Every meaningful change to the studio, newest first. Including the ones you cannot see: when the tools change how they write, that is listed here too, because a returning user deserves to know why this week's draft reads differently from last week's.
The screen at the end of a packet used to put about five hundred words of commentary between you and the document you came for. Three panels of notes, several numbers that did not add up, and no way to tell which sentence any of it was actually about.
It has been rebuilt around the document. Your resume and the download buttons are the screen now, and anything Asterion wants to tell you is a mark on the exact sentence it concerns. None of it appears until you ask for it.
A red underline means that line does not rest on your words alone: something was inferred, sharpened, or reasoned to from what you gave it. Open it and you get one short note saying what the claim is and what the risk is, with Confirm and Rewrite this sentence beside it. You can edit any line in place, and what you write replaces what Asterion wrote. If you change your mind, the original is still there.
When Asterion wanted a specific it could not honestly invent, a number or a scale, it now says so instead of guessing, and you can type that number straight into the note. It gets worked into the sentence properly rather than stapled onto the end.
Nothing blocks a download at any point in this. The buttons are there before any of the notes are, and they are repeated under the document, so finishing your edits does not mean scrolling back to the top to get your file.
The note Asterion writes about its own work used to open by describing its method: that it had reasoned some of this from your material rather than lifting it word for word. That sentence arrived on every packet and there was nothing you could do with it.
Notes are now twenty-five words or fewer and start at the substance. What the claim is, and what to check. There are two kinds instead of five, because there are really only two questions worth asking about a line: does it rest on your words, or does it not.
The numbers reconcile now too. There is one count on the screen, and it is the number of claims you have not settled yet.
Packets used to be kept for different lengths of time depending on what you were paying when you made them. A free one disappeared after thirty days. A subscriber's stayed indefinitely, including after the subscription ended.
There is one rule now. Every packet is kept for a year from the day you built it, and nothing about that changes when a subscription starts, lapses, or resumes. The subscription decides how many packets you can build. That is the only thing it decides.
This applies to packets you already have, not only new ones. If you had a free packet that was heading for the thirty day cliff, it is not any more.
Come back to a saved packet and it tells you the date it is kept until, shows which claims you settled and when, and lets you carry on from there. You can delete any packet outright whenever you want. Your dashboard also shows your packets whether or not you have a subscription, so a free one is no harder to find than a paid one.
Someone finished a packet, saw the last step on the progress rail still spinning, and asked whether anything had actually been saved. It had been. The spinner was the problem: the final step never marked itself finished, so a delivered packet looked like work still in progress.
That step completes now, and the screen says in ordinary words that your packet is saved and you can close the tab and come back to it. Come back and it tells you when you saved it and how long it is kept.
Separately, if you close the tab or lose power partway through building one, the job description and the answer you typed are waiting when you return instead of an empty form.
The Before you send block reads a posting for the instructions that get an application filtered out when they are missed. It was listing the same requirement twice whenever a posting stated a rule and then restated it as a reminder, which postings do constantly.
Lines about the same requirement are now grouped into one instruction, with every phrase the posting used kept underneath it. Nothing the recruiter wrote is dropped. Two warnings about one thing is how you learn to skim the whole block, which is the opposite of what it is there for.
Job headers on screen ran the organization and the job title together with no space after the comma, so a line read as Quartz Critters, San Francisco Bay Area,Founder rather than Area, Founder.
Downloaded .docx files were never affected, so anything you have already sent was correct.
When the AI behind the tools was briefly at capacity, you could get dropped mid-run and lose what you had entered. That is the worst possible moment to lose anything, because it is usually right after the longest bit of typing.
The tools now wait out a busy period and pick up where you were, using the service's own timing rather than guessing. You may see a short pause instead of an error.
A tester sent back a marked-up copy of every edit she made to her packet by hand, which turned out to be the most useful thing anyone has ever sent about the tool. Six of those edits were the tool's fault, and all six are fixed.
Your name line is now just your name: credentials that belong after it no longer get pulled up into it. Nothing infers your location any more, so a city you never gave it cannot appear. Job headers break at the boundary between your title and the organization instead of wherever the line happened to run out. Education keeps its honors. An employer's name is not split across lines, and the same content cannot land twice in one document.
If you generated a packet before this date and reused it, the next one you generate will be laid out slightly differently. That is deliberate, and it should need less cleanup.
Cover letters could go out with an unfilled placeholder in the greeting, the classic bracketed hiring manager's name. It happened often enough to be embarrassing rather than rare.
Letters are now checked for unfilled placeholders before you ever see them. If a name genuinely is not known, the letter opens in a way that reads as written rather than as a template somebody forgot to finish.
There is a new optional box on the intake screen: anything specific you want worked in. A short note, up to 500 characters.
It exists because people were downloading a packet, opening Word, and adding one line by hand every single time. Usually something the tool could not have known: that you live ten minutes from the office, that you have met the team, that you are relocating anyway.
What you write there goes into the cover letter only, never into the resume, and it is used as you wrote it rather than elaborated into a paragraph of motive you did not claim.
Each tool remembers what you have told it so the next one does not start from nothing. The problem was that it filed the same fact under a slightly different name each run, so one story about killing your own feature could end up stored three times in three wordings.
That mattered more than tidiness. Anything you had explicitly corrected could come back as though it were new, which meant the tools could tell you something you had already told them was wrong.
Now the tools can see what they have already recorded before they record anything else, so a fact you have given once stays one fact, and a correction stays corrected.
Ariadne, Asterion, Daedalus and Thalia now read from and write to one shared profile, so work you do in one is not thrown away when you open another. Answer something once and the next tool starts from there.
You are in charge of it. Everything held about you is visible in your account settings, and clearing it genuinely deletes it rather than hiding it from the page.
Something here break your flow?
Tell me. Changes that make the tools worse for how you actually work are the ones I most want to hear about, and they are the hardest for me to see from the inside. Write to [email protected], and link the entry if it is about a specific change.
