Questions and fixes

Getting in#

I can't sign in — "That email and password don't match an account". Either the password is wrong or the account was never created. Check for a confirmation email you never opened. Resist creating a second account: you'll end up with two empty workspaces.

"This account hasn't been confirmed yet". Open the link in the confirmation email, then sign in again. If it's not in your inbox, check spam.

A page says the workspace isn't connected yet. Setup hasn't been finished. If you administer this workspace, follow the installation guide. If someone set it up for you, tell them which page you saw.

Import#

It created two universities from one file. The name is spelled two ways in your sheet, or you typed a different name than the one already in your workspace. Fix the name, then dry-run again.

Rows were "flagged for review". They imported fine — something just looked unusual, like a malformed link or a name that closely matches someone already on your list. Open the professor and check.

Some rows had errors and didn't import. Read the issues list: it names the row number and the reason. Fix your spreadsheet and re-run.

Can I re-import the same file? Yes. Professors already in your workspace are skipped as duplicates rather than doubled. This is the normal way to add rows you appended to a sheet later.

Connecting an assistant#

Where does research actually run? In an AI assistant you already have — Claude Code, Claude Desktop, Claude on the web, Codex, Cursor, Windsurf, Antigravity, VS Code or OpenCode — not in this app. Chapter 10 sets it up, with a page for each one. Nothing in chapter 3 works until you have.

Do I have to deal with API keys? No. Signing in is the shorter path: give the assistant the address, approve it when it sends you back, done. Keys still exist and are still the right choice when you want two separately-scoped connections — see §10.3.

Which assistants work? All of them, plus anything else that speaks MCP over HTTP. Each has its own page with the exact steps: Claude Code, Claude Desktop, Claude on the web, Codex, Cursor, Windsurf, Antigravity, VS Code and Copilot, OpenCode, and anything else.

They do not all take the same JSON block, which is the assumption that wastes the most time. Cursor and Claude Desktop use mcpServers with url; Windsurf and Antigravity use mcpServers with serverUrl; VS Code uses servers; OpenCode uses mcp with "type": "remote"; Codex uses TOML. Use your own tool's page rather than adapting another's.

I pasted the config and nothing happened. Almost always the wrong key name for your tool — see the paragraph above, and your tool's page. Then restart the app, then check the address ends in /api/mcp. If Settings → Research Access shows the connection was never used, nothing reached the server at all.

I signed an assistant in, but I can't find a list of connected devices. There isn't one. Signed-in assistants appear in Settings → Research Access, in the same table as your keys, with a Signed in badge and the name the assistant registered under. If no row appeared there, the sign-in didn't finish, or it finished against a different account. Sessions under Account signs you out of the web app and has nothing to do with assistant connections.

"Verification must be independent." The assistant used the same key to check research that it used to submit it, which is refused on purpose. Issue two keys — one Submit research, one Check research — and tell the assistant which is which.

claude mcp list says "Failed to connect". Check the URL ends in /api/mcp, and check the key hasn't been revoked. Settings → Research Access shows every key and when it was last used — a key that has never been used means nothing reached the server.

Is my key safe in a config file? It only ever grants what you ticked, none of which can send email, and you can revoke it instantly. For Codex the key lives in an environment variable rather than the config file.

Research#

"Research batches aren't set up yet." Open Settings → Research Batches, tick Research batches enabled, and save.

"You have used this week's batches." You've hit your own weekly limit. Raise it in settings, or wait for the week to roll over. The limit exists to keep the pace sustainable — think before raising it.

"Every professor in the queue has already been researched." Nothing is left to work on. Import more professors, or widen which universities are eligible in settings.

A professor's status says "Failed — can retry". Click Retry on the batch page. If it fails repeatedly, skip it and move on — some faculty pages simply can't be read reliably.

A finding says "Sources disagree" or "Couldn't confirm". Exactly what it sounds like, and it's the system working. Open the sources and decide yourself. Never build an email on one of these.

Research is finished but the professor still says "Unchecked". Findings have been submitted but not yet confirmed against their sources, or they're waiting on your review. Check the batch page for the professor's row.

Research ran, but the professor's page barely changed. Almost always the publish gate. Checked findings wait in Review → Ready to publish until you publish them — see §3.4a.

A match won't attach a paper it's obviously about — "is not verified/promoted". Same cause from the other direction: a match can only cite a published publication. Publish that professor's findings in Review, then retry.

Can I add a university without a spreadsheet? Yes — Universities → Add university, type the name, done. Its departments and faculty are what the research finds. See §2.0.

I wrote a private note. Will it end up in an email? No. Private notes are input only: they steer what gets researched and what a draft argues, and are never quoted or hinted at in anything a professor reads.

Do I have to write the research prompts myself? No. Every university, department and professor page has a Research this button that hands you the prompt with the record's name already in it, and Prompts in the sidebar lists every job an assistant can be given. The dashboard leads with the one thing worth doing next.

Can I just ask questions about what I already have? Yes — "who fits me best here", "who is taking students", "why did this one score low". The assistant answers from the record rather than researching again, tells you when something has not been researched instead of guessing, and will ask which you meant if the request could go either way.

Matches#

No match on a professor who's clearly relevant. Matching needs verified research first. Check the Data badge; if it says Imported or Unchecked, put them through a batch.

The score seems too low. Look at the dimension breakdown and at your own profile. Vague research interests are the usual cause — "machine learning" can't be scored against anything specific.

The score seems too high. Check the confidence figure and the risks line on the angles. A high score with low confidence means a thin record, not a great fit.

Can I change which angle the drafts use? Yes. Click Use this angle on a different one, then regenerate the drafts.

Email#

Does the app send my emails? No, and it has no access to your mailbox. You copy an approved draft into your own email client and send it yourself.

A quality check is failing but I disagree. Approve it anyway. The checks are advice, not permission.

Can I get my edits back after changing a draft? Every version is kept — expand the version list under the draft to see them, marked as your edits or generated ones.

I forgot to click "Mark sent". Do it now, and correct the date by logging the message in the Outreach tab. Follow-up reminders and response rates depend on it.

How many follow-ups is too many? Two is a common ceiling, and your own limit is set in your profile. Silence after two is an answer.

Tracking and applications#

A follow-up isn't showing on my dashboard. The dashboard shows the next 14 days. Anything further out is under Tracking → Upcoming, which covers 30.

Why is my response rate so low? Check the sample size next to it. Also check that you've been marking drafts as sent and setting reply sentiment — an unlogged reply counts as no reply.

Should I open applications for universities I'm unsure about? Yes. One at considering costs nothing and keeps the deadline visible.

Data and privacy#

Who can see my workspace? You. Nothing is shared with other workspaces, and uploaded documents stay in yours.

What happens if I revoke a research access key mid-batch? Whatever it was working on stops. The professors involved go back in the queue and appear in a later batch. Nothing already verified is lost.

Can I export my data? Not yet from the interface. Ask whoever administers your workspace.

Still stuck#

Two things to try before asking for help:

1. Reload the page. Much of the app streams in as it loads, and a stalled section usually recovers.
2. Click the number that looks wrong. The drawer of records behind it very often explains it — most surprising figures turn out to be a small sample rather than a bug.

If it's still wrong, note the page, what you expected, and what you saw, and send that on.