How we know what we publish.

01What this is — and isn't

Colorado Works Better is an independent, public-records-based effort to help citizens understand how the state raises and spends money. It uses official documents and clearly-labeled analysis. It is not an official State of Colorado system, and it does not speak for any state agency.

Some public records are incomplete, delayed, or inconsistent. Where that is true, we say so on the page rather than papering over it. Our job is to make the public record legible — not to replace it, and not to claim more certainty than the record supports.

02How we label every number

We separate three different kinds of claim so you always know what you're looking at. Every figure on the site carries one of these chips.

The principle behind all four, drawn from our data rules: facts, calculations, and judgments are kept visibly separate. Facts come from public records. Calculations are labeled formulas. Any judgment carries the ANALYSIS label and is explainable on request.

NOT PUBLISHED is the sharpest kind of pending, and this site shows it. The publisher could report the figure and does not. The field stays blank, and no estimate goes into it.

Where a figure belongs and nobody publishes it, the page draws a block of watermarked question marks. An empty bar reads as a page that failed to load. The question marks read as the finding. That mark is only ever for a figure someone could publish and has not.

WITHHELD is the one refusal that is ours rather than the publisher's. The record exists and we hold it; we are deliberately not showing it, and the chip sits where it would have been with the reason beside it. A blank there would read as a figure we could not find. Everything published alongside it is unchanged, and the decision is reversible. It applies today to the vendor names on the project ledger.

The Honesty Framework, the open standard behind this site, defines all three.

A forecast is none of the above. It is not a blank, and it is not an estimate of ours. Colorado forecasts its own TABOR refund and publishes the figures, so Cost of Living shows three of them — FY 2025-26, FY 2026-27 and FY 2027-28 — quoted from Legislative Council Staff with their name on them. A forecast says so everywhere it appears: in the table cell, under the bar, in the chart’s readout, and in the description a screen reader hears. The bar is drawn as an outline where a certified figure is filled. A forecast is never added to a certified one, and when the certified figure arrives and differs, that is a corrections-log entry like any other.

The Honesty Framework sets out that rule too.

03Source confidence grades

Not every official record is equally current or equally precise. Each project field carries a confidence grade so you can weigh it.

Figures shown with a ≈ are program-level or rounded numbers from a budget or overview document rather than a single audited contract; we grade these B and label them accordingly.

04The rule that governs everything: appropriation ≠ spend ≠ performance

This is the single most important idea on the site. A budget document tells you what the legislature authorized. It does not, by itself, tell you what was actually spent, or whether a project is on time, over budget, or delivering value. Conflating these is the most common way spending dashboards mislead people.

Because of this rule, we will never publish a claim of waste, delay, or over-budget status unless the source record supports it — or unless the page clearly frames it as an open question for the state to answer.

05Reading the Long Bill's fine print: (I), (L), and the money nobody votes on

The Long Bill prints letter notations beside many line items. They matter enormously: they mark the difference between money the General Assembly actually controls each year and money it merely displays. Our dashboard's "Fine Print" section is built on them.

• (I) — informational only. The amount appears in the bill "for informational purposes" but is not appropriated by the legislature: TABOR-exempt enterprises, continuously appropriated cash funds, and most federal money. In FY 2026-27 this is ≈ $17.99B of the $49.56B headline — about 36¢ of every displayed dollar.
• (L) — local funds. Money belonging to local governments that flows through state accounts (≈ $180M in FY 2026-27).
• (M) and (H) — federal match mechanics. If federal funds shrink, the paired state appropriation ratchets down automatically. A quiet exposure: parts of the budget resize themselves to Washington's decisions without a new vote.
• General Fund Exempt. Referendum C money — General Fund revenue voters allowed the state to keep above the TABOR cap (≈ $3.9B in FY 2026-27). It is inside the "General Fund" totals but accounted for separately.

Two structural rules from the bill's capital and IT headnotes also shape how we present projects: capital and IT appropriations stay available for up to three years (so "one-time" money lands across several years, and multi-year projects return for phased appropriations), and a project whose lowest construction bid exceeds its appropriation "shall be redesigned to conform to the appropriation" — agencies must shrink the project, not ask for more.

06How we choose which projects to track

The portfolio focuses on major commitments over $10 million — the level at which a single decision meaningfully moves taxpayer money. The portfolio spans three categories — transportation (CDOT), buildings & facilities (capital construction), and technology (state IT) — and deliberately includes requests the state declined to fund and projects terminated before delivery, labeled as such.

• We start from official budget and project documents, not from news coverage or memory.
• We prefer discrete, identifiable projects. Some Long Bill items are programs or fund transfers rather than single projects; those are tracked differently or excluded until the record is clear.
• The batch grows over time. The count shown on the dashboard is the number we have verified so far — not a claim that it is the complete universe of projects over $10M.

07When we don't know, we say so

Colorado does not publish spend-to-date line-by-line for every project in a single feed. Rather than estimate a percentage to fill the column, we leave it marked PENDING until we reconcile it from expenditure data. A blank we can defend is worth more than a number we can't.

The same discipline applies to schedules. Where a public milestone schedule exists, we show it; where it doesn't, the field reads "not yet ingested" instead of a guess.

08Where the numbers come from

Every source below is public. Links open in a new tab. This is the reader's list; the complete running register of every state-wide and departmental system we read — including the ones that are currently blocked, and why — is generated from the pipeline itself and kept in the repository’s source register.

Each project's detail panel on the dashboard also lists the specific record(s) used for that project, with the date we accessed them. Vendors shown in a project's "Who is being paid" table come from award announcements and project records (CDOT, FHWA, agencies, legislative staff), each row linked to its source; per-vendor paid-to-date shows only published figures until reconciled from TOPS. The table's "Political giving" links open TRACER, the Secretary of State's official campaign-finance database — contributions are legal and public, and a search link implies no wrongdoing by any vendor or official.

09County reports

When you subscribe, the weekly email is tailored to your county. County profiles cover your representation and local context: your U.S. House district, your state Senate and House districts, the municipalities in your county and their population, and the major state projects in your region.

County profile sources

• U.S. Census Bureau — 2020 Decennial Census↗
• Colorado State Demography Office↗
• Colorado Independent Redistricting Commission — 2021 final maps↗
• Colorado General Assembly — district lookup↗

District assignments reflect the 2022 redistricting maps. Where a field is unavailable in the authorized data, the report says so rather than estimating.

10Updates & corrections

The dashboard is refreshed as new public records become available, with weekly change summaries once the ingestion pipeline is live. We are not affiliated with the State of Colorado, and we welcome corrections: if a figure looks wrong, tell us which project and which source and we will trace it and fix it.

The corrections log. A change that alters a published figure or a project's stage is never silent. Each one is recorded in the corrections log with the date, the subject, the value before and after, and the reason — including whether the error was ours, an upstream source changing, or two official sources disagreeing. A correction is made once, at the source of truth, so every surface of this site updates together.

How we measure traffic — the full disclosure. This has been true since Aug 1, 2026. This site counts page views using first-party, cookieless analytics: aggregate counts of pages viewed, referrer, country and device class. No cookies, no fingerprinting, no identifiers, no cross-site tracking, and never any content you type. Calculator inputs stay in your browser. Why: watching aggregate patterns for anomalies, and for signals about what to improve. This replaced the earlier “no analytics” wording, and the change is logged in the corrections log — disclosed before the first count was collected.

How we handle what you send — the full disclosure. This has been true since Sep 3, 2026. Ratings and comments are stored with no name, address or identifier, and are published only as totals, never as individual quotations; names, emails, phone numbers and street addresses are stripped from a comment before anyone reads it. A county sign-up or a follow keeps your email address and what you chose to follow, and nothing else about you. A confirmation link is sent before anything else, every email carries a one-click unsubscribe, and no email is sent until you click. We do not record IP addresses, and we do not track whether an email was opened or a link was clicked. Nothing you send is used to train or tune any model. The store is controlled by this project and by no other company. This replaced the earlier “demo only” wording, and the change is logged in the corrections log — disclosed the day collection began.

We also measure how quickly pages load for real visitors: the timing of this page rendering in your browser, and with each timing the browser, device class, device operating system, network speed and country. No cookies, no identifiers, nothing that follows you from one page to the next, and never any content you type. Why: a page a reader cannot load is a page that publishes nothing, and we cannot fix what we cannot see.

What we count when you use the page. We also count a few interactions, not just page views: when a citation is followed out to its publisher; when the Honesty Framework is opened and from which page; when a data table is opened; how the question player on one prototype page is used; which link is followed, and from which menu or part of the page; how far down a page a reader gets, by section; how far the guided tour is followed; whether a ZIP code typed on this site matched one county, crossed a county line, or was not in the file — never the code itself; that updates were asked for, and from where; and which broad region you are reading from. Four rules bind all of it. Every event has a name from a fixed list, and anything not on that list is dropped. Nothing you type is ever recorded — not a ZIP code, not a search, not the words of a comment; where we count one of those, we count only that it happened. Location stays coarse: a county, a region or a state — never a postal code, never coordinates, never an address. A county is named only where your postal code sits wholly inside one and enough people live there for the answer to be about a place rather than a person. Location is worked out where your request first reaches us: a county where your postal code sits wholly inside one and enough people live there, otherwise a group of counties, more often a whole state, and outside the country nothing but that. The closer location your network reports is read there and thrown away in the same breath — it never reaches this page, it is never written down, and nothing about it is kept between one page and the next. If your browser sends the Global Privacy Control (GPC) signal, or the older Do Not Track (DNT) signal, we honor it. Nothing about your visit is counted: no page view, no page-speed timing, no interaction. We do not ask our edge where you are, so no region and no county is ever known. The two counting scripts are not loaded at all. We sell nothing and share nothing with anybody, under any setting. The address bar is stripped of anything after the path before a count is sent, so a value you typed cannot reach us even by that route. Nothing is counted while somebody from this project is previewing a feature that is off for the public: those visits make no request for a location, send no event, and are not included in the page-view count.

The evidence archive. Every source document we cite — bill text, agency page, staff analysis — is archived as an immutable copy on the day we access it. Links can rot and pages can be edited after the fact; the archive means our citations cannot. Paired with nightly data snapshots, it keeps the question “what did this site say on date X, and based on what record?” permanently answerable.