A Claim Needs a Verdict Next to It

The file is 227,330 bytes over 1,072 lines and 33,093 words, and it is not an essay. It is a ledger. Inside it sit 161 question blocks, 155 source lines and 155 verdict lines, one verdict under almost every claim. The legend at the top allows exactly 2 verdict values and no others: VERIFIED when the source states the claim directly, INTERPRETIVE when the sentence is framing, inference or theory synthesis. Counted in place, 112 of the 155 verdict lines carry a VERIFIED mark, 53 carry an INTERPRETIVE mark, 11 carry both, and 1 carries neither. Strip the overlap and the file reads 101 verified-only against 42 interpretive-only. Two labels, 155 decisions, and the whole point is that a reader can recount them without opening a single source, because the label sits on the same line as the claim it grades. That is the first rule of a working research file: nothing gets to sit in the middle column.

One line refuses the 2 labels. Line 885 reads: REPORTED, NOT PRIMARY-READ, flagged for the fact ledger. That is a third category, and it appears exactly once in 155 verdict lines. The file will not call a second-hand report verified, and it will not hide the report either, so it names the exception and moves on. Behind it sits the ledger itself: 78,840 bytes, 11,547 words and 127 numbered rows, contiguous from 1 to 127 with no gaps and no duplicate numbers. Eight of those 127 rows carry an explicit not-primary-read flag: rows 107, 108, 110, 111, 118, 124, 126 and 127. Eight out of 127 is 6.3 per cent, and 6.3 per cent is a number a reader can hold in their head while reading the rest. The audit did not stay in a chat log. It is 78,840 bytes on disk with the exceptions numbered, which is the difference between saying you checked and leaving something behind that can be checked.

Of the 155 blocks, 39 are written as Claim with a bracket after it, and the bracket is where the writer says out loud that this sentence is framing, assessment or synthesis rather than a finding. The other 116 open with a plain Claim. That split is not decoration. Take the film spine the file is built to defend: the most defensible 1-line portrait of its subject is assembled from 3 named studies, and the block carrying it is marked INTERPRETIVE, with every component traceable to a verified block underneath it. So the synthesis is allowed to exist, it is allowed to be the best sentence in the file, and it is never allowed to pass as a citation. That is the whole craft in 1 move: the interpretation gets its own label rather than its own paragraph, because a paragraph can be skimmed and a label cannot. Thirty-nine admissions on 1 page is not weakness. It is the only reason the other 116 are worth anything.

The questions are coded, and the codes are countable. Across the 161 blocks there are 35 distinct family codes: F1 through F8, D1 through D7 including a D3b that carries the experimental test of the same question, E1 through E8, and A1 through A7 including the selection boundary. Those blocks live under 37 top-level sections, 32 of which are marked ANSWERED in the heading itself. Then 4 iteration sections close the thing out. Iteration 8 closed the last 4 construct gaps, iteration 9 closed the last 2 named constructs and answered the boundary question, iteration 10 was a pre-terminal verification sweep with no new sources, and iteration 11 was the terminal sweep. The last line of the file reads 127 sources, 119 primary-read, 161 claims, all 11 seams construct-complete, zero open questions, 3 consecutive zero-novelty ticks. Four sweeps, and the last 2 added nothing, which is exactly what a sweep is for.

227,330 bytes. 1,072 lines. 33,093 words. 1 ratio.

Across the file there are 147 year citations in the form of an author and a parenthesis. Only 6 of the 155 source lines carry a URL. So for the overwhelming majority of the claims, the address of the evidence is a name, a year and a journal, which is enough for a reader to find the paper and not enough to take the sentence on trust. That is where the verdict earns its place. A bibliography answers whether somebody wrote something. A verdict answers whether the sentence in front of you is what they wrote. Both counts are printed on the same lines, and they do not reconcile exactly: 155 source lines against 127 numbered ledger rows leaves a 28-line gap that the file leaves standing rather than smoothing. The gap is worth more than a tidy number, because a file that rounds its own arithmetic is a file that will round a claim. The probability that an unlabelled framing sentence survives a review and gets quoted as a finding is not 0, and the only thing that has ever stopped it is a line that prints its own verdict.

Here is why the discipline pays, on the record. Entrade built 200 wood-gasification machines across the UK and Europe, and 200 is a count somebody can walk and check. A EUR 75 million industrial group was restructured, and a counterparty can recount it from its own records without asking anybody. The book published on 8 April 2026 runs 393 pages, and anyone holding it can confirm that in 1 minute. Three claims, three verdicts, all three checkable by a third party. Now put the method beside the numbers: 227,330 bytes, 1,072 lines, 33,093 words, 161 claims, 155 verdicts, 127 ledger rows, 8 flagged exceptions and 2 allowed labels. The research file is not trusted because it is long. It is trusted because it prints the ratio of what the sources said to what was inferred from them, on the same line, every time. A claim with no verdict next to it is an opinion with a bibliography. One line, one label, and a human at the end. Keep more of yourself, for longer.