Private evidence options
Use public pages, internal notes, metadata, aggregates, or redacted samples to start and deepen assessments.
Public-evidence demand chain
Every outside-in signal keeps its exact source URL and observation date, and expires when it becomes stale. Proposed sources, signals, and likely priorities pass through review gates before they count in a real workspace. Accepted public-demand hypotheses can influence the ranking of opportunities already supported by internal evidence only within a visible cap; they cannot silently invent projects. The public URL teaser runs in an isolated throwaway user scope that is deleted after the brief is built.
Workstation capture, opt-in and local-first
Workstation capture drafts workflow facts from an employee's own app/window activity. Raw events are summarized on the employee's machine; only a redacted aggregate summary (or, once reviewed, a patch proposal) ever reaches Foundry. Denylists for apps, domains, and folders are applied before any event is tokenized, and nothing merges into a company model without a human accepting it.
Human review controls
No recommendation should change just because extraction guessed a fact. Reviewers decide what is accepted into the model.
Controlled deployment boundary
Teams can run analysis against approved LLM endpoints and keep decision workflows aligned with internal governance.
Auditability for stakeholders
Scoring rationale and sequencing logic remain visible so sponsor conversations focus on assumptions and evidence, not black-box outputs.