How Quana Labs keeps its records
Identifiers, revisions, evidence, AI disclosure and corrections
The rules the lab keeps: every entry gets a permanent identifier, addresses never break, changes are logged as lettered revisions, claims carry an evidence status, AI involvement is disclosed, and results that turned out to be wrong stay on the record.
Quana Labs works on whatever is interesting, so its records have to hold together across very different kinds of work: a disassembly listing, a plant-growth log, an agent evaluation and an over-engineered weekend experiment should all be findable, citeable and honest about how sure they are. This entry sets out the rules that make that possible over years and thousands of entries. It is entry number one because everything else depends on it.
Identifiers
Every entry receives a document identifier when it is first added: QL- followed by a four-digit number. Numbers are assigned in order, are never reused, and never change, even if the entry is renamed, moved to another field or superseded. This page is QL-0001. Experiments, findings and every other kind of entry share the one sequence, so an identifier alone is enough to find anything.
Each identifier has a permanent address of the form /n/1/ that always redirects to the entry’s current page. Entry pages also have readable addresses, such as /entries/lab-conventions/; if one of those has to change, the old address is kept as an alias that redirects to the new one.
Classification
Each entry belongs to one field, and optionally one section within it. The class code combines the two: RE.2 is the second section of Reverse engineering. Fields are deliberately few and stable; new interests usually start as a section of an existing one. This field holds entries about the lab itself, and Rabbit holes holds whatever does not fit anywhere yet.
Everything cross-cutting is a tag. Tags come from a controlled vocabulary: a tag is registered, with a facet (platform, engine, language, technique, tool or topic), before any entry can use it. That is what keeps a large archive from growing hundreds of near-duplicate labels.
Kinds of entry
An entry is one of six kinds, and the kind sets expectations for its structure:
- Investigation: a worked line of inquiry, with question, method, observations and conclusions.
- Experiment: a test with a stated hypothesis, a control that could disprove it, and a recorded result.
- Finding: one specific, citeable result, with its evidence and the limits of what it shows.
- Reference: technical documentation kept current.
- Essay: long-form explanation, history or argument.
- Note: a short field note.
Investigations, experiments and references number their sections, so a passage can be cited as QL-0001 §3.
Revisions
An entry is never silently rewritten. The first published version is revision A; each substantive change after that is the next letter, recorded with its date and a one-line note in the entry’s revision table. Typo fixes do not earn a letter. The change log lists every addition and revision in date order, and has its own feed.
When an entry is replaced by a better one, it is marked superseded, with a link forward, and kept. When it simply stops being maintained it is archived.
Evidence
Entries that make a claim carry an evidence status. The standard for the strongest one is fixed:
Weaker statuses are honest labels, not failures: observed means seen and recorded, hypothesis means proposed and untested, inconclusive means tested without a decision. Refuted means tested and shown false, and refuted entries are kept on purpose and listed with the other findings.
Verification at Quana Labs is done by the author. When someone else checks an entry, it lists an independent review with their name and date; no entry implies outside verification it did not get.
AI involvement
AI tools are part of how the lab works, and their part is disclosed rather than hidden:
- An entry’s AI use says whether AI tools played no part, assisted (drafting, code, analysis, search, with the author checking the result) or generated most of it. A one-line note says what the AI did.
- A hypothesis proposed by an AI system is marked AI hypothesis, not hypothesis, until a person has tested it. AI hypotheses are listed separately on the experiments page as leads, not claims.
- Whatever the AI use, the author is responsible for every claim an entry makes.
This entry is itself marked AI-generated: it was drafted as a starting point, and it becomes the lab’s charter as the author revises it.
Questions
Entries record the questions they leave open. Each question stays attached to its entry and is collected on the open questions page; when a later entry answers one, the question is marked answered and linked to the answer.
Corrections
Errors are corrected in place, with a revision note that says what was wrong. If an error changes a conclusion, the evidence status changes with it, and the revision note says so.