diff --git a/method/form-language-prompt-template.md b/method/form-language-prompt-template.md
index fafbd16..f3f9dae 100644
--- a/method/form-language-prompt-template.md
+++ b/method/form-language-prompt-template.md
@@ -285,14 +285,17 @@ Scale notes for the project this grammar will serve: {SCALE_NOTES}.
   away, the same way a pattern-language narrative generated downstream
   should be checked against its selected pattern subset.
 - Salingaros described his original prompt as asking the LLM to "locate
-  open-source images of traditional buildings" and synthesize from them.
-  It's unclear whether that literally happened, as opposed to the model
-  drawing on general training-data familiarity with named buildings and
-  presenting it in the language of research — locating and verifying
-  images is a materially harder task than an LLM can reliably do without
-  a real, working search/browsing tool. Treat the Appendix precedents as
-  indicative, not verified, unless the model you're using actually has
-  that capability and you've confirmed it used it.
+  open-source images of traditional buildings" and synthesize from them,
+  and has confirmed the session used a real web-search-enabled LLM
+  configuration, not training-data recollection alone. Even so, treat
+  "synthesized from imagery" with some caution: current LLM tooling makes
+  it more likely the model synthesized primarily from textual search
+  results and descriptions of buildings than from genuine visual analysis
+  of downloaded images, whatever the prompt literally asked for. Treat the
+  Appendix precedents as text-grounded rather than confirmed
+  image-grounded, unless you've specifically verified that the model/tool
+  you're using performs real visual analysis of retrieved images and not
+  just text search.
 - This document is a formal/tectonic vocabulary for description and
   illustration, not a construction specification: literal historical
   materials it names (rammed earth, adobe, stone vaulting) are
diff --git a/method/paper/paper.md b/method/paper/paper.md
index 1e04795..b469468 100644
--- a/method/paper/paper.md
+++ b/method/paper/paper.md
@@ -178,7 +178,9 @@ required both.
 For this project, Salingaros produced a project-specific form-language
 grammar, the *Ecuadorian Form Language Grammar*, developed after
 discussion with a Quito-based designer and self-builder and synthesized
-by an LLM from open-source reference imagery. Rather than a set of
+by an LLM, using real web search, from open-source reference material —
+though see §3.3 for a caveat on how much of that synthesis was likely
+text-mediated rather than genuine visual analysis of imagery. Rather than a set of
 historical facades to imitate, it is written as a *generative* vocabulary:
 a stratified tectonic sequence from dark Ecuadorian andesite at the
 plinth, through thick limewashed tapial and adobe at the lower and middle
@@ -218,12 +220,18 @@ To check that this generalized, rather than merely described, we ran the
 template a second time on a project unrelated to Ecuador: *Harbor House*,
 a travellers' inn intended to relieve pressure on homeless services, sited
 in Sheffield, South Yorkshire — a different typology, region, and
-climate, produced by the paper's first author. This second run also used
-an LLM configuration with real web search available, so climate figures
-and precedent references were checked against live sources and cited
-inline rather than drawn purely from training-data recollection,
-demonstrating the template is compatible with a stricter, source-checked
-mode of use as well as the original's. The resulting document
+climate, produced by the paper's first author. This second run used an
+LLM configuration with real web search available, so climate figures and
+precedent references were checked against live sources and cited inline
+rather than drawn purely from training-data recollection — a condition
+Salingaros has confirmed the original Ecuador run shared, not a stricter
+mode unique to this validation run. Even with real search available in
+both cases, current LLM tool capability makes it more likely that both
+runs synthesized the form language primarily from textual search results
+and descriptions of buildings than from genuine visual analysis of
+downloaded images, notwithstanding the original prompt's explicit request
+to work from open-source imagery (§3.2) — a caveat that applies equally
+to both runs, not a point of difference between them. The resulting document
 (`projects/harbor-house/docs/sheffield.md`) follows the same section shape as the
 Ecuadorian grammar while producing an entirely different vocabulary — dark
 gritstone and coursed rubble rather than andesite, slate and Yorkshire