What HeardTogether is, and isn’t.

Why these two lanes share a site

A council ordinance amending a Planned Development for a 79.3-acre data center and a 503-message chat transcript fabricating Roman numerals on the Voynich Manuscript do not look like the same problem. They are.

In each case, an authority — municipal in one, computational in the other — produced a decision or a body of claims much faster than ordinary readers could verify, in vocabulary specifically designed to look authoritative, and embedded in a structure (an agenda packet, a YAML block with checksums) that makes downstream reuse easy and original-source verification hard. In both cases the simplest defense is the same: assemble the actual documents, label quotation versus interpretation, and ask precise questions.

The civic case files are not an argument against data centers. The AI-safety teardowns are not an argument against AI. They are records on which more careful conversations can be built.

Load-bearing commitments

1. Primary sources outrank summaries. Every factual claim on this site is anchored to the originating document. Where the source is a PDF, the page links to the PDF and quotes verbatim with line numbers. Where the source is a chat transcript, the page quotes the exact assistant message and identifies the conversation.
2. Quoted material is marked as quoted; interpretation is marked as interpretation. The blockquote style on this site is reserved for verbatim source text. Anything outside a blockquote is the maintainer’s framing, never presented as a finding-by-the-source.
3. No analytics, no third-party scripts, no LLM-in-the-loop on the site itself. The civic case files have been built and are maintained by hand. The AI-safety teardowns were drafted by humans using HAIL’s SlopFilter / ECP-1 framework. No model writes content here.
4. Falsifiers are listed openly. Each case file ends with a list of open follow-ups that, if resolved against the page’s claims, would require updates. The page tells you where it could be wrong.
5. The hard safety guardrails apply equally to everyone — including the maintainer, the residents’ groups, the AI labs whose model outputs appear in the library, and the elected officials named in the civic files. No doxxing, no threats, no targeted harassment. See Rules.

What this site is not

• Not a 501(c)(3). A nonprofit-formation template exists for this kind of work, but HeardTogether is not currently incorporated. It cannot accept tax-deductible donations.
• Not the residents’ group. The Lavon civic case file links to Lavon Families Against the Data Center / Protect Lavon as a peer resource. It does not mirror that group’s content, does not coordinate with its organizers, and does not represent it.
• Not affiliated with any AI lab. The Ghost Pattern Library publishes forensic analysis of model outputs from Grok (and, eventually, others). HAIL is the analyst, not a vendor for any of the models examined.
• Not legal advice or psychological advice. Pointers to the Texas Open Meetings Act, Texas Public Information Act, and the published literature on AI safety are pointers to public sources. They are not lawyering or clinical guidance.
• Not a feed. Updates happen on a slow cadence. Pages are dated. There is no scrolling timeline, no notifications, no engagement mechanic.

Operating discipline

The site is maintained on a few simple rules:

• If a fact can’t be linked to a primary source, it doesn’t go on the site.
• If a primary source becomes unavailable (a city takes down a PDF, a model conversation is deleted), the original is mirrored to the on-disk cache so the citation chain survives.
• Corrections that point to a primary source are made promptly. Corrections without a source are not.
• Each case file has a “last refresh” timestamp visible at the top. The civic case file is refreshed daily by an automated script that pulls new council agendas; the AI-safety library is refreshed when new specimens are cataloged.

Origins

HeardTogether’s structural conventions — primary-source library, plan-builder, moderated-submission, no-LLM-in-the-loop — were drafted in 2025 as a template for civic engagement in a Texas town facing organized harassment around school-board and police-accountability issues. That template went undeployed; the load-bearing principles survived.

The site went live in its current shape on 2026-05-23 when two real workstreams — the Lavon data-center situation and the Grok ghost-corpus analysis — arrived at the same time and turned out to be the same problem in two different costumes.

The site is built and maintained by Ed Honeycutt, a software engineer in Princeton, Texas. HAIL (Honeycutt AI Labs LLC) is the analytic shop behind the SlopFilter framework. Neither is a media organization. This site does not seek funding, attention, or advocacy partnerships.

If you spot an error, see Contact for how to flag it.