How a Colorado water report is assembled

1 What this document is

A Water Table report answers one question: what do the public records say about the water on this piece of ground. That claim is easy to make and hard to verify, so this document sets out what is actually behind it. It names every feed, every check, and every rule the report follows about the difference between something a public body wrote down and something we worked out.

It is written for three readers. A buyer deciding whether the report is worth its price. A lawyer or lender who needs to know what a statement in it rests on before relying on it. And anyone who wants to reproduce a number in it from the sources themselves.

Every number in this document is generated from the running product on the date above. The argument is written by a person. The facts are read out of the code and the warehouse, so this page cannot describe a version of the product that no longer exists.

2 The record

Nothing behind a report is proprietary. Every source is a public agency feed and anyone may query any of them. The work is that no two of them share an identifier, they publish on different schedules, several publish no statewide layer at all, and none of them answers the question a buyer actually has.

public feeds read

48

separate publishers

16

parcels from county assessors alone

147,382

counties the statewide layer misses

6

Counting organizations, not labels

The provenance table names a body and then the specific dataset, so one agency appears on several rows. Those collapse to one organization before being counted. Reporting the label count would be the easier thing to do and would overstate the number by roughly half, and a reader who noticed the duplication would be right to discount everything else on the page.

One key

Every record resolves to a Public Land Survey System section, a nominal square mile, because it is the only identifier the agencies already share. The diagram above names four real ones on a single section, each filed against a different kind of key. None of those keys appears in any of the other records. All four resolve to the same square mile, and each does it a different way.

That sets the resolution of every statement in the report, and the report says so. A count of wells is a count across a square mile that may hold one ranch or forty subdivided lots, which is why the document carries how many parcels the section is cut into.

The sources that are in no catalogue

Colorado has no single authoritative parcel layer, and the statewide composite is thinnest in the counties where rural land actually trades. The missing counties come from the assessors themselves, which means finding them. One provenance row reads Fremont County Assessor GIS, and Chaffee County published by the City of Salida, because that is where Chaffee County parcels are actually served from. No API directory contains that fact.

3 What had to be read rather than downloaded

Some of the record is not published as data at all. It exists as scanned paper that an agency images and indexes by nothing, or behind a service that will not answer a machine on a schedule. Where that record carries something a report needs, it was taken anyway.

The driller's own log

The state images the completion report a driller files for a well. The tabular record does not always carry what is written on that page. Those images were read and set against the database's own values.

Logs read

463

Supplied a water level the database lacks

117

Supplied a yield the database lacks

54

Disagree with the database on depth

15

The last row is the one worth dwelling on. These are wells where the state's tabular record and the driller's own signed log do not say the same thing. Every one is a real permit that can be looked up.

The scan above is the document behind the first row of that table. The state’s tabular record puts that well at 100 feet. The registration filed in 1958 has 400 written on it in pen, along with a yield and a static water level the database does not carry at all. It is cropped to the well data block: the full page is a public record and the report links to it, but it also carries a family’s names, their address and their signatures, and none of that is needed to make the point.

Where the two disagree the report surfaces the disagreement. It does not average them and it does not silently prefer one. A scanned value that contradicts the database is a fact about the record, and hiding it would be the easier product decision and the wrong one.

The filings the published index does not carry

Water court resumes are the monthly notices of every application filed in each water division. The judicial branch indexes recent years. Everything older exists as a scanned notice inside the state engineer's document repository, readable one page at a time and indexed by nothing.

water court filings in total

20,922

recovered from the scans

8,209

carried by the published index

12,713

years the scans cover

24

Those years were read out of the repository and de-duplicated against the published index, so the figure separates what was recovered from what was already available. Every recovered filing carries a link to the document it was read from, which is the only honest way to publish a number that came out of a scan.

5 What a finding is allowed to say

Every statement in a report carries four things: how much weight it deserves, the evidence it rests on, the statute behind it where one exists, and its basis. The basis is the part that matters most, and it is printed on the finding itself rather than buried here.

record

A public body wrote this down and the report is repeating it.

inference

The report worked this out from the records, and says how.

absence

The record is silent, and the silence is itself the finding.

The third is the one most products will not print. On rural ground an empty rights table is not a gap in the report. It is the answer: nobody has decreed water here. Reporting that as nothing, or omitting the section, turns the most important fact about a parcel into whitespace.

Weight is separated from certainty on purpose. A finding flagged for a closer look and one that is merely worth knowing can both rest on the same quality of evidence, and a reader deciding what to do next needs to see both facts about it.

None of these levels is a valuation. A flag says the record holds something a buyer should resolve before closing, not that the ground is worth less. What it is worth is the reader's judgment and their lawyer's, and the report deliberately stops at showing them what the state wrote down.

6 What the report refuses to do

These are decisions not to produce numbers that would look authoritative and would be unfounded. Each of them costs the product a section that competitors do print.

• No interpolated water table. A surface drawn through sparse measured wells reads as a measurement of a specific property and is not one. Depths are reported as the depths of the wells they came from, with the count of wells behind them.
• No inferred legal access. The federal road layer does not distinguish a public county road from a private subdivision road, so a finding about whether a parcel can be reached would be a guess dressed as a record.
• No filing presented as an outcome. An application can be withdrawn, denied, or granted for less than was asked. The report says what was filed and by whom, never what it will become.
• No regional average presented as local truth. Where a measurement covers a basin rather than the section, the report says so in the sentence, and names whose monitoring network it came from.
• No silent gap. Where a record exists and holds nothing for this ground, that is printed as a result rather than left out.

7 How it stays true

Two reports on the same square mile must agree with each other, and a report must not change because the software was rebuilt. That is an engineering property rather than an editorial one, and it is enforced rather than intended.

• The assembled document is frozen and compared byte for byte. Any change to the code has to prove what it did to the finished report, on real sections, before it ships.
• Aggregates are rounded where they are computed. A sum over floating point numbers adds in whatever order the engine chooses, so an unrounded total is a number that moves between builds. Two of those were being used to rank things, which meant the ranking moved too.
• Any query that picks one row out of several must say how it ordered them. An unordered pick returns a different answer on a different machine, and it does it silently.
• Every source carries the date it was last read, and the report prints it. A number with no vintage is a claim about the present tense that nobody checked.
• A check that scans nothing fails. Automated guards assert they actually examined something, because a guard that matches nothing passes for the wrong reason.