# DEVIATIONS — enquiry-path-census-2026

1. 2026-09-30: first run of the background command failed to parse in bash (nothing measured). Files were re-created with the Write tool and the same pre-registered definitions; no definition changed.
2. 2026-09-30, measured: 219 hosts -> 201 measured, 8 js_shell (under 150 words in raw HTML), 10 unmeasurable (blocked, 4xx/5xx or challenge). Denominator for all rates = 201.
3. Validation (method 2: real headless Chrome, JavaScript executed, rendered DOM): 30 random hosts with a raw-HTML direct route -> 30 of 30 confirmed. The single no_route host showed a route when rendered (JavaScript-injected), so raw-HTML "no_route" is a floor-miss: 1 of 201 raw, 0 of 201 after rendering that host. Not extended to all 201 hosts.
4. Raw-HTML results are a FLOOR for route presence (chat widgets and JS buttons are invisible to it). "Response promise" and "privacy link" are also floors.

5. 2026-09-30, after collection: CLAIM resolved with the measured figures (original planned wording below). No definition changed.
   ORIGINAL: CLAIM: Only SingRank can say how many of 219 Singapore service-business homepages give a visitor a phone, WhatsApp, email or form route without leaving the page, counted the same day with one fixed method.
6. 2026-09-30, after the main results were seen: a price_shown column was joined from the price study (research/_citation-sample-20260925/measured.json, v2 rules and hand review) so the article could cross-tabulate route, price, reply promise and privacy link on 199 sites measured in both. It is descriptive, not pre-registered, and the article says so.
7. No dated immutable copy with SHA256 of the plan was made before collection (the skill was being built the same day); the plan file was written before measurement and has not been edited except item 5.