The Record Standard

Why a Standard

Field Assembly publishes three kinds of work. A reference work records what institutions publish about their own rules. An assessment work records a judgment about a body of evidence that does not agree with itself. A market record records what a set of published figures did over a period, and what they can and cannot establish. All three are only as good as their discipline, and discipline claimed is worth less than discipline you can check. This page states the commitments every Field Assembly work is kept to, so that a reader, an advocate, or an opposing party can audit them rather than take them on trust.

These commitments existed as working rules inside each project before they were written down here. This page codifies them; it does not invent them.

1. What This Governs

Every work that carries the line "a Field Assembly record, kept to the published record standard" is governed by this document: what its pages may state, how its sources are captured and kept, and how its history can be verified. Sections 2 through 7 apply to all three kinds of work. Sections 8 and 9 add what assessment works and market records each require beyond them. Each work also publishes its own dates, reviewer status, and change log on its own pages; where a page has not yet had its human review, the page says so. Each work also names, in its own repository, the version of this standard it was collected under, and earlier versions stay published so that a reader can see which rules a given observation was made under.

2. First-Party Sources Only

Every fact on a page traces to what the institution itself published: its own pages, its own documents, its own dates. Third-party guides, news coverage, and general knowledge are never sources, never fill a gap, and never confirm anything, even when they are certainly right. Where two of an institution's own sources disagree, both are quoted and the disagreement is recorded; nothing is reconciled on the institution's behalf.

3. Verbatim, Verified Mechanically

Quotations are exact, artifacts included. Every source is captured in full into a retained packet before a page quotes it, and a checker proves, character for character, that every quotation on a published page is a contiguous substring of its captured source; that every phone number, email address, and external hostname on the page appears in the captured sources; and that advisory language appears nowhere outside quotation. A page that fails the checker is not published. When a check fails, the page is fixed; the packet is never padded to make a fact pass.

4. Two Dates on Everything

Every source carries the date the institution gave it (or the recorded fact that it published none) and the date it was retrieved. Every page states when its sources were last checked, and re-verification runs on a stated cadence. A stale statement is quoted with its date and called historical; silence since is never treated as confirmation.

5. Absence Is a Finding

Where an institution publishes nothing on a question, the page says so and names what was searched. An absence stated at the scope actually searched is information a reader can act on; an absence left blank is a reader guessing; an absence papered over with material from another state, another institution, or general knowledge is a fabrication. The works record what institutions fail to say with the same care they record what institutions say.

6. Nothing Is Destroyed

Captures are kept. When a re-verification pass finds a source changed, the new capture is retained beside the old one, and the record of every pass (what was checked, when, with what result) is kept separately from the captures themselves. The history of what an institution's pages said, including versions the institution has since replaced, stays queryable. Nothing in a capture store is ever edited; mistakes are superseded by later captures, on the record.

7. Dates You Can Prove

A capture date asserted by the person who holds the capture proves little. So each work's capture corpus is hashed into dated manifests joined in an append-only chain, and each chain entry is timestamped by parties outside Field Assembly's control: cryptographic timestamp tokens from independent authorities, plus a proof anchored in the Bitcoin blockchain. Altering any past capture or manifest breaks every later link.

The consequence is checkable by anyone: that a given capture existed, in exactly its current form, no later than a given date, can be verified with standard tools by someone who trusts neither Field Assembly nor any single timestamp provider. Each work's repository documents the verification steps.

8. Assessment Works: Judgment on the Record

A reference work quotes sources that are authoritative by definition; whether the institution is right is not the work's burden. An assessment work has the opposite problem. Its sources are studies and reports that contradict each other, and its product is a judgment: what has been asked of a literature, what has been answered, and what was conspicuously never reported. A judgment needs discipline a quotation does not.

So every appraisal is made twice, independently, against a written threshold for agreement, and where the two passes disagree the disagreement is reconciled on the record rather than quietly resolved. The evidence under an appraisal is bound to it by the same character-for-character check that binds a quotation to its capture. Judgments are versioned: when one changes, the earlier judgment stays visible with its date, because a reader is entitled to see that the view moved and when.

Ratings are never displayed bare. A level on a scale means nothing without the scale, so the scale, its definitions, and what it was applied to travel with it. A scale borrowed from someone else is named as theirs, and a scale Field Assembly built is never dressed as an established one. Where the evidence disagrees with itself, the disagreement is published rather than averaged away. And a question nobody has tested is itself a finding, stated as such and kept distinct from a question Field Assembly has simply not assessed yet.

9. Market Records: Figures With Their Working

A market record reads published figures — a statistics bureau's index, an agency's market report, a mandatory-reporting series — and states what they did over a period, compared with which period, what evidence might explain it, and what the figures cannot establish. Its reader is usually someone holding one price claim and no way to check it.

The evidence here is the source's file itself. A market record retains the response exactly as served, byte for byte, with its headers, its address, its retrieval time, and the source's own stated period, addressed by hash. Nothing is extracted into a packet body, because a statistical release is its file. A retrieval that fails is recorded as a failure; nothing is substituted from somewhere else, and a later run supersedes rather than replaces. Every figure that enters the record is parsed from a retained file by a recorded method, and a parser that cannot find what it expects fails the source rather than guessing at it.

An edition may not quote a figure that is not in a computed set regenerated by tool from those parsed rows, and that computed set is published with the edition. Every calculation the edition leans on — a percentage change, an average, a unit conversion, a comparison period — is stated in a public data dictionary with its formula and its rounding, and a figure Field Assembly calculated is distinguishable from a figure a source reported.

Comparisons hold product, grade, size, unit, channel, geography, and market stage constant. An index is never presented as a price, an import unit value never as a retail price, a public market measure never as a particular supplier's cost; where a comparison mixes dates or stages, the edition says so. An edition may describe two series moving together and may not say that one caused the other unless a source says so, quoted. Several reports from one institution are not independent corroboration of each other.

Editions keep their address. A published edition holds its permanent address, its publication date, and its version; a correction appears as a dated revision with a new version while the original text stays visible, and is noted in the next edition. Where a market record makes its own price observations of named sellers, those sellers appear in editions by letter, with the mapping on the work's methodology page.

What a market record will not do is as important as what it does. It does not state whether a contractual adjustment is permitted. It does not find that a price is unfair. It does not forecast. It reports market evidence and leaves the argument to the person making it.

10. The Record Is Free

The record itself (the pages, the quoted sources, the change logs, the capture history) is free and complete, with no accounts, no intake, and no analytics. If Field Assembly ever offers paid services on top of these works, payment attaches to formalization, such as certifying a capture for a proceeding, and never to access. No part of the record sits behind a fee, and no reader is profiled for having read it. Why access stays free — the economics behind this commitment, and the succession design behind the works — is stated on its own page, Permanence, and that statement is itself externally timestamped.

11. Corrections on the Record

Errors are corrected in each page's change log, dated, with the wrong version acknowledged rather than silently replaced. Corrections go to hello@fieldassembly.net. The works are independent of every institution they document, of detector and software vendors, and of law firms; each page states the independence relevant to its own subject.

12. What the Standard Does Not Claim

The works present potentially relevant official sources. They never determine which rule governs a situation, never calculate deadlines, never recommend arguments, never predict outcomes, and never give legal advice. What any source means for a particular case is for the institution and its processes to determine.

A market record adds three refusals of its own. It never states whether a contractual adjustment is permitted under anyone's agreement, never finds that a price is unfair, and never forecasts. Published figures can show a buyer what a market did; what that means for one contract is between the parties to it.

Field Assembly is also not a web archive, and makes no claim of breadth beyond each work's stated scope. It is not a capture service that can be hired after the fact; a record either was being kept when the moment passed or it was not. And it is not a monitoring product; the works exist for the reader in the middle of something, and for the later moment when what a page said, on a date, has to be shown.

13. Changes to This Standard

This standard is versioned and dated, and its own history is kept to its own rules: every released version is retained unchanged, each release is hashed into an append-only chain and externally timestamped, and changes are listed here with their dates. A work states which version it was collected under, and observations made under an earlier version are not brought into retroactive compliance with a later one. The point of versioning is that the record honestly shows which rules were in force when.

Version 1.0, August 30, 2026. First publication, codifying the working rules already followed by Rules & Record, Gathered Work, and Board & Border.

Version 1.1, September 14, 2026. Adds assessment works, section 8 on this page. A minor release: it extends the standard without changing what an observation is. Like 1.0 it is descriptive, stating practice the founding assessment work already followed.

Version 1.2, September 23, 2026. Adds market records, section 9 on this page. Also a minor release, and also descriptive: every requirement in it is practice the founding market record, Provision Record, already followed. It makes one change of form rather than of kind. For a market record the retained observation is the source's file as served rather than text extracted from a page, because a statistical release is its file. That applies to market records only; it does not change what an observation is for any other work.