Observed evidence & intelligence controls
Kariant now separates what it read from what it generated. Where a page shows observed evidence, every fact carries the page it came from and when it was read — and you decide how deep, and how expensive, each run should be.
Generated vs observedlink
Most intelligence on Kariant is generated: a model reads your activity and writes a briefing, a SWOT, a set of competitor alerts. That is useful, but it is the model's recollection, not a measurement.
Observed evidence is different. Kariant fetches a page you have already linked — your organization website, a competitor you track, a listing you submitted — and reports what that page actually says right now. Observed evidence is always shown beside generated content, never blended into it, so you can tell the two apart.
Where you will see itlink
- check_circleFor You — reads your selected organizations' own sites so you can check the briefing against the source.
- check_circleDashboard — reads your organization site plus every competitor you have tracked in Kurator.
- check_circleOrganization detail — reads the organization, its subsidiaries, and any competitor sites stored on the record.
- check_circlePlatform → Organization — verifies which root organization URLs in your hierarchy actually resolve.
Reading sources is always a clicklink
Nothing is fetched when a page loads. Opening a dashboard is not permission to spend, so the first read happens only when you press Read sources. After that the result is reused for about six hours, which is why a second read is usually instant.
When a page cannot be readlink
Some pages will not let us read them, and Kariant says so rather than quietly leaving them out. You will see the reason in plain language — the site asked crawlers not to read that page, it returned a bot check, it is behind a paywall or a login, or the domain did not resolve.
In every one of those cases nothing is inferred to fill the gap. A page we could not read and a page with nothing on it are different facts, and Kariant keeps them separate.
Intelligence controlslink
Four settings decide how much Kariant spends thinking and reading on your behalf. They sit in the Intelligence controls panel on each surface.
- check_circleModel tier — Economy for routine refreshes, Balanced by default, Deep for teardowns and real decisions.
- check_circleThinking budget — how much extended reasoning the model may use. Off is fastest.
- check_circleContext depth — how much of your history is included. Minimal is cheapest; Full uses long context.
- check_circleEvidence depth — how many live pages are read: none, up to 3, up to 8, or up to 20. Only the deepest setting will render pages that need JavaScript, because that is several times more expensive.
Whose settings applylink
If you are working inside an organization, the settings are shared by everyone in it, so a team keeps one budget posture. Otherwise they are yours alone. You can also save a different setting for a single surface — deep on the work that matters, shallow everywhere else.
Changing these settings is always free. The cost line in the panel is a real estimate of what the next run will spend, not a rough guide.