Here is a sentence that has been on the Sjá page since monitors shipped:
Re-scan URLs on a schedule — hourly, 6-hourly, daily, weekly.
And here is what it delivered: daily, if the window happened to be open.
Those are not the same sentence, and the gap between them is the worst shape a bug can take, because a monitor that didn't run looks exactly like a monitor with nothing to report. Green bell, no alerts, everything fine. You only find out when you go looking for a history and discover the history has holes shaped like the days you had the app closed.
0.7.0 is the release that closes that gap — and then does the same thing again for the citation probe, which had been a button you had to remember to press.
The honest way to make "daily" true is for something to run when the app isn't running. On macOS that means a launchd agent, and launchd agents are a thing you should be asked about rather than given.
So it's off. Turn it on in Monitors or in Settings, and Sjá installs an agent for your user that runs a new subcommand every fifteen minutes:
elyra-sja tick
It opens the database, runs every monitor that's due, notifies you about regressions and errors, records that it ran, and exits. Turn the setting off and the agent is removed. Its output goes to:
~/Library/Logs/Elyra Sja/tick.log
That's the whole feature, and it would be a two-line changelog entry if the interesting part weren't underneath it.
A second process touching the same database as your open app is exactly the kind of thing that works for a month and then eats a week. Four guarantees, each of which is a decision:
A monitor runs exactly once, even with the app open. Both the app and the agent ask "what's due?" and both can get the same answer at the same moment. So claiming is a compare-and-set: monitors::claim writes last_run_at only if the row still holds the value it just read. The loser of the race gets nothing back and moves on. No locks, no leader election, no "please don't open the app while the agent runs".
tick never creates or migrates the database. The app is the only migrator. If tick finds a pending migration it logs open the app once and skips — it does not guess, and it does not helpfully run a migration you haven't seen. A background process that upgrades your schema unattended is a background process that can lose your data unattended.
It prints to stderr only, which is what launchd routes to that log file. Nothing invents its own logging path.
The app re-registers the agent when the binary has moved. An app update relocates the executable; an agent pointing at the old path silently does nothing, forever, while the setting still says on. So on every launch, the app checks and rewrites. An update must not be able to quietly break a schedule.
And the test for all this drives the real binary's tick against an empty HOME, then asserts it exits 0 and created nothing. Not a mock. The thing that will run on your machine at 03:15.
$ tail -f ~/Library/Logs/Elyra\ Sja/tick.log
[09:15:02] 3 monitors due
[09:15:04] elyracode.com/eterm — score 85 → 85, no change
[09:15:06] elyracode.com/sql — score 82 → 74, alert raised
[09:15:06] done in 4.1s
0.6.0 added the citation probe: type the question a user would ask, and Sjá puts it to the answer engines with your own keys and tells you whether you're cited, at what position, and who was cited instead.
The card that announced it said something slightly awkward in the last line: what to read is the trend and the competitor list, not one answer. Which is true — a single probe is one sample of a non-deterministic system — but it quietly asked you to build that trend by hand, by remembering to press a button on a Tuesday.
Now you don't. Schedule repeats the question daily or weekly. Both the open app and the background runner pick up due schedules, a few per pass, because unlike a monitor a probe costs real money. And every scheduled run compares each engine's verdict with its previous one:
Citation alert · "fastest way to run Laravel without PHP-FPM"
⚠ citation_lost — Perplexity
Cited on 8 Sep, not cited on 12 Sep.
Cited instead: frankenphp.dev, roadrunner.dev, laravel.com
✓ citation_gained — Gemini
Not cited on 8 Sep, cited on 12 Sep at position 2.
That's the shape of the thing you actually wanted: not "are we cited today", but the day it changed, and who replaced us.
Gemini joins ChatGPT, Claude and Perplexity — through the Gemini API's Interactions surface with the google_search tool, which per Google's own grounding documentation is the closest API path to what AI Overviews and AI Mode do.
We added it because a competitor review found the obvious hole: Google's answer surface is the one most of your traffic will meet, and we weren't asking it. The key, the model (gemini-3.8-flash) and the endpoint are settings, because these surfaces move; the parser reads both the Interactions shape and the classic generateContent grounding shape, in case your endpoint points at the older one.
Gemini needs its own key under Citation probe in Settings. Like the others, it bills a few cents per probe to your account, not ours.
This one came from using the thing.
An answer engine can talk about you by name and never link to you. "For PHP in-process, Askr is one option…" — no URL, no citation. Under the old model that counted as not cited, which is wrong in a way that matters: being talked about and being ignored are not the same result, and treating them alike hides the middle of your funnel.
So each probe now records whether the answer used your project name, the Organization your page declares, or the domain's own label — whole words, any case, any script:
ChatGPT cited position 2 of 5
Claude mentioned, no link
Perplexity not cited cited instead: frankenphp.dev, roadrunner.dev
Gemini cited position 1 of 3
Shown in cyan in Trends, and — the part we were careful about — kept out of both counts. A mention is not a citation, and it is not nothing. Folding it into either number would have made a metric that quietly lies in one direction or the other.
We'd rather be blunt about this than have you discover it on a card statement:
A daily probe on four engines is four API calls a day, every day, until you pause or delete it under Monitors.
A weekly schedule on one engine is a rounding error. A daily schedule on four engines across a dozen pages is a real line item. The interface says so at the point where you press Schedule, and everything is listed under Monitors where you can pause it.
It installs itself if you're on 0.4.2 or later. Two migrations run on first launch (probe_schedules, and mentioned/mentions on citations). Then:
Turn on Run in the background if you want your monitors to mean what they say.
Add a Gemini key if you want the fourth engine.
Schedule the one question you'd most like to be the answer to — weekly is plenty — and leave it alone for a month.
Then, one Tuesday, you'll get a notification telling you that Perplexity stopped citing you and started citing somebody else. That notification is the whole product.
Close the window. It keeps working now.