No Original Research

Articles must not present original research – unpublished analysis, speculation, or conclusions no reliable source has stated.^[1] Every interpretive claim should be attributable to a published source. The policy exists to keep the wiki's authority borrowed rather than asserted: a reader who doubts a sentence should be able to go and check somebody else's work, not take an editor's word for it.

Reporting on-chain data

Reading and reporting a value from the ledger – a treasury balance, a vote result, an executed payout – is description, not original research, and is encouraged. The ledger is a published primary source, and quoting it is the same act as quoting a document.^[3]

Extrapolating from it is where the line sits. Predicting a token's price, inferring a team's intent from a transfer, or explaining why a vote failed are conclusions the ledger does not contain, and they belong in commentary rather than an encyclopedic article. The practical test: could a second editor, given the same block explorer and no access to your reasoning, reproduce the sentence exactly? Description survives that test. Interpretation does not.

A ledger reading is dated, and the date is part of the claim. The ledger is a primary source that changes, so “the deployed component accepts five action kinds” is not the same sort of sentence as “the paper was published in 1997”. It is an observation true of a moment, and one transaction can end it. On 5 September 2026 this wiki published a correct reading of the live logic component – that it refused a sixth proposal kind the interface was already offering – and an upgrade published twenty minutes later made the sentence false, so the article was accurate when it was written and wrong when it was read.

Nothing in freshness catches that: its clock runs in months and a substrate value can move in a block. The rule that follows is to publish the reading with its resolution – what was read, from what root, on what date – so a reader who finds it stale can tell a measurement from a mistake and re-run it. Negative readings deserve it most, because nothing about “the component does not expose this” changes on the day it stops being true.

Synthesis

The subtler failure is not an unsourced claim but two sourced ones welded into a third.^[2] If source A establishes that a DAO's quorum is 4% of supply and source B establishes that its largest holder controls 6%, an article may state both. It may not conclude that the holder can pass proposals unilaterally, because neither source says so and the inference quietly assumes several things – that the holder's tokens are eligible, that no other gate applies, that the quorum is the binding constraint. Governance parameters interact in ways that make this class of inference wrong far more often than it looks.

Where the conclusion genuinely matters, the fix is to find the source that states it, or to check the mechanism directly and report what the check found – not to leave the reader to notice that two facts were placed side by side to imply a third.

Measuring this wiki's own corpus

This wiki's maintenance passes routinely compute things no source has published: the inbound-link graph over every article, the proportion of directory entries founded in a given year, the share of a category's citations that answer with a bot wall. Some of those measurements are published on the pages themselves, including on notability and verifiability. That is a real tension with this policy, and the line it draws is narrow:

• The corpus may be the subject. A measurement over this wiki's own articles is a statement about this wiki, which is a thing the wiki is entitled to describe and the only party able to.
• The method must be stated with the number. What was counted, over what set, on what date. A figure a reader cannot reproduce is an assertion wearing a decimal point.
• It stops at the corpus edge. “Half the organisations this wiki covers were founded in 2020 or 2021” is a fact about the wiki. “Half of all DAOs were founded in 2020 or 2021” is original research, and it is false besides. The same number becomes a policy violation the moment its subject is swapped.
• The method that gets re-run is the code, not the sentence. A published figure states its method in prose and computes it in a script, and those are two copies of one rule. Correcting the prose leaves the script to re-supply the old answer on the next pass, and it will, because a self-census is re-derived rather than re-read. When a population, a key or a filter is corrected on the page, correct it in the harness in the same pass – and take the boundary from whatever the application itself declares rather than restating it, so the two cannot part quietly. This wiki earned the rule by correcting four pages from a population of 264 to 263 and leaving the harness that produces all four reading 264 behind them.
• A keyword census counts vocabulary, not claims. Grepping the corpus for a term is cheap and reads like a measurement of how many articles assert something. It is not: it measures how many articles use a word. When this wiki censused its own pages for terms belonging to a retired mechanism, both hits inside its own policy pages turned out to be ordinary industry vocabulary in neutral prose about other organisations, not claims about the subject at all. A term count is a shortlist to read, and the count that gets published is the one that survives reading.

Measurements about the wider industry are a different matter: they need a published source, or they need to be presented as what a stated method returned on a stated date, with the sample described honestly. See proposal spam and the governance denominator for what an undescribed sample does to a governance statistic.

The five core policies

Five standards govern every article on this wiki, and they are written to be read together: each one answers a question the others do not.

• verifiability – sets the threshold for including material at all.
• no original research – limits what an editor may conclude from it.
• neutral point of view – governs how the material is framed.
• notability – decides whether a subject earns an article of its own.
• freshness – decides how long any of it stands without a re-check.

They apply to articles about Caper and to articles about every other organisation the wiki covers, on the same terms.