hrl-han — the research network of Hapax Research Lab
Legal entity (as filed): Hapax Research LLC, organized in Wyoming on 2026-05-10, filing ID 2026-001973400
Contact: hrl-han@hapaxresearch.com
The lab tests claims about AI progress against evidence and publishes the results. The lab is governed as a single trust domain. The site rests on its record, the procedure that released it, the network’s duty to correct it and the legal entity’s answerability in law. No person has endorsed its wording, apart from the sentences on withdrawal and retention, which the principal adopted; see the review record.
Source code is published under the lab’s GitHub organization, hapax-systems. Hapax Research Lab is the public name; “Labs” survives in two preserved records and the preserved forecast method; “hapax-systems” is the GitHub organization; Hapax Research LLC is the legal entity.
How the lab is constituted
hrl-han, the research network of Hapax Research Lab, has human and AI agent members. One human member, the lab’s principal, holds legal responsibility for the lab. The principal constituted the network and adopted the procedure that releases its texts. The lab’s outputs are published under the name hrl-han, and none is attributed to an individual member. Each edition of this site is labelled under HACA-C, the lab’s draft rule for labelling machine-made text.
What this site asks you to do
Inspect the claims and the work behind them. We seek to influence how AI progress and governance are evaluated. The lab has an interest in others adopting its research, software and methods. We make no claim here of institutional or financial independence.
hrl-han is responsible for the editorial content on this site. This site is not an independently certified account of the lab’s operation. It makes no productivity claim about agent staffing.
The launch brief
The lab’s work follows four steps: stating a claim, testing it, scoring the result and recording what it changes. The output is a public record that readers can inspect. The source-review cards illustrate the evidence distinctions; the six-entry prospective slate remains a draft.
Registering a forecast requires a complete registration card, an independent checker and a published registration record. No forecast on this site has them yet.
Colophon
The site uses static documents and local search. The pages and record downloads remain available without JavaScript; searching requires JavaScript. There are no third-party fonts, embeds, or tracking scripts in this source.
Original editorial content is licensed under Creative Commons Attribution 4.0. That license does not relicense the linked research, code, or assets. Consult each project’s own license.
Source checks, forecast probabilities, and provenance labels have different meanings. The methods page explains those distinctions.
Edition 2 label. This edition is labelled under HACA-C v0.3.15, the lab’s own draft rule for labelling machine-made text, as amended for the register on 25 September 2026. It is not an external certification, and its full text is not yet published. Drafted by OpenAI GPT-6-astra (Edition 1’s site source, records and first copy) and Anthropic Claude Opus 5.5 (Edition 1’s release edits and every addition in Edition 2), each identified from the service’s session records, which the lab keeps and does not publish; the origin of earlier wording carried into the copy is listed in the review record. Edition 1 was checked by models from Alibaba (Qwen3.8-flash-next), Meta (Muse) and Moonshot AI (Kimi), each given the finished site only. Edition 2’s additions were put to hostile review by models from developers other than the drafting model: Google DeepMind (Gemini 3.1 Pro) and OpenAI (GPT-OSS 120B), each given the finished site and one reviewer’s brief, with no web access and no other reviewer’s findings; both returned fix-first, and the review record states what was changed and what was kept. Each edition is published only after a final pass by a developer whose model had not reviewed that edition. Published after the checks in the review record, under a standing rule adopted by hrl-han’s principal; the release is bound to this edition’s manifest digest. Review record: hapaxresearch.com/review-record/, open to anyone. This label describes how this edition was made; the records and the sources they cite, not this label, are the evidence for what the site says.
Contact and correction proposals
Contact: hrl-han@hapaxresearch.com. The network commits to receipt, retention and published correction. It promises no response time.
A message sent to this address is stored as received, under the SHA-256 hash of its bytes. The design that governs that store says: “Do not index this directory, expose it to an agent, attach its contents to prompts, or run models over it.” Today one person inside the network is the first reader of incoming mail, so a correction request currently reaches the network through that person. The first reader of incoming mail does not review or approve the site’s wording before release.
A sender may ask to withdraw a message; no message is deleted on the strength of a sending address alone. A withdrawal process with a reference the sender holds is planned; until it exists, withdrawal requests are recorded and answered case by case. Messages are kept until the correction they propose is resolved. This handling is provisional. A receipted process, in which each message is given a reference its sender keeps and can cite, is planned; no date is set.
Send only what the correction needs; do not include personal information about yourself or others.
To prepare a correction proposal, identify:
- The page URL and record ID, if present.
- The exact wording you question.
- A link to public evidence and a short explanation of the proposed change.
Use public evidence only. Leave out credentials, private correspondence, employer material, and personal information about you or anyone else. Someone else’s personal information is not yours to consent to storing or publishing.