, a scanned archive with no metadata, or a modern document management system with full-text search. The statutory floor does not drive extraction quality. That is purely a function of what a county chose to invest in when it digitized, and what it chooses to invest in now.
The implication is that records retention compliance is a starting point for understanding what a county has, not a predictor of how usable it is for extraction. The gap between "retained and accessible as a document" and "queryable as structured data by parcel" remains entirely a function of the extraction infrastructure that a county has or has not built. That gap is what we work to close.