Research question
Does the evaluator's landscape match a practitioner's in a category the practitioner sold for five years?
Why it mattered
The first run where the coder had real operator ground truth, in Sky's Cintas domain. It tested whether the extraction agrees with lived commercial knowledge rather than with itself.
Method
Evaluator: Google AI Mode, signed in, retrieval on. A four-turn sequence starting from "best uniform rental company," then an FR-garment automotive follow-up, then extraction, with two turn-one draws captured. Sample: one draw per arm. No card was frozen; this run predates the card format.
Direct observations
- The evaluator narrated eliminations with reasons.
- The user-buyer gate (technician comfort and adoption) and the procurement-leverage gate (buyer tiering) both appeared unprompted.
- Closing behavior shifted from asking to offering artifacts.
Interpretation (not established)
Ground-truth agreement in an operator's own category is evidence that the evaluator reads something real about the category, or that the category is well documented and both reader and model learned from the same corpus. E-018 later made the second reading harder to dismiss.
Outcome and remaining uncertainty
Supported: the practitioner's landscape and the evaluator's substantially matched, and gates surfaced unprompted. The shared-corpus alternative was not ruled out, so the agreement is consistent with real reading and with common sourcing.
Evidence and methods limitations
Single evaluator, one draw per arm, single coder with direct operator priors in this category (a source of both signal and bias), no codebook.
Contribution to the research program
Surfaced the role-posture signal, where the evaluator requests information early and offers artifacts late, later folded into C-14.
Follow-on
E-010 and E-011 ran the other Cintas-domain cards.
Created from the canonical Upstream Zero research archive (July 2026). Raw outputs are retained privately and are not published.