The Companion, version one

Twenty minutes, with a coffee. Palmer's image: a man sitting down with the morning paper. That is now the design constraint, and every timed section of an issue carries its own minute count so the promise is checkable. The room is the one part with no clock on it, on purpose. Total: twenty minutes. Anything that does not fit gets cut rather than compressed.

Every link the reader needs. The first draft told the reader to read two pages of a discussion section that sits inside a 216 page PDF. That is the failure the whole product is against: telling someone to engage with a source and then not handing them a usable door into it. Every issue now opens with a named door, a time cost, and a line saying what is optional.

Three questions, written as questions. The six were labelled "The claim," "The check," "The load-bearing assumption." Palmer's read: that is machine writing, and it reads like a worksheet. The labels are gone. Three questions, in plain sentences, in his voice. The self-check that closed the six survives as one line under them, since it costs nothing and it is the only line that measures anything.

A room, free, that he already has. A Discord server. It is what newsletter operators actually use for this, and one operator's Discord holds about a fifth of her paying subscribers (Inbox Collective). Discord's current free tier and any member limit were not checked in this session, so confirm both before the invite link goes in an issue. One channel per issue, nothing else, opened at issue one. The same piece carries the warning worth heeding: a community started as a growth tactic reads as one, and the sign a newsletter is ready is that readers already talk to each other rather than only back at the writer. One channel is the cheap way to find out whether they will.

The vocabulary. New section, and it is the payoff Palmer named: three terms per issue, defined in one line each, that the reader can use correctly in a conversation that week. It is the difference between having read something and being able to hold a position on it on Thursday.

Palmer, 2026-09-07, on the sentence "Hold that word 'often' carefully, because it is where the whole argument turns": that is what AI writing sounds like to him, and it annoys him. The fix he gave, which is now the rule for this product and for anything else written under it:

Because correlation does not necessarily equal causation, let's approach that word "often" with a critical lens instead of taking it at face value.

The rule underneath it: name the move. Where a sentence is pointing at a critical thinking operation, say what the operation is called. Do not gesture at it with writerly phrasing and leave the reader to feel that something important just happened. The writerly version flatters the writer. The named version hands the reader a tool they can use on Thursday, which is the entire product.

This is a style fix and it is also the pedagogy. Palmer's own account: the most foundational thing he learned at school was that correlation does not equal causation, and it was drilled in. A letter that teaches critical thinking should teach in the same way, by naming the move and then running it on something real. The vocabulary section already worked this way; now the prose does too.

Changed in this pass: the EEG sentence above; the Wharton comparison now names randomized controlled trial and says what randomization earns; the R and W self-check now names self-regulation; issue 002's note names the fact and inference distinction instead of saying stories are where the sloppiness lives; issue 003 names the confidence and accuracy gap; and issue 001's vocabulary swaps statistical power for correlation and causation, since that is the move the item actually turns on and the plain version of the power point already sits in the background prose.

Palmer, 2026-09-07. Two changes.

Every minute count comes off. The per section clock and the "2 min" labels are gone from the page and from every issue. Twenty minutes is stated once, on the masthead, and it still binds the writing. Visible timers on each section make a letter read as a worksheet, which is the opposite of a man with a coffee and the paper.

The sections are renamed. "The item" becomes The story. "The background" and "The vocabulary" merge into Background & Taxonomy, since the taxonomy is the naming of the terms and the terms belong with the background that earns them. Three terms a week, still there, still the Thursday payoff, now sitting under the section that sets them up. The rest keep their names: Three questions, The forecast, Write it down, The room, From Palmer.

Masthead

THE COMPANION. Twenty minutes a week on the thing everyone is talking about. From Intelligence Meridian.

Hero

You want to understand AI. Reading about it is the slowest way.

Every week there is a new model, a new claim, a new study. You read about it, you nod, and by Thursday somebody asks what you think and you find you are repeating a headline back to them.

The Companion is twenty minutes on a Monday. One real thing from the AI world, the background you need to think about it, three questions, and a paragraph you write yourself. By Thursday you will have a position you can defend, in words you chose, on the subject that is going to come up anyway.

Your first two months are free. No card.

The argument, in three paragraphs

Here is the finding that this whole thing is built on. In 2024 a team at Wharton ran a trial with close to a thousand students. The ones given free run of GPT-4 did 48 percent better on practice, and then 17 percent worse than students with no AI at all on the exam where the tool was taken away. The ones given a version that only offered hints did 127 percent better on practice and held level on the exam. It was the same tool in both groups. What changed was whether it was allowed to do the thinking. (Wharton)

A Microsoft Research survey of 319 knowledge workers found the same shape in adults at work: the more people trusted the AI, the less thinking they reported doing. The more they trusted themselves, the more. (Microsoft Research)

So the way to understand AI is to keep doing the thinking while you use it, on purpose, on a schedule, with something real in front of you. Twenty minutes a week is enough. This is that twenty minutes.

What arrives every Monday

The story. One real thing: a paper, a launch, a claim somebody made in public. Linked properly, with a named door into it, so you read the source instead of a summary of a summary. The source links are optional; everything the questions need is in the letter.

Background & Taxonomy. What you need in order to think about it at all: who made it, what they measured, and the names of the things involved. Three terms every week, defined in one line each. You cannot reason about something you know nothing about, and you cannot argue about something you have no words for.

Three questions. The same three shapes every week, on different things, until the shape is a habit you carry into rooms that have nothing to do with AI.

The forecast. One claim that resolves within thirty days. You write a number down. Next month you find out. It is the only part of the AI conversation that keeps score.

Write it down. A paragraph, yours, by reply. Palmer reads every one and answers the best in the next issue. Nothing you write is graded by a machine, ever.

The room. A Discord channel for each issue, free, open to everyone on the list. Answer somebody. Or read and say nothing, which is also allowed.

What it is not

It is not a news digest; you can get twelve of those. It is not a course; there is no certificate and nothing to complete. It is not an AI that tutors you; the questions were drafted with a machine's help and edited by a person, and the machine goes no further than that. It is twenty minutes a week with the part of thinking you are about to hand to a tool.

Who it is for

People who use AI every day for work and can feel the muscle going. Musicians, designers and writers who want to know what these tools are doing to their craft before the tools decide for them. Students and recent graduates who miss the seminar room. And anyone who is curious about AI and tired of being talked at about it.

The free two months

Eight issues, free, no card. At issue eight you get a plain question: did this change how you think about AI, yes or no. If yes, what comes next is priced then. If no, you leave with eight paragraphs in your own hand and a forecast score, which is more than most subscriptions leave you with.

The line under it

Eight Mondays. Twenty minutes each. Free.

[Subscribe: email field, one button, "Start Monday"]

Cadence. Monday 07:00 ET. One issue a week, in blocks of eight, and the eighth is a return week: score the forecasts, reread issue one.

The twenty minutes is a contract. Stated once on the masthead and nowhere else. Palmer's call, 2026-09-07: a clock on every section reads as a worksheet, and a letter that keeps telling you how long each part will take is not a letter. The constraint still binds the writing. If a draft runs long the background gets cut, never the questions, and an issue that takes forty minutes has broken the only promise the masthead makes.

Production. The same weekly pass that produces Meridian Passage. The machine assembles item candidates and drafts the background, the vocabulary and the three questions; Palmer picks the item, rewrites the questions in his own words, and writes the note. Budget: two hours on top of the Passage pass.

Where it lives on the site. A blog collection named The Companion, unlinked from navigation and password protected while it is behind the scenes, with the section page above it. Email off the same collection. Squarespace's current password protection, collection and email campaign features were not loaded or checked in this session; confirm each in the account before the build starts, since the plan depends on all three.

The room. A free Discord server, one channel per issue, invite link in every issue and on the section page. It stays one channel per issue until people talk to each other without being asked. No roles, no bots, no voice channels, no second server. Palmer is the only moderator and the only rule is the one under the questions: say what you think and say what would change your mind.

Name the move. The standing writing rule for every issue, from v3. Where a sentence points at a thinking operation, name it: correlation and causation, base rate, randomized trial, inference, self-regulation, considering the opposite. Writerly gestures at an important moment get cut. A reader should finish an issue holding a named tool rather than a feeling that something clever happened.

Replies. Every reply read. The best one or two quoted in the next issue, with permission. The reply is the private door and the room is the public one, and people who will never post in a room will write an email.

What is measured. Opens are noise. Four numbers: replies per issue, forecasts logged, people still opening at issue eight, and how many people post in the room without being prompted. If replies are zero for three weeks the questions are wrong. If the room is silent for three weeks the room closes and nothing is lost.

What version two would add, only if version one earns it. Replies visible to other subscribers, then a small paid tier, in that order. Never a platform with a monthly bill before there is a monthly income.

Subject line: Everyone has an opinion on this study. Almost nobody has read it.

The story

In June 2025 a team at the MIT Media Lab published a study called "Your Brain on ChatGPT." Fifty-four people wrote essays: some using ChatGPT, some using a search engine, some using nothing but their own head. All of them wore an EEG cap. Within about a day the coverage had turned it into "AI makes you dumber," and that is the version almost everybody is carrying around.

Three doors, shortest first. The abstract on arXiv is the authors' own paragraph. The FAQ on their page, brainonllm.com, is the fullest honest account of what they found and what they say it does not show. The full paper runs 216 pages with appendices and you do not need it today. Everything the questions require is below, so all three doors are yours to take or skip. Read the press coverage afterwards if you want to see the gap.

Background & Taxonomy

Three things, and then you know enough.

What the machine measured. An EEG cap records electrical activity at the scalp. Researchers look at how strongly different regions fire together, which they call connectivity. More connectivity often gets read as more engagement. Because correlation does not necessarily equal causation, approach that word "often" with a critical lens instead of taking it at face value.

What they found. The group the paper calls Brain-only showed the strongest connectivity. They typed on a Mac in a plain text editor, forbidden from using a model or any website. The search engine group sat in the middle. The ChatGPT group showed the weakest. The ChatGPT writers also felt the least ownership of what they had written, and when asked minutes later to quote a line from their own essay, many could not. A smaller group of eighteen came back for a fourth session with the conditions swapped, and the ones moving from ChatGPT to writing unaided looked less engaged than people who had been unaided all along. The authors call the accumulated cost cognitive debt.

What the critics said. The paper is a preprint, which means it was posted publicly without going through peer review first. In January 2026 a commentary argued that a study built this way would need around 159 participants to have a fair chance of detecting the effect it went looking for, that some of the comparisons rest on two to four essays per group, that the authors limited the analysis to a thousand statistical tests without saying clearly how they chose them, and that lower connectivity does not necessarily mean less thinking; it can also mean a shift in attention, a change of strategy, or lower motivation (the commentary is here). The authors have revised the paper since it went up.

One more piece, because it changes the picture. The Wharton trial from the argument above is a randomized controlled trial: people were sorted into the groups by chance, which is the design that earns the right to talk about cause at all. A thousand students, unrestricted GPT-4 hurt later performance, hint only did not (Wharton). Put the two side by side while you answer. They are different kinds of evidence, not two votes on the same question.

The three terms, so you can say something exact on Thursday.

Preprint. A paper the authors have posted publicly before peer review. It is a real result and it is an unchecked one. Saying "it is a preprint" is a reason to read it carefully, and not a reason to dismiss it.

Correlation and causation. Two things moving together, against one thing making the other happen. Almost every AI headline you will read this year turns the first into the second. This is the move to make first, every time, on everything.

Cognitive debt. The authors' own phrase for the cost that builds up when you let a tool do the thinking and then have to think without it. Their evidence for it is the eighteen people who switched conditions, which is worth knowing before you use the phrase at dinner.

Three questions

1. Say what the study found in one sentence, the way you would actually say it out loud to a friend. Then write the sentence the headlines used. Put them side by side. What did the headline add that the researchers did not?

2. The finding leans on the idea that less brain activity means less thinking. Suppose that turns out to be wrong. What is left standing? Write down what survives, because something does.

3. Make the case that the critics are nitpicking and the study is basically right. Three sentences, best you can do quickly.

Now mark each of your three answers R or W: R if it came from something you read this week, W if you worked it out yourself. That is a self-check, and it is the one habit the research calls self-regulation. Count the Ws.

The forecast

By 6 October 2026, will the MIT preprint have been accepted for publication in a peer reviewed journal, as reported on its arXiv page or on brainonllm.com? Write a number between 0 and 100. Write one sentence saying what would move it ten points. Keep it somewhere; issue eight scores it.

Write it down

A paragraph, by reply. The prompt: the headline said one thing and the paper said something narrower. Where else this week did you accept the headline version of something and never go back for the rest?

The room

[Discord invite] The channel for this issue is #001-your-brain-on-chatgpt. Post your one sentence from question one. Reading other people's is the fastest way to see how much room there is between "what the study found" and "what everyone says it found."

From Palmer

I picked this one first because everyone has an opinion on it and almost nobody has opened it. Fifty-four people, a preprint, and a measurement that had turned into "dumber" by the time it reached you. I do not think the study is junk and I do not think it shows what people say it shows. It is small, the finding is real enough to take seriously, and a randomized trial twenty times the size found the same shape when the tool was allowed to do the work. Holding both of those at once is the skill. Tell me which of your three answers you actually worked out.

Subject line: Four companies shipped the same feature in one month. Ask why.

The story

Between late July and early September 2025, four companies shipped versions of the same thing. OpenAI added study mode to ChatGPT (announcement). Anthropic opened Claude's learning mode to everyone (report). Google put Guided Learning in Gemini (TechCrunch). Perplexity added a Study Mode for students (changelog). Each one does the same basic thing: instead of answering you, the model asks you questions and gives hints.

Read the OpenAI announcement. It is short, it is the clearest of the four, and it admits the most.

Background & Taxonomy

Socratic questioning is old, and the version most teachers use comes from Richard Paul, who sorted the questions into six kinds: clarifying, probing assumptions, probing evidence, asking for other viewpoints, tracing consequences, and questioning the question itself (summary). A study mode is a chatbot wearing that list.

The evidence that it works when a machine does it is thin and specific. The best result is the Wharton trial from issue one: a hint only version of GPT-4 avoided the harm the unrestricted version caused and left students level with the control group. Level, not ahead. A Harvard physics trial with 194 students found a tightly constrained AI tutor beat active learning classes by a wide margin, and the reviewers concluded that a good AI tutor is a strategically unhelpful one, because it withholds the answer (review).

Two facts to carry into the questions. OpenAI's own announcement says study mode is built from custom instructions rather than a new model, and that it can be inconsistent. And all four are free or bundled. Nobody is charging for the Socratic layer.

The three terms, so you can say something exact on Thursday.

Socratic method. Teaching by asking rather than telling, so the learner produces the answer. It is slower on purpose.

Scaffolding. Support that is deliberately removed as the learner improves. A tutor that never removes it has built a crutch.

System prompt. The standing instructions a company writes for a model before you ever type anything. Study mode is mostly a system prompt, which is why it can slip.

Three questions

1. Write what each company says study mode does for you, then write what it does mechanically: asks questions, withholds answers. Are those the same sentence? Say where they part.

2. All four launches assume you will choose the slow mode when the fast one is a single toggle away. What is your honest evidence, from your own week, about whether people do that?

3. Make the strongest case that four companies moving in one month is a good faith response to the evidence rather than cover against the "AI makes you dumber" story. Three sentences.

Mark each answer R or W. Count the Ws and compare with last week.

The forecast

By 6 October 2026, will any of the four companies publish a usage number for its study or learning mode: share of sessions, daily users, anything countable? Number between 0 and 100, and one sentence on what would move it.

Write it down

Turn study mode on in whichever tool you use, and ask it something you actually need to know this week. Describe what happened in a paragraph, and say whether you turned it back off.

The room

[Discord invite] #002-study-mode. Post the moment you gave up and asked for the answer, if you did. Everybody did.

From Palmer

[Last week's replies, quoted with permission, or a line saying nobody wrote back and what I make of that.] This week the item is a pattern rather than a paper, which is a different kind of thinking. Four companies moving in one month is a fact. Why they moved is an inference, and confusing the two is the most common failure in tech commentary. Question three is the hard one, and it is hard because the honest answer is probably both.

Subject line: What did you hand over this week?

The story

You. One piece of work you did with an AI tool in the last seven days: an email, a plan, some code, a mix note, a caption, a summary of something you were meant to read. Pick the one where the tool did the most. Have it open in front of you.

Background & Taxonomy

In 2025 Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers and collected 936 first hand examples of them using generative AI at work. The finding was clean: the more confident people were in the AI, the less critical thinking they reported doing; the more confident they were in themselves, the more. And the shape of the thinking changed. People described doing less of the work and more of three other things: checking the output, fitting it into everything else, and steering the task (study). It is self reported, so it measures what people believe they did, which is its own kind of interesting.

A 2026 study of 589 students and early career workers drew a line between two ways of handing work over: taking the output as finished, and taking it as a starting point. Both felt equally helpful at the time. Only the first predicted lower motivation and the sense of having given the thinking away (study).

The oldest idea here comes from a psychologist named Robin Hogarth. Some environments are kind: the feedback is fast and honest, so experience teaches you. Some are wicked: the feedback is slow, missing or misleading, so experience teaches you nothing or the wrong thing (explainer). Using an AI tool for knowledge work is wicked. The output looks finished whether or not it is right, and you rarely find out which, so your confidence grows without your accuracy growing with it.

The three terms, so you can say something exact on Thursday.

Cognitive offloading. Letting something outside your head do part of the thinking. A calculator is offloading. So is a shopping list. The question is never whether, it is what.

Kind and wicked environments. Whether the world tells you when you were wrong. Chess is kind. Hiring is wicked. Prompting is wicked.

Automation complacency. The documented tendency to check a machine less the longer it has been right. It shows up in pilots and radiologists, and now in you.

Three questions

1. Could you explain that piece of work to somebody, right now, without the tool and without rereading it? Try it in your head for thirty seconds. Say honestly where it went thin.

2. Name one thing in the output you checked yourself and one thing you accepted. How did you decide which was which, or did you never actually decide?

3. Make the strongest case that handing this one over was exactly right and doing it yourself would have been vanity. If the case is strong, say so.

Mark each R or W. This week it should be nearly all W, because the item was you. If it is not, that is the finding.

The forecast

Pick your own this week. Find a claim in the AI news that resolves by 6 October 2026, write it in one sentence, give it a number, and send it in your reply so it can be scored in issue eight.

Write it down

Redo one small piece of that work without the tool. A paragraph on the difference: was it worse, slower, or yours.

The room

[Discord invite] #003-what-you-handed-over. Nobody has to say what the work was. Say what you found when you tried to explain it without the tool.

From Palmer

[Replies from issue two.] This is the one the whole letter is for. The papers are interesting, and the papers are also a way of holding the question at arm's length. The Microsoft study asked people to describe their own handovers. This week you do the same, in writing, where you can see it. Keep the wicked environment idea. Nothing about the tool will tell you when it has quietly taken a skill off you. Only a habit like this one will.

004, the forecast issue. Five public predictions about AI made in the last year by named people, with resolution dates. The background is base rates and Tetlock's finding that an hour of training in probabilistic reasoning improved forecasting for a year, and that finding the reference class was the single most useful move (evidence). Vocabulary: base rate, reference class, calibration. Write it down: pick the prediction you most want to be true and argue it is wrong.

005, the argument map. One widely shared essay about AI and work, drawn as boxes and arrows instead of read as prose. The background is argument mapping and why it has the largest measured effect of any critical thinking method (meta-analysis). Vocabulary: premise, inference, load bearing. Write it down is replaced by a photograph of the map, posted in the room.

006, the correlation. The Gerlich study: 666 people, a correlation of minus 0.68 between AI use and critical thinking (study). The background is one thing only, what a correlation can and cannot say, and why a number that big from a self report survey should slow you down rather than speed you up. Vocabulary: correlation, confounder, self report. Write it down: explain the finding to a friend twice, once so it scares them and once so it does not, using only true sentences.

007, the disagreement. A disagreement the reader had about AI in the past month, with a real person. The background is considering the opposite: a 1984 experiment found that asking people how they would judge the evidence if it pointed the other way reduced bias, and worked far better than asking them to be fair (study). Vocabulary: steelman, motivated reasoning, considering the opposite. Write it down: the other person's case, well enough that they would sign it.

008, the return. No new item. Score the issue one forecast and your own from issue three. Reread issue one with what you know now. Write it down: three lines, what changed, what did not, what would change it. Then the plain question, did this change how you think about AI, yes or no.

You read about AI most days. So does everyone in your field, and you have noticed it has made almost nobody sharper. It has made people fluent. Fluent is the tell. Somebody who has never thought a thing through sounds exactly like somebody who has, for about four minutes, and then a real question arrives and one of them has nothing.

The Companion is twenty minutes on a Monday, built for the moment after the four minutes. One item, not the whole feed. The background you need, so you are reasoning instead of reacting. Three words you can use exactly. Three questions in the same three shapes every week, until the shape is a reflex you take into meetings that have nothing to do with AI. A number you write down and get scored on, because this is the one field where everybody predicts and nobody keeps score. A paragraph in your own hand, read by a person. And a room where other people are doing the same thing, so you can see how differently a careful reader can land.

The payoff is not a certificate. On Thursday somebody says the thing everybody is saying about AI, and you know exactly which part of it is true, which part is a headline, and what would have to be shown for the rest to hold. You say that out loud in three sentences. That is what twenty minutes a week buys, and almost nobody in the room has spent it.

Two months, free. Start Monday.

A fresh read of the prototype page by STRANGER, 2026-09-06, per _context/02_AGENTS/STRANGER.md. No file access, no history with the company. Recorded here unedited except for the fixes it caused, which are listed under Gate below.

Stopped at: "Every week you get handed a new model, a new claim, a new study saying your brain is either saved or cooked," because that is the exact register of the hype the page claims to be an antidote to, and it arrives before the reader knows what the thing is. Kept reading only because the next paragraph gave a concrete shape. What STRANGER thinks the company sells: a weekly email that picks one AI story, gives enough background to argue about it, makes you answer three questions and write a paragraph back, and keeps a running score on predictions, so the product is the habit of forming your own opinion rather than the information. First eight issues free, and a price after that which has not been decided. It reads like one guy named Palmer with a Discord, which is either the charm or the risk depending on your mood.

• Confusing. The minute block read as broken text ("2Item / 4Background") until the section below it explained itself.
• Costly. Does not believe the twenty minutes. The sample issue points at a four minute FAQ inside a two minute block, then asks for three multi-sentence answers in eight minutes, then a paragraph. STRANGER's estimate: forty-five. Knowing the exact minute count of every part made the underestimate read as deliberate rather than sloppy.
• Would sign up. Yes. The line that decided it: "Nothing you write is graded by a machine, ever."
• Machine tells. Three phrases with the smoothed cadence of a model doing aphorism: "Thinking is the one thing you cannot outsource without losing it," "Both the alarm and the eye roll are lazy," and "Same tool. The only difference was whether the machine was allowed to do the thinking." The second one sits in Palmer's note, which is the one place on the page that has to sound like a person wrote it fast.

Run 2026-09-06 per _context/02_AGENTS/GATE.md. gate.sh passed on both files before this pass, so the style half is clean and this is the substance half.

Result: partial pass, four rules unsatisfied. All four are fixed above; this section is the record of what they were.

1. Rule 3, traceability. Discord described as free with no member limit, with no source and no inline marker. Fixed by removing the pricing claim from the sentence and stating that Discord's free tier and member limit were not checked in this session.
2. Rule 3, traceability. Squarespace 7.1 features (password protection, blog collections, Email Campaigns) stated in the present tense with no source, in a document whose front matter says no vendor surface was loaded. Fixed by naming them as unconfirmed and putting the check before the build starts. This is PRA-018 exactly, and the front matter disclaimer did not cover a flat present tense claim two hundred lines later.
3. Rule 3, traceability. The correlation of minus 0.68 appeared in the page's issue outline with no sample size and no link. Fixed by adding both.
4. Rule 3, internal consistency. The document claimed every section carries a minute count while the room carries none. Fixed by narrowing the claim.

Checked and clean: the other ten figures (48, 17 and 127 percent, 319, 54, 159, 216, 194, 589, 1,000 tests) each carry a link in the file where they appear. The minute counts add to twenty and match the clock. Nothing in either file claims an action already taken without Palmer's approval, so rule 8 is clean and the prototype banner says so on the page. GATE's note on voice: the closing pitch section reads closer to marketing than the rest of the record, which is defensible because it is labelled as a pitch rather than passed off as internal reasoning.

What GATE expects a reader to challenge first: the Discord and Squarespace claims, since the document admitted no vendor site was checked and then stated both companies' features as fact.

Fixed this pass: every source now has a named door and a time cost, the labelled question format is gone, six questions became three written as sentences, the room is a free Discord with one channel per issue, the twenty minutes is a stated contract with a per section clock, the vocabulary section is new, and the four GATE findings above.

Still open, and needing Palmer: - Discord. Confirm the free tier and any member limit, then create the server and one channel. Nothing has been created. - Squarespace. Confirm password protection, the blog collection and email campaigns in the account before the build. - Price after the free two months. Not set anywhere on purpose. - The twenty minutes. STRANGER estimates forty-five for the sample issue as written. The fixes above cut the required reading and tightened question three, and the only real test is Palmer sitting down with a coffee and a timer and running issue one himself. That is the acceptance test for version one.