Ask Kabbalist

Google's June post said "power users" would keep the ability to install apps from unverified developers. Here is what the Android sideloading advanced flow does with that promise: a one-day wait, an identity check, a warning on every install, and one quiet catch. Switch it off, and unregistered apps installed this way stop updating.

Why does the Android sideloading advanced flow matter now?

It matters now because app registration became required on 30 September 2026 for participating stores in Brazil, Indonesia, Singapore and Thailand, and Google says the requirement expands globally in 2027 and beyond.

Today's reach is narrower than the headline. Google's Help page starts in four countries, so a phone in the US or UK sits outside the first phase. Neither Google page says plainly whether a browser-downloaded APK is checked yet. The FAQ says unregistered apps "can only be installed or updated when the advanced flow is enabled or by using ADB". Our reading: that is the direction of travel, not today's rule everywhere.

Google's FAQ confirms a "one-time, one-day wait", a fingerprint, face unlock or PIN check, then a choice to enable it "for 7 days or indefinitely". A warning with an "Install Anyway" button still appears. Digital Trends reported on 19 August 2026 that the switch hides in Developer options and a restart comes first. Call it a checkpoint. No, a waiting room is closer: nothing is blocked, you just cannot enter yet.

Our view: phone makers now change phones through dated, staged rollouts. Google moved Android phones from Assistant to Gemini from 4 September 2026, and Apple's iOS 27 Siri beta ships with daily caps it will not number. The detail that matters tends to sit on a help page, and here it does: the wait length is in the FAQ, not the June post.

Four numbers settle most of this. Three are Google's, as read on 23 October 2026; the fourth is our own count across three Google documents.

Take the wait. With no skip command yet, only a promise to "consider this for future versions", it becomes something you arrange ahead, like renewing a passport, not on the day you need it.

Which of these five situations is yours?

Most readers fall into one of five situations, and the sensible move differs for each: a hard deadline, a sideloaded app that needs updates, a developer build, an Android 7 phone, or store-only use.

Find your row, then read across. The right-hand column gives the reason, taken from Google's pages as read on 23 October 2026.

Worked example: assume you finish the restart Monday at 09:00, the wait runs from then, you confirm Tuesday at 09:00 and pick 7 days counted from confirmation, and the app releases every 14 days. Neither Google page says when those clocks start. Monday 09:00 plus 24 hours is Tuesday 09:00, your earliest confirmation. Seven days later the window closes. A release lands inside it on 7 of every 14 days, and 7 divided by 14 is 50%. So a 7-day setting is a coin flip for the next update and no help after it.

Your own sum takes a minute: write down when you will need the app, subtract a day, then a few hours more for the identity check, which Google does not time. That is your latest start. If you sideload once a year, 7 days beats indefinitely, because the open door is the cost.

Whether you are late depends on where you live: the first deadline arrived 104 days after Google's announcement and has passed in four countries, while everyone else waits for 2027. Dates come from Google's 18 June 2026 post and Digital Trends' 19 August 2026 report; the day counts are our arithmetic.

Where do Google's own pages disagree?

They disagree on which Android versions are covered, and only the FAQ states how long the wait is, so no single Google page gives the whole rule, which is why we read all three side by side.

The version split is the sharper problem. An Android 7 phone is covered by the FAQ and absent from the Help page. We will not pick between them; an Android 7 owner should plan for the stricter FAQ reading until Google reconciles the two.

The safety case is thinner than the friction. Google's 18 June 2026 post defends the checkpoints as a way to resist coercion scams, yet it publishes no malware figures. In our opinion, whether a one-day wait stops scams is unproven in public. Narrow starts also rarely stay narrow: DigiYatra became mandatory for international flyers at four airports.

• The warning stays: the FAQ says you still see an unverified-developer prompt, so each install keeps its extra tap.
• Restarts are unclear: the wait is called one-time, but nothing says whether switching the flow off and on repeats it.

This week, pick the one sideloaded app you would miss and ask its developer whether it is registered; if the answer is no, start the wait before you need it. Before setting anything to indefinitely, read Google's Help page first, because that setting leaves the door open for every other unverified app.

The promise fit in three words: ChatGPT free unlimited. You picture an evening of questions with no counter ticking down. Then an image request stalls, or a voice chat drops to a smaller model, and the word turns out to have been doing a lot of work for a single kind of use. OpenAI's announcement was true. It was also narrow.

What does ChatGPT free unlimited actually cover?

Only everyday text chats are unlimited: OpenAI's help page, read on 21 October 2026, says Free and Go users get them subject to abuse-prevention safeguards, while files, images, voice and data analysis keep separate limits. What vanished is the old step-down. PCWorld reported on 7 August that Free users had been held to 10 to 40 text messages every three to five hours on flagship models, then moved to a lighter one. That cliff is gone, which is the real saving, and the reason a long evening of typing no longer ends mid-thought. Uncounted is the more honest word than unlimited. Set it beside what a paid plan buys before it throttles, where the counter still runs.

The model is the second catch. Free accounts get Luna, which Notebookcheck calls the lowest-cost model in the new GPT-6 lineup. It reported on 8 October that GPT-6 Luna began reaching Free and Go chats that day, a day after paid plans received Sol. Yet OpenAI's own help page still names GPT-5.6 Luna as the Free default, so which one answers you depends on whether the rollout has reached your account. Google went the same direction on 9 October, when its free tier dropped to Flash-Lite. At both companies, free now means the small model.

Four numbers decide whether Free fits your week: how fresh the model is, what the sticker price hides, how much room your files get, and what limited used to mean. The help page and Engadget's 12 August hands-on supply the facts, and two of the cards are my own sums from published dates and limits.

The old ceiling is my own sum, and it assumes the windows ran back to back, which no real evening does. Read it as the outer edge of what limited used to mean. The practical gain is not the big number but the end of the mid-conversation downgrade, when a hard question you had saved for later got answered by a lighter model.

Which cap will you hit first?

The cap you hit first depends on what you ask for: plain typing will not hit one, while files, long voice sessions and hard reasoning questions are where Free users meet the limits.

Find the line that sounds like your week, then read across. The table rests on OpenAI's own wording rather than testing, so where it says no number is published, it means exactly that.

Notice what the last column cannot give you: a usage number. OpenAI states none for files, images or voice, so the notice that appears at the limit is your only meter. Engadget reported in August a single weekly limit, viewable under Settings, then Usage, but the current help page publishes no figure for it.

Where the free tier still stops you

The free tier stops you at the tools around the chat, not the chat itself: uploads, images, voice, data analysis and the model's own ceiling, with the price paid in ads and training data. The sharpest dispute is about the model, not the meter. The Decoder wrote that free users "no longer get any access to OpenAI's frontier reasoning models", and OpenAI answered that it was "not accurate to represent this as free users being restricted to a weaker experience". My view, and I hold it loosely: both can be right. A small model can beat last year's default and still be the wrong tool for a tax return or a medical question, and no benchmark on a vendor's own page settles that for you.

Free also costs attention and data. Engadget reported on 12 August that sponsored ads appear below answers and that free accounts are enrolled in model training by default. If you paste a client contract, switch off "Improve the model for everyone" under Settings, then Data Controls, first. Apple has the same problem of an unpublished allowance, as the iOS 27 Siri beta with daily caps Apple will not number shows.

• Think runs on Luna, not Sol, so pressing it deepens the answer without upgrading the model.
• A GPT-6 label in the menu is not a promise: OpenAI says it may not be available for your account yet.
• Upgrading resets your usage rates on Plus, yet OpenAI's help page gives no Plus figures to plan around.

Decide this week whether Free still fits. Take your last ten chats and mark each one that used a file, an image or your voice; it takes a minute. If two or fewer did, stay on Free and press Think for hard questions. If most did, you are the user the caps were written for, so try Plus, listed at $20 a month in August reporting, for one billing cycle and cancel if you never miss the extra headroom.