October 2: grant preparation added locally
Staff can search up to ten federal Grants.gov results, import public evidence,
and prepare a tailored brief and application draft using actual questions,
word limits and approved public nonprofit facts. Manual public-source grants
can use the same preparation workflow; broad private-foundation discovery and
automatic application-question extraction are not implemented.
Citations, missing facts, eligibility and deadline checks stay visible. Exact-version
human review and approval are required. Repeat-generation protection is implemented;
the application does not submit grants or send donor communications.
Owner-recorded local verification: 390 unit tests and a 22-page build,
with policy, formatting, lint and type checks passing. Earlier browser verification
covered 25 cases and four widths. The audit-only follow-up had focused independent
checks; hosted authenticated and live-model checks remain open.
Implemented locally
Human approval retained
Hosting verified; app release still gated
October 2 hosting reads verified the active account, app directory and tracker
subdomain mapping. Public checks reached app sign-in and found both tracker routes
still displaying the September 28 page before this update.
The actual deployed source, database identity and effective runtime flags remain
unverified. A connected staging database is healthy but lacks two existing source
migrations. App release requires schema compatibility, verified backup and restore,
and hosted access checks. This dashboard update does not authorize database changes.
Live AI requires the reviewed account, exact model, server credentials, spend and
data controls, approved public HVF facts and real-model evaluation. No live AI,
donor import, grant submission or provider activation occurred in this work.