How it was builtand where it disagrees with itself

The sample

Twelve categories of the advertising and media supply chain, ten companies each, chosen for share of attention rather than share of revenue. Plus twenty companies that sell on “Outcomes” directly, read deeper.

Home page first, then platform, product, pricing, customers, about and press where they exist. The frame is in sample_frame.csv and the twenty in outcomes20_frame.csv.

What the words mean here

Five definitions carry every count in the study.

What the second pass found

Each sector statement was split into six parts — what the word means there, the claim pattern, the evidence pattern, the transfer read, the commercial risk and the white space — and each part was ruled on separately. 78 rows in sector_propositions.csv.

1. 13
   hold outright

   Every one of the thirteen claim patterns — the counting itself.
2. 46
   hold with a boundary

   The statement stands once the limit is printed with it.
3. 19
   partially hold

   Mostly the transfer reads: how far a sector’s proof travels.
4. 0
   fail

   No sector statement was withdrawn.

Four headlines the second pass rewrote

Each one claimed more than the capture could carry. The replacement is narrower and holds.

1. 01
   The public promise stops before the remedy.

   Replaced “Nobody’s fee moves.” 103 companies publish no pricing page and no contract was read, so the market-wide version cannot be carried.
2. 02
   The adjudicators are lightly adjudicated.

   Replaced “least adjudicated.” Measurement and verification carry 3 third-party figures across 95 outcome claims, but retail media carries 0 across 44.
3. 03
   The proposition often appears before the evidence.

   Replaced “the front door asserts, the case study establishes.” 230 of 536 arguments rest on evidence elsewhere on the same site, which is not the same as a case study.
4. 04
   Operational outcomes stand in for market outcomes.

   Replaced “production stands in for performance.” Creative tech counts assets; workflow counts pipeline and process time. Those are two different substitutions.

What the capture could not see

Five gaps apply across the whole read. Each one bounds a particular kind of statement, and each is in uncaptured_evidence.csv with MAVEN’s ruling beside it.

Eight places the record disagrees with itself

Printed rather than reconciled away. Where a figure on a page differs from a figure in a file, the file is the record.

1. 01
   Two figures for the same idea

   433 of the 740 “Outcome” claims state a figure. 8 more guarantee one. 441 is the count of claims that state or guarantee. Both appear; they are different definitions, and 433 is the one used for “states a figure.”
2. 02
   Two different 90s

   90 of 536 arguments are unsupported where they stand. Separately, 71 carry no evidence anywhere on the site. Nothing links the two counts.
3. 03
   A front-door figure that was withdrawn

   An earlier draft read “55 of 131 home pages use the word.” 137 home pages were captured, and neither number reproduces. What does reproduce: 41 of 137 home pages print the word inside a captured claim, and 116 of 137 carry a claim about an outcome.
4. 04
   The adjudication ran at the sector level

   All four MAVEN columns are blank on all 2,213 claim rows and all three on all 536 pillar rows. The second pass ran on 78 sector propositions, 13 matrix placements, 5 patterns and 5 capture gaps. Those blank columns are dropped from the files published here.
5. 05
   MiQ is counted two ways

   companies.csv records a business-result guarantee on MiQ’s page. The claim rows code the same sentence implied, not guaranteed. The eight guarantee-coded claims therefore exclude it. Both readings are in the files.
6. 06
   Three verbatims failed their check

   2,184 of 2,213 matched the captured page exactly, 26 matched ignoring markdown, and 3 did not match. Those three are flagged in verbatim_check and are quoted nowhere.
7. 07
   The matrix is judgment

   The thirteen positions on claim and proof are a reading of the sector memos, not measured scores. MAVEN’s ruling: the relative topology holds, the coordinates are not scores. The counts sit in the same row so each placement can be re-argued.
8. 08
   Sector names differ between files

   sector_stats.json, the adjudication CSVs and the deck each use a slightly different spelling. segment_id — S01 to S12 and O20 — is the key that holds across all of them.

Two research notes

The reading behind the precedent and the agentic comparison. Each one names its sources and what it could not establish.