• The studio keeps finding companies for the wrong business: usually a workspace that was renamed but never repointed. The name is a label; the site the studio read is what every run is written from. Change it under Workspace settings → Workspace configuration → Website, and tick clear what the old site taught it if this is now a different company; your setup answers and learned voice are the other half of what keeps the old hunt alive. See Workspace settings.
• Results marked "simulated": the deploy has no model key configured; the plumbing ran but nothing was researched. On the production studio this should never appear; if it does, that is worth reporting.
• Blank decision-maker names: deliberate, not a bug. A name is filled only when a credible source supports it; blanks are left for your own enrichment rather than guessed.
• Bracketed placeholders in a draft: facts the studio could not verify and refused to invent. They are marked wherever they appear; the pencil on the block is where you fill one in before sending.
• "Sign in with a passkey" does nothing, or says no passkey was used: either you have not made one on this device yet, or you dismissed the browser's prompt. Sign in with Google or an emailed link, then add one from Your account. A passkey lives on the device that made it: adding one on your laptop does not put one on your phone unless your browser syncs them for you.
• The passkey button is not on the sign-in page: the browser cannot make passkeys. Recent Chrome, Safari, Edge and Firefox all can; a very old browser or a locked-down work profile may not. The emailed link works everywhere.
• "This passkey looks like it has been copied": the key answered with a counter that went backwards, which is what a duplicated security key looks like. Remove it in Your account and enrol a fresh one; if you did not duplicate anything, tell us.
• "Include the country code" on the recovery number: a number without one cannot be dialled from where support would be dialling it. Write it as +49 176 1234567.
• "Nothing here was opened just now" on a free run: the first step described the company from what the model already knew rather than from their live site. It is unusual; the glance opens the homepage. When it happens the studio says so instead of heading the block "what we read on your site", because a guess dressed as a reading is worse than a guess. Read what it says with that in mind, and correct the hunting sentence in step two if it is off; the companies it goes and finds come from that sentence, not from the guess.
• A scan that stops without answering: the site took longer than twenty seconds to open, and the studio stops waiting rather than hold you on a spinner. Ask again in a minute; most sites answer the second time. If the glance had already read a page before it ran out, it answers from that page and says only what it actually read.
• "acme.com would not open, so it was not read": the site did not answer at all: nothing listening, a certificate that never completed, a page that refused the reader. The scan says so rather than describing a page it never opened, and the domain is asked again an hour later. If the site is up for you, try once more after that hour; if it is down, no reading is the right answer.
• "A website, not a name, like competitor.com": the field takes an address, not a company name: ueno.co, www.ueno.co and https://ueno.co/work all work, "Ueno" does not. Every real suffix is accepted, .co and .inc among them.
• A run that sits on one stage for minutes: research reads the live web and is the long stage; the activity line under it ("2 searches · 9 sources read") is the sign it is alive. A run has fifteen minutes; one that hits that limit is marked did not finish on the screen you are watching it from, with Rerun right there. Everything else the studio runs behind a screen keeps the same clock: a competitor report, a call prep, an inbound verdict, a Find more, a portfolio upload and a site rescan all read did not finish past fifteen minutes, with the press that starts a fresh one beside them. A rescan that was cut off no longer blocks the next one; press Try again.
• The research behind your first screen: it is the one run that starts itself again. If it is cut off, or never starts at all, it restarts once and says nothing; only a second failure says "The research did not finish, twice", and Rescan your site at the foot of Your brief is the way on from there. Your answers count either way; the questions never waited on the research.
• "The studio hit a busy moment": the model provider was overloaded. The run retries by itself (it says so in its log) before it ever gives up; one that fails anyway offers Rerun (same target, same ticks) and Share this bug with Prospektor, which sends that run's error and step log to us for analysis, only on that click.
• "The research call ran longer than the studio allows": the research step filled its whole budget before it finished. Rare, and Rerun is the fix: it starts again on the same target and usually lands.
• "Suggestions could not be loaded" on the home screen: the line names what the studio actually hit, and Try again re-asks without reloading. If it keeps saying the same thing, Support has the wording it needs.
• "Drafted from a partial brief" on a pitch: the run started before the workspace's research or interview finished. The pitch is real but worth less; sharpening the brief and running again is the fix. A brief you have written yourself never carries the label, whatever the research did.
• "Read from public sources" under your brief: your own site would not open when the studio went to read it, so the brief was assembled from what is published about you elsewhere: indexed pages, directories, filings, press. It is thinner than a brief read off your own pages, especially the voice rules, so it is worth correcting by hand. Sites hosted on Squarespace, Wix and similar platforms sometimes refuse automated readers outright. Use Rescan your site at the foot of Your brief if the site has changed since; a rescan that gets in replaces the label.
• "Your pages come back empty without a browser": your site is drawn in the browser (a single-page app), and the studio's reader has none, so it reads what is public about you instead, the same way as above. A target company built that way gets the same treatment: its pitch says so in the compliance findings, lists none of those pages as read, and never scores the empty pages as evidence about the company. The site is asked again on the next run rather than remembered as unreadable.
• The brief says your site could not be read, while a rescan is running: fixed. The banner reads what is happening now: it says a rescan is in flight while one is, and says a reading is waiting once one has landed, with a link to it. If you still see a stale failure alongside a running rescan, that is worth reporting.
• A paused workspace: the subscription behind it lapsed. The lock screen's resubscribe button reopens the same studio with everything kept. Shared links pause with it and come back with it.
• "The studio's records aren't answering. Reconnecting…": a storage outage on our side; nothing is lost, and you do not have to do anything. The studio keeps asking and opens by itself the moment the records come back, usually inside a few minutes. Only if it is still out after six minutes does it stop and offer you a Try again button.
• The sign-in page during an outage: it loads and works normally. It needs none of the records, so a storage outage no longer keeps you off it.
• Someone can't sign in: their address is probably not on the workspace's allowlist; the owner adds it in Workspace settings → Access → Who has access. Removal takes effect immediately, and so does adding. A storage hiccup is never the reason you are told you have no access: those say so in their own words, in the reconnecting banner above.
• A share link stopped working: it was revoked or expired. Mint a new one from the pitch; Workspace settings → Sharing shows every link's state.
• A recipient says a link is "not available right now": the workspace is paused, so every link it has sent is paused too. Reactivating brings them all back; nothing was revoked and nothing was lost.
• No suggestions on the home screen: normal for a brand-new workspace while research lands, and the shelf refills over time; running pitches and recording outcomes are what feed it.
• A portfolio row that says "waiting its turn": normal. Three companies are researched at a time so none of them is rushed; the row starts by itself when a slot frees. You can close the tab.
• A portfolio row that says "could not be finished": the workspace exists and works, but its research did not land. Open it and use Rescan your site at the foot of Your brief.
• "Skipped …: already has a workspace" on an upload: your agency is already on that company's workspace, so a second one was not created. Switch to it from the workspace switcher.
• "The studio cannot send this run as it is configured": a fault at our end, caught before the run started, so nothing was spent and nothing was lost. Every run is checked against the shape the model will accept before the job is created; a run refused this way would have failed several minutes in. It is logged for us automatically. Try again in an hour, and tell us if it is still there.