Agreement hashes now follow the Markdown source
Our agency and broker agreements used complete HTML files as their original artifacts. The fingerprint therefore identified the HTML document, including its presentation markup.
On November 5, we made Markdown the source for new versions of both agreements. The site publishes those bytes unchanged and generates the formatted reading pages from them. The source fingerprint now belongs to the document text rather than the reading page's layout.
Hash the bytes you publish
The agreement manifest names the source URL, version, role, and SHA-256 digest. For the current agency source, the tuple is:
{
"agreement_type": "agency",
"version": "agency-2026-11-05-v2",
"document_ref": "/partners/agreements/agency-2026-11-05-v2.md",
"document_sha256": "32da7cb79f3ca46ee3f2f9fe39904fd676a22e475a84d2efe33ed4132736152f"
}
The browser fetches the named source and verifies its digest before submitting an application acceptance. It hashes the raw bytes, including their line endings. Rendering the Markdown first would identify a different artifact.
The response serves text/markdown and carries Cache-Control: no-transform.
A browser-injected script or an edge transformation would otherwise change
the bytes behind a valid source URL. The release checks fetch the public
manifest and sources and compare their digests with the committed and
registered tuples.
Keep the accepted originals
We introduced v2 sources instead of replacing the registered v1 HTML files. The old files still have their original bytes, URLs, and fingerprints. Stored acceptance records continue to point to them.
Before the conversion shipped, we compared the rendered clauses and link targets of both formats. The version line was the only text difference. The database compatibility map recognizes those exact old-to-new tuples. It does not make every old version acceptable for a later source or a different hash.
That distinction matters when a future agreement changes its clauses. A format-equivalent predecessor is a specific exception, not standing permission to apply with any registered version. Tightening new-application validation before a substantive version remains a separate prerequisite.
We registered the two new source identities before publishing the new manifest. After deployment, we checked the served Markdown bytes, generated readers, archived HTML originals, and existing acceptance records. The historical records and the program's enrollment and servicing settings stayed unchanged.
Let the reader change without changing the record
The reading pages supply section navigation, an AI summary, a source link, and document history. Current summaries jump to the agreement text. Older readers link to the latest reading page while retaining their own original source link.
A mandate card can also link to the application agreement on record. It must handle both Markdown and retained HTML URLs. We caught an HTML-only filter during review and added browser checks for both formats before publication.
The summary is a reading aid. Its saved provenance binds it to the source fingerprint, and the agreement text remains authoritative. Presentation, summary, and acceptance identity each have a clear role.
The reader announcement links to both agreements and their current Markdown sources.

Comments
Comments are reviewed before publication. The Author badge marks replies from the author.
Your test session has expired. Stop testing to use comments and sign-in normally, or open a fresh test link to keep testing.
Stopping testing reloads this page in normal mode.
Read approved comments on the discussion page.
Open the discussion page