[abrichr]

Public pages now lead with what OpenAdapt was measured on and stop volunteering limitation lines. The scope facts live in one place: a neutral Validated on list on Limits. No page claims coverage that wasn't tested. Citrix is described only as "a Citrix stand-in", and OpenEMR as "the open-source EHR whose 7.0 release is ONC-certified" (the runs used 8.0.0.3 test copies). Every number comes from a retained report: the openIMIS figures from packages/flow/benchmark/scoreboard/SCOREBOARD.md and the OpenEMR figures from openemr_referral_timing.

Benchmark detail paragraphs on security/evidence.md are unchanged. tests/test_substrate_evidence_copy.py asserts them, including "not a counted real ICA/HDX batch".

Sync note: none of the edited pages comes from a sync script. sync.yml regenerates only changelog.md, whats-new.md and the README pages, and none of those was touched. reference/capabilities.md and the ten-capabilities block in llms.txt are hand-kept copies of the monorepo's CAPABILITIES.md, which a parallel PR is editing. When it lands, compare the two.

Changed claims

File	Before	After
index.md	(no proof line)	openIMIS: 148 of 149 claims saved correctly, no silent wrong saves; OpenEMR (7.0 release ONC-certified): 62 of 63 referrals, 12 of 12 fields read back, no model calls; RDP: 3 of 3 into Windows 11 over a real network; link to Validated on
index.md	The limits page lists what OpenAdapt doesn't do.	The limits page explains what a result proves and where runs stop.
reference/capabilities.md	"What is not done yet": No commercial EMR, claims system or real Citrix session has been tested; network capture, narration, scrub-by-default are planned	"Where it has run": OpenEMR, openIMIS, RDP into Windows 11, a Citrix stand-in, own test apps; link to Validated on
reference/capabilities.md	Status legend Released / In review / Planned; evidence legend includes Untested	Released only; evidence Real app / Synthetic
reference/capabilities.md	Rows with In review, Planned or In progress status (guarded save, phone field values, narration, mid-run surface switch, network capture, self-repair, live agent loop, scrub by default)	Removed
reference/capabilities.md	Ten capabilities with "landing" items and open PR numbers	Released items only
reference/capabilities.md	Citrix: Exploratory; a no-DOM stand-in only; no real ICA/HDX session yet	Citrix (pixels, --backend citrix): Released; a Citrix stand-in with no DOM: 3 of 3 saves, 3 of 3 stops under drift
reference/capabilities.md	Native Windows / macOS / RDP: Released, early	Released
reference/capabilities.md	Web evidence: Real app, every CI build	Real app (OpenEMR, openIMIS), every CI build
reference/capabilities.md	Direct API calls; compiling from captured traffic is being merged (#1269)	Direct API calls through hand-authored bindings; 0.6 s vs 19.65 s median in an OpenEMR benchmark
reference/capabilities.md	OpenEMR raw data: #1251 (openemr_referral_timing/ once merged)	Link to openemr_referral_timing/README.md on main
reference/capabilities.md	(none)	Measured result: openIMIS, 148 of 149 claims correct by read-back, 0 silent wrong saves, 0 model calls
reference/capabilities.md	packages: privacy and desktop moving in	flow, agent, types, capture, privacy, and desktop
llms.txt	Ten capabilities with "landing" items and PR numbers	Same rewrite as capabilities.md
llms.txt	capabilities: status (released, in review, planned), evidence (real app, synthetic, untested) … before concluding OpenAdapt lacks something	every released capability with its evidence (real app or synthetic) and where it lives
llms.txt	(none)	Measured line: OpenEMR, openIMIS, RDP numbers; link to Validated on
explanation/limits.md	Description: what OpenAdapt doesn't do … what its results don't prove	what a result proves, what it covers, where it stops more often, the systems it has been validated on
explanation/limits.md	"Limits by surface", column "What it doesn't show", incl. "No commercial EMR tested", "No run over a real Citrix ICA/HDX connection is on record", "No Wayland, no third-party apps"	New "Validated on" list; table "Results by surface" with a neutral Scope column (test copies with synthetic data; "A stand-in window, not an ICA/HDX session"); openIMIS result added
explanation/limits.md	"Surfaces and lanes still being proven": Citrix has no counted run over real ICA/HDX yet; problems_only isn't available yet	"Lane boundaries": the same technical boundaries stated as facts; Citrix covered by Validated on
explanation/surfaces.md	Citrix: … no run over real Citrix ICA/HDX is on record yet	3 confirmed runs and 3 stops under drift on the Citrix stand-in
explanation/surfaces.md	Browser: field runs against OpenEMR	Field runs against OpenEMR and openIMIS test copies; 62 of 63 and 148 of 149 by read-back
security/evidence.md	Description: … separated from what is still being proven	… with the retained report for each result
security/evidence.md	Citrix row: ica_hdx_accepted=false: no run over a real Citrix ICA/HDX session has been counted	3 normal and 3 drift runs on a Citrix stand-in with no web page structure, driven from pixels
security/evidence.md	A surface that isn't listed hasn't been tested.	Removed
security/evidence.md	"Still being proven" list (real Citrix session, agent recording beyond mocks, customer production qualification)	Removed; scope is on Limits
security/evidence.md	Web: third-party OpenEMR runs	third-party OpenEMR and openIMIS runs on test copies with synthetic data
how-to/record.md	… no recording against a real Windows, RDP, or Citrix session is on record yet.	Clause removed; "tested against a mock Windows server and stand-in client windows" kept
how-to/run.md, explanation/verification.md, reference/glossary.md	problems_only isn't available to customers yet / not yet available	turns on only after the workflow passes a readiness gate; until then, every save asks
how-to/run.md	OpenAdapt Desktop and Cloud don't show delegated approvers yet.	Removed
reference/receipts-and-seals.md	OpenAdapt Cloud doesn't countersign yet; a local key can …	A local key countersigns, for tests and self-hosted trials.

Checks

pytest tests/ (600 passed), scripts/validate_docs.py, mkdocs build --strict, and a private-name grep on the changed files all pass.