Open data downloads

Why the data is published at all

A reference table you cannot take away with you is a reference you have to keep coming back to on somebody else's terms. Everything assembled here, the equipment specifications, the price observations, and the off-grid law verdicts, is published as machine-readable files under a permissive license for the same reason the calculators publish their formulas: work that cannot be checked has to be taken on trust.

It also makes the material more useful. Somebody sizing a system wants the spec table in a spreadsheet next to their own load list. Somebody researching a region wants the law verdicts filtered to a set of counties rather than clicked through one at a time. Somebody studying the market wants the price history rather than this month's snapshot.

What each dataset actually contains

The specification files hold what manufacturers publish for the equipment off-grid builders actually buy, normalized onto common columns so that units of different sizes can be compared: batteries, panels, inverters, charge controllers, generators, and pumps. Each row carries a confidence flag recording whether the figures came from a manufacturer datasheet, a primary retailer page, or a class-typical estimate, and the date it was recorded.

The price observations file is append-only. Every price ever recorded stays in it, with the seller, the amount, whether it was a listing or a reference price, the date observed, and the source. Nothing is overwritten, which means the file is a history rather than a snapshot and can be used to see what the market has actually done.

The law verdicts file holds the eight standardized off-grid verdicts for every state, more than three thousand counties, and the New England towns that hold land-use power, with status, confidence, citation URL, and review date on each. It is a large file and it is the most useful single artifact on this page for anybody researching more than one region.

How to use it responsibly

Read the confidence flags. They exist because the underlying evidence genuinely differs in quality, and a table where every cell looks equally authoritative invites conclusions the evidence does not support. A verdict marked unclear means research could not confirm an answer, which is a finding rather than a gap being hidden, and it should not be treated as permission.

Read the dates. Specifications are revised, model names persist across silent changes, prices move constantly and regionally, and ordinances are amended without notifying anyone outside the county. A dated observation is a fact about a moment rather than a standing truth.

And treat the law data as research rather than as legal advice. It summarizes what public sources say and cites them; it does not replace the office that will issue a permit. Anybody building an application on top of it should carry that qualification through to their own users rather than quietly dropping it.

Stability, and what will not change

The URLs on this page are stable. Columns are added when new material is recorded and are not renamed or removed, so a script written against these files should keep working. Where the meaning of a field changes, it becomes a new field rather than a redefinition of an existing one.

Corrections to published material are logged publicly with dates rather than applied silently, which means a result derived from these files last year can be traced to the state of the data at the time. That matters for anybody citing them.

Citation, licensing, and what is asked in return

The license is permissive: use it, republish it, build on it, including commercially. What is asked is attribution and a link, which is ordinary academic and open-data courtesy and which also lets your own readers find the caveats attached to the numbers.

If you publish something derived from this data, a note of the date you took it is worth including, since all three datasets grow and change. And if you find an error, reporting it is the most valuable thing you can do: corrections are logged publicly rather than edited away, and readers who have checked a figure against its source are the most reliable correction mechanism this site has.

What is not published, and why

Nothing about readers. No account data, no usage data, no email addresses, and no aggregate behavior, because none of that is anybody's business and because publishing it would be a betrayal of the people who use the site.

Nothing that is not ours to publish. Manufacturer datasheets are linked rather than mirrored, and copyrighted source documents are cited rather than reproduced. The data here is the normalized record of what those sources say, which is the part that took work to assemble.

And nothing that pretends to a confidence it does not have. Where a figure could not be established, the field is empty rather than filled with a plausible estimate, because an estimate that is indistinguishable from a measurement a week later is worse than a gap.

Getting started with the files

The comma-separated files open directly in any spreadsheet application and in any statistical or scripting environment without preparation. The JSON documents suit anybody building something programmatically, and they carry the same fields plus the metadata describing when the set was assembled.

For most practical uses the workflow is simple: download the specification file for the category you are shopping in, sort by the ratio that matters for your design, and filter to rows with a verified confidence flag if you want only manufacturer-sourced figures. For regional research, filter the law verdicts file to your state and read across the eight topics, which surfaces patterns that clicking through pages one at a time does not.

The larger files are large because they are complete rather than sampled. The law verdicts file in particular covers every county in the country, and it is the one worth downloading rather than querying repeatedly.

Where the numbers come from, in one place

Specifications are transcribed from manufacturer datasheets or primary retailer specification pages, recorded with a date and a confidence flag, and linked back to the source on each product page. Where a manufacturer publishes a figure without stating its test conditions, it is recorded as published rather than quietly normalized, because inventing the missing conditions would make the table look tidier and be worth less.

Prices are observations rather than estimates: an amount, a seller, a date, and a source URL, labeled as either a listing price or a reference price. The monthly benchmark ranges are derived from those observations rather than asserted, so the underlying evidence for any range is retrievable.

Law verdicts are researched from primary government sources: statutes, administrative codes, county ordinances, and agency guidance. Each carries a citation, a confidence level, and a review date, and counties that have not been researched individually inherit their state framework and are marked as inheriting rather than presented as researched.

If you are building something on top of this

You are welcome to. A few things will make it work better. Cache rather than fetching repeatedly, since the files change on a slow schedule and the law dataset is large. Key on the stable identifiers rather than on display names, which can be edited. Carry the confidence flags and the dates through to whatever you build, because dropping them turns a qualified dataset into a set of confident assertions.

For law data in particular, carry the qualification through to your own users. These are research summaries with sources, not legal advice, and every page on this site that presents them says so. An application that presents them as settled answers is misrepresenting them, however accurately it copies the values.

If you build something useful, saying so is welcome. Several of the improvements to these datasets have come from people who tried to use them for something and found a field missing.

Reporting a problem with the data

Errors in a dataset this size are inevitable, and the useful thing is that they get found and fixed rather than that they never happen. A specification transcribed from a superseded datasheet, a price recorded from a listing that turned out to be a placeholder, or a county verdict that was accurate when researched and has since been amended are all ordinary and all worth reporting.

What makes a report actionable is the identifier and the source: which row, and what the correct value is according to which document. A link to the manufacturer sheet, the ordinance, or the agency page is enough, and it saves the research being done twice.

Corrections are logged publicly with the date rather than applied quietly, so anybody who cited an earlier version can see what changed and when. That record is on the corrections page, and it is also the reason these files are safe to build on: a silent change to a dataset is indistinguishable from an error in whatever was built with it.