The Receipts Room: every claim, with its proof
Opened July 12, 2026 · updated by addition only — nothing gets deleted · También en español
Every company says "trust us". We prefer to say: check us. This room gathers the receipts for what we claim in public: cryptographic fingerprints of our internal laws, dates of recorded acts, exams that can be re-examined, and failures we documented when they happened. It is the rule that governs us inside — the Law of the Receipt: only what is verified and lived gets published — applied to our own storefront.
What a published fingerprint proves (honesty first)
Our internal documents are not public — they are the house's asset. What IS public is their SHA256 fingerprint: a string that identifies one exact file, byte by byte. Change a comma in the document and the whole fingerprint changes; fabricating it backwards is computationally infeasible. Publishing the fingerprint on a dated page pins down two things for any present or future auditor: that the document existed in this exact form on this date, and that if it is ever shown (to a buyer, an auditor, a court), anyone can check nobody retouched it afterwards. We don't ask you to believe us: we leave the signed check for whenever you want to cash it.
The receipts
R1 · We have a real operating constitution, sealed by sovereign act
33 articles governing every AI agent in this company: human sovereignty, a no-exceptions cost rule, anti-illusion-of-capability, work integrity. It is not a decorative PDF: it is law applied daily, and the rules we publish (the 7 rules against agent churn) come from it.
SHA256 fingerprint of the text in force:
B0E2E24F44700040F807C32F5F97426B184759607381957A441874FF6453746DR2 · Our law reforms itself by its own procedure — and leaves a trail
The operating bylaws (35 chapters, the constitution's mechanics) were amended ONCE, by the procedure they themselves define: sovereign formula + act + re-fingerprint. We publish both fingerprints — the historical and the current one — because a governed reform is demonstrated by showing the before and after, not by hiding it:
Current (with the amendment):
172FEB8E1C9980946C54D56FD92F78E0939A4A207092C0038A0C54E372608383Historical (the base version, archived — nothing gets destroyed):
4889123E4F9E9F176AE11A74BE36AB0C37DA7B771B38CC814AC1E56A17FD7309R3 · A BLIND successor passed the method's exam — twice
The question every AI-operated company must be able to answer: does the standard survive if you swap the AI? We tested it twice in one week: an agent with zero prior context — only the role's manual the first time, only the sealed law the second — solved 3 out of 3 real cases from our incident bank: it re-detected a self-modified validator, blocked a fake installer by demanding its fingerprint, and told the CEO the uncomfortable truth without makeup. In the second exam, the blind agent additionally detected, on its own, that the law it was using had not yet been ratified — applying the law to the very document containing it. The full method is published: verifying AI work without being technical.
R4 · The house's third line was also born with law and fingerprint
The line that guards our method (the one signing this room) has its own ratified bylaws — with the Law of the Receipt inside, a dated anti-theater checkpoint (October 2026: if by then this line has produced nothing real, it gets declared out loud), and its death criteria written down.
SHA256 fingerprint:
45ECBB03A73CA804B29AB648312DE184C589B5D6B63793978D0BBC35165D25E6R5 · We publish our failures in the same font as our wins
The Work Journal has a fixed rule: every weekly entry includes "what failed" — because a chronicle without failures is advertising, and this is a record. Entry 1 already tells of an automatic backup that lied in green. The failures cited in our playbooks (the edited validator, the empty screenshots, the ~1,000 lines of patches, the 11 hours of paperwork) happened to us for real, dated, and are recorded in our internal incident bank — every published rule cites its own.
R6 · Truth labels are used outwards too
[VERIFIED] = checked against reality by someone who didn't do it. [CLAIMED] = word not yet checked — and we say so. This discipline appears on our public pages because it is the internal one: no piece of the system may edit its own approval criteria, and success is always declared by whoever did not produce the output.
How to audit us
- Today: this page is dated and the internet's archives will keep it. Every new receipt will arrive with its date. If a fingerprint published here ever "changed" without a reform act explaining it, that would be exactly the fraud this system makes visible.
- Against a document: whoever comes to see an internal document (a buyer of the method, an auditor, a professional) can recompute its SHA256 with any standard tool and compare it with this room. It matches, or someone is lying — and the math doesn't negotiate.
- Against the work: what we publish is checked against what we do: the Journal narrates week by week, building in public explains the channel's rules.
The rule of this room
Add only. No receipt gets deleted or rewritten — if one is ever superseded (a reformed law, a new version), the old receipt stays, marked historical, as R2 already does. The room inherits the internal rule that lineage is part of the asset: an editable history is not a history — it's marketing.
25 United Capital · a human directs, AI operates, written rules govern — built in public at 25united.com.