An archived source is still a source

Every source registered so far has a live publisher URL. This one does not, and cannot.

The department publishes one State Foundation Funding Calculator and replaces it in place. There is no archive directory, no versioned filename series, no “prior years” page. When the FY2027 model went up, the FY2026 model stopped existing at any address the department serves. That is not an outage to wait out; it is the publisher’s release model.

What is lost with it is not decoration. The two transportation rates — the per-rider and per-mile amounts of R.C. 3317.0212(C) and (D) — are trimmed means of what districts reported spending, so they are the formula’s only endogenous parameter and the only one that moves without an act. parameter/transportation-cost-rates has carried “the rates before FY2027” as an open field since it was written. One prior year is recoverable. It is recoverable only from the Internet Archive.

And this repository has a recorded lesson about exactly that. the-capital-channel-has-no-second-publisher states it: a Wayback 200 on a portal proves the crawler stored a shell, not that it captured anything. That lesson is not theoretical here. Every archived copy of the FY2025 traditional calculator — two capture dates, two filenames, four URLs — is exactly 1,048,576 bytes, truncated at one mebibyte mid-stream, and fails a zip integrity check. The archive holds a file that is the right name, the right content type and the wrong length, and answers 200 for all of it.

So the question is not whether to trust the Internet Archive. It is whether a retrieved file can be shown to be the publisher’s, by something better than the status code that has already lied once.

Approved, narrowly and with a check.

One source, fy26-calculator, under the existing dew-foundation connector, its url a web.archive.org/web/<timestamp>id_/ capture of the department’s own FY2026 workbook. It is not a new connector: same publisher, same office, same page, same release calendar, same epistemic status. Only the address is the archive’s, and that is a fact about where a copy survives rather than about what the document is.

The rule this establishes. An archived URL is admissible when three things hold, and all three are recorded on the source rather than left to judgment:

1. The publisher no longer serves the file at any address, and the reason is the publisher’s release model rather than a transient failure.
2. The retrieved bytes are pinned by SHA-256 like every other source, so the archive cannot quietly serve something else later.
3. The file proves it is the publisher’s by reproducing the publisher’s own arithmetic.

The third is the one that answers the FY2025 trap, and it is the reason this is approved rather than declined. A digest pins what was retrieved; it says nothing about whether what was retrieved is the real document. So the builder recomputes a published figure from the workbook’s own inputs and fails if it disagrees: each district’s [A1] Per Rider Based must equal its weighted ridership times the rate read from the sheet header. A truncated file has no district rows to check and fails; a wrong file fails on the first row; the 1 MiB FY2025 capture fails before that, at the zip.

That is a stronger claim than any live source in this registry currently makes about itself, which is the appropriate asymmetry: an archived source should have to prove more.

What is not approved. Not a general license to prefer the archive. A live publisher URL is used wherever one exists, and verify continues to check the live sources against their digests. Not the FY2025 traditional calculator, which is unrecoverable and stays that way in the record — the catalog entry names its length so a later reader does not spend the afternoon this one cost. And not backfilling FY2022 to FY2024, which the archive does not hold at all.

A note on what this does and does not close. It gives the rate series a second year, not a series. Two observations cannot separate a fuel response from anything else, and the two rates move in opposite directions across them — the per-rider up 4.88% and the per-mile down 2.73% against a 17.5% rise in diesel between the cost years they are drawn from. What the second year buys is the ability to say that, which the corpus could not say at all with one.

A second file clears the same three conditions, and it is a FY2025 file. fy25-cs-calculator is the FY2025 community and STEM school simulator, archived whole on 22 February 2025 at 1,347,923 bytes, and it is registered under dew-foundation on exactly the rule above.

It is worth being explicit that this does not reverse the “what is not approved” line, because the two files share a fiscal year and nothing else. The one that stays unrecoverable is the FY2025 traditional district calculator, whose every capture is truncated at a mebibyte. The one admitted here is a different workbook, from a different page of the department’s site, and the archive happens to hold it intact. The distinction is the crawler’s, not the rule’s — which is the point of writing the conditions down rather than deciding file by file.

The third condition is met in the same shape as fy26-calculator’s and with the same builder discipline: the workbook writes its own formula out as Equity Supplement [a*$650], and the extractor recomputes every row against it. All 317 site-based schools pay $650 times enrolled ADM with zero mismatches; the 18 e-schools and 8 STEM schools pay exactly zero; the two sheets that compute the line independently agree on all 343 schools; and the department’s own statewide row, $54,846,889.86, is the sum of the rows beneath it. A truncated or substituted file fails all four.

What it buys the corpus is one parameter, not a series: the FY2025 equity supplement rate now has a department model behind it rather than the act alone. The middle year does not, and the registry is the wrong place to look for it — FY2026’s $500 was never published as a community-school model the archive could hold. The corpus node formula-component/fsfp-community-school-equity-supplement records that as closed rather than pending.