Review

The gate between "an agent found this" and "this is on my record". Nothing an assistant produces reaches a professor's page without passing through here.

Full screen

1 — The three tabs.

The first two are queues of decisions, not notifications. Nothing on either tab was thrown away — it was kept and flagged, so the call stays yours.

2 — Publish all. Publishes every finding on the tab at once. Reasonable when you have read them; the per-record buttons exist for when you have not.

3 — One professor's findings. Each card is everything one research run produced for one person: their name, their university, and how much is in it — 10 facts, 4 publications.

4 — When it was checked, and how much work went in. Checked Aug 16 is the date a second, independent agent verified these claims against their cited sources. 4 research runs is how many passes contributed. A record checked long ago is worth a second look before publishing.

5 — Show what would be published. Expands the full list — every individual fact and publication, with its source. The number in brackets is how many items. This is the part worth using. Publishing without opening it is trusting a summary; opening it takes a minute and is the whole reason this screen exists.

6 — Publish. Puts these findings on the record. From that moment they appear on the professor's page, and matches and drafts are allowed to cite them.

7 — Reject. Keeps them off the record. They are not deleted — a rejected finding stays recorded as rejected, so the same claim arriving again is visibly a repeat rather than something new.

Why this gate exists at all#

Two separate reasons, and both matter.

The platform will not let an agent publish its own findings. Not "does not by default" — cannot, whatever it is told. The one rule the whole system rests on is that whatever finds a fact is refused when it tries to confirm that same fact, and publishing is the end of that chain.

And it is the bottleneck people miss. Verified findings sitting unpublished silently block everything downstream: matches cannot cite them, so scores come out thin, so drafts have less to work with. If research seems to have run and nothing improved, look here first.

Possible duplicates#

The second tab holds pairs the importer or a research run could not confidently separate — two records that may be one person. Each has a Needs a decision and a Settled state. The reason is always shown, and the reasons differ: same name at the same university, conflicting publication sources, a personal page that points somewhere unexpected.

Left alone, a duplicate does real damage — it splits one person's evidence across two records so neither scores well. But merging two people who genuinely share a name is worse and harder to undo, which is why this is a decision rather than an automatic cleanup.

---

← Batches · Your profile →

Collaboration material#

The third tab is the one that is not a gate, and the difference matters.

Everything listed here is already on its professor's page and already usable. A research finding is a claim about the world and should not be relied on before somebody checks it; this is advice about how to approach a person, and holding it back until approval would reproduce exactly the problem the findings queue has — work sitting unseen, blocking what it was written for, with nothing on screen to say so.

So the tab exists for the opposite reason: so a run's output is not missed because nobody happened to open that professor again. Keep marks it read. Dismiss keeps it too, rather than deleting it, so a later run can see that section was rejected once.

Two labels are worth reading:

• Deliberately empty — a run looked at that section and found nothing small and genuinely useful to say. That is a correct answer, not a gap.
• N flagged — points whose blunt, written-for-you wording tripped the tone check. They were kept because that version is yours, not theirs. The version meant for the professor is checked far more strictly and never reaches you if it fails.

See Contribute, Questions and Prepare for what each section holds.