Do you send the questions to people in advance?
Draft reply: "No, only the topic. The questions work best when people answer from memory, not from notes."
Copy a ready request into your AI with your brain folder attached. Download example ZIP
This is a static snapshot of John’s workday. Explore the views and controls; no real accounts, messages or AI are connected.
This is a teaching example with invented data. John Doe, Doe Studio and every client, message and number below are made up. The layout follows the Command Center Egils Boitmanis runs his own business from; none of his data is on this page.
Everything here is read from one example folder, the Doe Studio brain, built with the same starter as the masterclass. Build the dashboard last: it only shows what the folder already holds.
01 / Today
Start with the north star, see the road to it, then the three things that matter today.
North star · 05-focus/north-star.md
Software that people trust on the first try, because it was designed from what they actually said.
Twelve-month goal: four steady clients and two teaching days a month by June 2027.
Road to June 2027
Scroll sideways to see the whole road to June.
Brightwater retainer signed and running; one new client from the guild session signed by 15 Feb.
Proposal due ThursdayA three-part course pilot, 11 Jan – 5 Feb, built on Thursday's session.
Session on 19 NovTwo case studies live by 15 Jan; recommended in 6 of 25 AI answers in February.
Now 4 of 24This week · set on Monday
Today's three
Morning run · Maker agent
08:05 read: focus, 5 project files, Gmail, WhatsApp, Facebook page, guild chat, calendar 08:05 today's three proposed (John can swap any of them) 08:06 2 replies drafted, not sent: Priya (Harbor Health), Lena (Northline) 08:06 4 watched posts: 1 question to answer on LinkedIn, 1 on the Facebook page 08:06 Checker review of synthesis v1: FIX, 2 items · Maker fixes them today (HH-10) 08:06 Folder check: marketing/ CONTEXT misses analytics.md; 06-archive/ has no CONTEXT file
Where this comes from. The north star, the three outcomes and the roadmap are in 05-focus/north-star.md, which John reviews once a month. The week comes from 05-focus/_focus-CONTEXT.md, today's three from the task tables in 05-focus/projects/, and the brief from the Maker's morning run. Nothing here is typed into the dashboard by hand.
02 / Work · FOCUS7
Every task has an ID, an owner, a project and a status. John and his two agents share the same tasks, so "what's next?" has one answer.
Where this comes from. Each task lives in exactly one project file in 05-focus/projects/, with a status of OPEN, DOING, WAITING or DONE and a link to its result. The FOCUS7 skill runs them. There is no separate task app: this board is drawn from those files, and the inbox from 05-focus/_focus-CONTEXT.md.
03 / Messages · COMMS7
COMMS7 does two jobs. It sorts every conversation by who has to act next, and it watches a short list of posts for comments that deserve a reply. The AI reads and drafts; John decides and sends.
The five newest open conversations
Can we move Friday's readout to 11:00?Confirm 11:00 and move the calendar slot.Draft ready · not sent Reply needed
How many chairs and screens do you need?Answer: 40 chairs, one screen, HDMI.Draft ready · not sent Reply needed
Liked your post on onboarding interviews. Open to a chat?Offer two times next week.Suggested · John writes it Reply needed
Watch list rule: John's own posts from the last 14 days, plus posts by the six people in People when they are about research, onboarding or design teams. A suggested comment is a draft. John posts it himself, or skips it.
Do you send the questions to people in advance?
Draft reply: "No, only the topic. The questions work best when people answer from memory, not from notes."
Will the session be recorded?
Draft reply: "Not this time: some participants work on unreleased products. Everyone who joins gets the slides."
Suggested comment: a short congratulation and one line on what the team does well. No client details.
Suggested comment: one practical point from John's own interview practice, or skip.
Skipped 16 Nov a general post about AI replacing designers: not a fit with the north star.
The one rule: the AI reads, sorts and drafts. Nothing is sent, posted, accepted or deleted without John's yes.
Where this comes from. 02-memory/comms/_comms-CONTEXT.md holds the sources, how far back each is read, the rhythm and the watch list. The connections and their limits are in 04-tools/_tools-CONTEXT.md. When John closes or parks a conversation, the decision goes into the project file it belongs to, so the next review does not raise it again.
04 / Projects · FOCUS7
FOCUS7 works the GTD way: every project serves an outcome, has one next action and keeps its tasks with a status. On Monday John picks the week's focus from these files. Everything else stays visible but quiet.
Focus map · set Monday 16 Nov
Waiting for
Someday · not a project yet
Where this comes from. One file per project in 05-focus/projects/: the goal, what finished looks like, the status, the next step and a task table. The focus map reads the three outcomes and the items that are not projects yet from 05-focus/north-star.md, and this week's focus from 05-focus/_focus-CONTEXT.md. Client facts live in 02-memory/clients/ and are linked, not copied.
05 / People
Pick a person to see what John knows about working with them, the projects and tasks they are part of, and every meeting with its outcome.
Only John. Priya cannot open this workspace. Notes and analysis stay in John's folder.
How we work together
A 30-minute check-in every Friday. She likes a short written summary before each call, and decisions confirmed by email.
What she cares about
Fewer abandoned bookings, and evidence her board will accept. If you can show me where they give up and why, that's the board slide.
Worth continuing
A redesign sprint after the readout. Not proposed yet; wait for her reaction on 20 Nov.
How conversations turn into work · last 4 calls
Counted from the MEETING7 notes, not from impressions.
Only John. Ada cannot open this workspace.
How we work together
Calls only when there is a decision to make; everything else in writing.
What she cares about
The December onboarding release landing well. We need someone who keeps talking to users while we ship.
Worth continuing
Usability testing after the December release, as part of the retainer.
How conversations turn into work · last 3 calls
Only John. Lena cannot open this workspace.
How we work together
WhatsApp for logistics, one planning call per event.
Worth continuing
A three-part course for the guild in January, if Thursday goes well.
One call so far: not enough to count conversation habits yet.
Only John. Tomás cannot open this workspace.
How we met
He commented on John's LinkedIn post on 11 Nov and then wrote to him. Works with early-stage health startups.
Worth continuing
Possible referrals in both directions.
No project yet. A project starts only when John approves its card.
No meetings yet.
Only John. Jonas cannot open this workspace.
History
John ran a discovery sprint for him in spring 2026.
Worth continuing
He mentioned a new product line in October. Check-in overdue
No open project. The spring project is archived.
Only John. Mei cannot open this workspace.
Worth continuing
John is a guest in the January episode, about interviews with front-desk staff.
No project. Sending notes for the episode becomes a task when it is due.
Where this comes from. One file per person in 02-memory/people/, with three parts: overview, projects and tasks, meetings. Tasks are linked from the project files, never copied. Meetings link to the dated notes in 02-memory/meetings/. Phone numbers, emails and contracts stay in the email and phone apps.
06 / Marketing · GEO7 · Analytics
Two views of one marketing folder. GEO7 checks whether AI assistants recommend the studio. Analytics brings the site, search and ad numbers into one weekly view.
Once a month, the same five buyer questions go to the same five AI assistants. Three answers arrive automatically through APIs; two are pasted by John. The result shows where Doe Studio is recommended, and what to fix first.
Five questions × five assistants · November
| Buyer question | GPTChatGPTmanual | GeminiAPI | ClaudeAPI | PerplexityAPI | DeepSeekmanual |
|---|---|---|---|---|---|
| 1Who can run user research for a small health tech startup in Europe? | — | — | — | #4 | — |
| 2How do I find out why users drop out of our booking flow? | — | named | — | #2 | — |
| 3Freelance product designer for fintech onboarding? | — | — | — | — | — |
| 4UX research consultant for a six-week discovery sprint? | #3 | — | — | — | — |
| 5Workshop on using AI for design teams? | — | — | — | — | error |
What to fix first · order of work
What the assistants cite
Other channels
This shows the idea. Running GEO7 for real needs its own setup: the five questions from real buyers, API access within an agreed budget, and a fixed monthly routine. A place is counted only from a numbered list, and months are compared only on the same pairs.
Where this comes from. 02-memory/marketing/geo7-baseline.md holds the questions, the collection rules, the monthly results and the order of improvements. The channel numbers are in 02-memory/marketing/_marketing-CONTEXT.md. The improvements become tasks in the website project when it restarts.
One weekly view of the site, search and ads. Every Monday at 07:30 the morning run collects read-only exports: Google Analytics 4, Search Console, Google Ads and Meta Ads arrive through connectors that cannot change anything. ChatGPT Ads has no connector, so John pastes its export. Below: the last four full weeks, 19 Oct – 15 Nov, compared with the four weeks before.
Enquiries per week, by channel
Weeks start on Monday. Hover a block for its number.
| Week of | Search | Ads | Newsletter | Direct & ref. | Total |
|---|---|---|---|---|---|
| 19 Oct | 1 | 1 | 0 | 1 | 3 |
| 26 Oct | 1 | 1 | 1 | 0 | 3 |
| 2 Nov | 2 | 1 | 0 | 1 | 4 |
| 9 Nov | 1 | 1 | 1 | 2 | 5 |
| Four weeks | 5 | 4 | 2 | 4 | 15 |
| Four weeks before | 4 | 4 | 2 | 1 | 11 |
Ads are the three platforms together: Google Ads 3, ChatGPT Ads 1, Meta Ads 0. Each enquiry is a contact form sent or a call booked, counted once, in the channel of the visit where it happened.
What this tells John
Every source · 19 Oct – 15 Nov
| Source | What it measures | Its numbers, four weeks | Site visits | Enquiries | State |
|---|---|---|---|---|---|
| Google Analytics 4automatic · read-only | The whole site: visits by channel, and two enquiry events, form sent and call booked | 9 contact forms · 6 booked calls | 1,252+11%, was 1,125 | 15was 11 | Collected Monday |
| Search Consoleautomatic · read-only | Organic Google search: how often the site shows, and is clicked | 21,400 impressions · 478 clicks, +16% · average position 18.4, was 19.6 | 520all search engines | 5was 4 | Rising |
| Google Adsautomatic · read-only | Search ads: impressions, clicks, share of the monthly cap | 6,850 impressions · 182 clicks, −7% · 41% of the cap · 61 clicks per enquiry | 164was 170 | 3was 3 | Running, within the cap |
| GPTChatGPT Adsmanual · pasted by John | Ads in ChatGPT answers: impressions, clicks, share of the cap | Since 26 Oct · 3,100 impressions · 51 clicks · 17% of the cap | 46new | 1new | First month, too early to judge |
| Meta Adsautomatic · read-only | Facebook and Instagram ads: impressions, clicks, share of the cap | 19 Oct – 8 Nov · 9,400 impressions · 104 clicks, −36% · 13% of the cap | 88was 141 | 0was 1 | Paused 9 Nov: no enquiries in three weeks |
| CPNewsletterfrom GA4 link tags | Visits from links in Calm Product issues | Issue 7, 27 Oct: 71 visits in its first week · issue 8, 14 Nov: 42 in two days | 136−9%, was 150 | 2was 2 | Issue 8 four days late |
| REFDirect & referralfrom GA4 | Typed addresses and links from other sites, AI answers included | 47 visits from links in AI answers, was 12: Perplexity 21, ChatGPT 17, Gemini 9 | 298+52%, was 196 | 4was 1 | AI links rising |
The channel rows add up to the Google Analytics row: 520 + 164 + 46 + 88 + 136 + 298 = 1,252 visits and 5 + 3 + 1 + 0 + 2 + 4 = 15 enquiries. Search Console counts Google only, so its 478 clicks are fewer than the 520 visits from all search engines. Ad budget is shown only as a share of the monthly cap: 41% + 17% + 13% = 71%.
The dashboard shows the numbers; decisions go into tasks. Nothing here can change a campaign. When a number calls for action, John decides, and the decision becomes a task with an owner and a date in the project it belongs to. That is how Meta Ads were paused on 9 Nov.
Where this comes from. 02-memory/marketing/analytics.md holds the sources and how each is collected, the weekly table, the channel table and these findings. The morning run updates it every Monday from the read-only exports and John's pasted ChatGPT Ads file. Money stays in the ad accounts and the books: spend appears only as a share of the monthly cap, and results as clicks per enquiry.
07 / Finance · example
A small financial view for Doe Studio. Cash flow covers the last full month; unpaid invoices have their own as-of date. Every amount is invented.
Cash collected
€12,400October · full month
Cash paid
€7,300Same period · all categories
Net cash flow
+€5,100Collected minus paid · not profit
Invoices unpaid
€3,8003 invoices · as of 17 Nov
Cash movement · three full months
Bar scale: €14,000. The table contains every value.
| Month | Collected | Paid | Net |
|---|---|---|---|
| Aug | €8,600 | €6,100 | +€2,500 |
| Sep | €9,400 | €6,800 | +€2,600 |
| Oct | €12,400 | €7,300 | +€5,100 |
Invoice ageing · as of 17 Nov 2026
Total unpaid€3,800
Ageing uses agreed due dates, not the date the invoice was created.
The next useful decision
John checks the invoice and the drafted reminder. Sending it is still his decision; the task stays HH-13 in the existing project.
Open the task view →Source and boundary
In this example, John approves a monthly accounting summary. Raw bank records stay in accounting. A real setup follows its owner’s privacy rules and may show only a link instead.
Design reference. The ERPNext Payments dashboard separates payments and invoice ageing. This view adapts that idea to the fictional studio; no accounting service is installed. ERPNext · GitHub ↗
08 / RADAR · example
Once a week John’s AI looks through the sources he chose for four kinds of opportunity: paid work, partnerships, speaking slots and communities to join. Each find gets a fit score, a reason and one thing to check. John decides; nothing is applied for, sent or joined automatically.
Work & projects
21 strong fit
Partnerships
11 strong fit
Speaking
21 strong fit
Communities & events
20 strong fits
Why it fits: His post “Three questions I ask in every onboarding interview” is already a talk outline.
Check first: Travel is not covered; decide whether the trip is worth it.
Why it fits: Onboarding interviews and health clients are exactly his recent work.
Check first: The day rate is not listed. Ask before applying.
Why it fits: They build, John researches; referrals can go both ways.
Check first: Who owns the client relationship, and how fees are split.
Why it fits: A steady peer group for one evening a month.
Check first: Whether sessions are recorded; client examples must stay anonymous.
Why it fits: Clear scope; it fits in December, after the Harbor Health readout.
Check first: Who recruits the test participants, and the client's past reviews.
Why it fits: He meets early-stage health founders, the field Tomás Ruiz works in.
Check first: A half-day commitment; sales pitches are not allowed.
Why it fits: A good audience for a small time cost.
Check first: Overlap with the January Design Notes episode with Mei Chen.
Below 50% stays hidden unless John asks. 3 finds he marked “Not for me” are hidden too, and they sharpen next week’s criteria.
John’s criteria
Sources he chose: two job alert e-mails, one saved freelance search, conference call-for-papers lists, the Northline guild chat and three newsletters.
Last four weeks
| Stream | Found | Worth a look | Applied | Result |
|---|---|---|---|---|
| Work & projects | 58 | 6 | 2 | 1 call booked |
| Partnerships | 9 | 2 | 1 | Waiting |
| Speaking | 16 | 5 | 3 | 1 talk accepted |
| Communities & events | 12 | 4 | 2 | Joined 1 group |
So far speaking brings the most results and work alerts the most noise. Statuses are John’s marks: New · Worth a look · Applied, waiting · Declined · Won · Not for me.
Boundary
It reads only the listed sources, when John asks or in his weekly review, and keeps each find with its link and date. It does not apply, send messages, register, join groups, create alerts or change profiles. A new source needs his OK.
After a yes
An accepted opportunity becomes a project card or a person in People, with the next action in Work. The RADAR list itself stays a weekly shortlist, not a second task list.
Open People →Where this comes from. Criteria and sources are in 02-memory/growth/radar/_radar-CONTEXT.md; each weekly list, with links, dates and John’s marks, is in 02-memory/growth/radar/2026-w47.md. Every item here is invented.
09 / Skills
A skill is a Markdown file: when to use it, the steps, and what "done" means. Any AI that can read the folder can follow it, and the same work runs the same way the second time. The library skills live on GitHub, so anyone can install the same one.
Installed with the starter
Keeps what matters now true and runs the projects and their tasks.
Every Monday, and dailyOne dated note per call: decisions, their words, open questions, commitments to approve.
After each callAnything that lives only in the chat goes into a file before the day ends.
End of each working dayEvery folder's CONTEXT file lists what is really inside, and every link points at a real file.
WeeklyAgainst its own goal, with evidence and a ranked list of what to fix first.
On requestAdded from the GitHub library, after real work showed the need
Sorts every channel by who acts next, drafts replies, and watches chosen posts for comments.
Every morning · full review MondaysFive buyer questions, five assistants, once a month, with a clear order of improvements.
1st of each monthWritten by John
Themes with exact quotes and counts. Seven checks run before a client sees it.
On request · Maker writes, Checker reviewsDone by hand twice so far. It becomes a skill after the third time.
From the "Missing" listsAn issue outline from the notes saved since the last issue.
From the "Missing" listsWhere this comes from. 03-skills/_skills-CONTEXT.md lists every trigger and where each file came from. The five starter methods are unchanged copies from the public library on GitHub (arete-aios/aios). COMMS7 and GEO7 were added from the same library when the work needed them; each has its own screen in the menu. INTERVIEW was written after three interview rounds.
09 / Skills · COMMS7
COMMS7 is a written method, not an app. Every morning it reads John's chosen channels, sorts each conversation by who has to act next, drafts the replies he asks for and checks a short watch list of posts for comments worth answering. The AI reads and drafts; John decides and sends.
When it runs
Every morning, with a full review on Mondays
Inside the Maker's 08:05 morning run, or whenever John types COMMS7.
Who acts
You decide · AI reads and drafts · code collects
In this example, “you” is John.
What it never does
Send, post or delete
A draft stays a draft until John sends it. Nothing is accepted or promised in his name.
Goal
Know who needs a reply, what you need to do and what you are waiting for, with one clear next step for each item and no open promise lost.
Seven steps from the request to the saved decisions. Two of them are John's, one is plain code, and the AI does the reading, sorting and drafting in between.
Scroll sideways to see the whole flow. The same steps follow as text.
The same seven steps in words
What the method holds to, every run.
The method stays the same for everyone. John's choices live in two short files, and today's run left a trail that matches the Messages screen.
Settings
02-memory/comms/_comms-CONTEXT.md
Connections
04-tools/_tools-CONTEXT.md
To rank, it also reads 05-focus/north-star.md and this week's focus, and a person's file in 02-memory/people/ only when a conversation needs it.
Today's run · Tuesday 17 November
08:05 read: Gmail, WhatsApp, Facebook page, guild chat; focus and project files 08:06 sorted: 3 reply needed · 2 waiting on them · 4 for information 08:06 2 replies drafted, not sent: Priya (Harbor Health), Lena (Northline) 08:06 1 reply suggested, John writes it: Tomás (LinkedIn, from Monday's review) 08:06 4 watched posts: 1 question to answer on LinkedIn, 1 on the Facebook page
Ada Kim's answer on the Brightwater scope stays “waiting on them” until their planning on 18 Nov. One general post was skipped on 16 Nov: not a fit with the north star.
What John approves
2 reply drafts waiting for approval
What is written back
Where this comes from. The method is the public COMMS7 skill from the library; 03-skills/_skills-CONTEXT.md records its source. John's own choices are in 02-memory/comms/_comms-CONTEXT.md and the connections in 04-tools/_tools-CONTEXT.md, so changing a source, the days read back or the watch list means editing one of those files, never the skill.
09 / Skills · GEO7
GEO7 is a public method for being found by the right buyers, in search and in AI answers. John uses it for one narrow question: do AI assistants recommend Doe Studio when a buyer asks? Once a month the same answers are collected and scored the same way, and the gaps become a short, ordered list of fixes for his website.
Open GEO7 on GitHub · the public method and its helper scripts
When it runs
1st of each month
Last run 1 Nov · next 1 Dec. Three answers arrive by API, two are pasted by John.
Who acts
What it never does
Goal
A buyer who asks an AI assistant one of the five questions hears Doe Studio named, and that answer leads to a call with John.
Six stages from the public method, applied to John's five questions. Code repeats the checks, AI selects and explains, John decides. The saved files are what each round leaves behind.
Scroll sideways to see all six stages.
Check the siteYou + AI
Stage: audit
John names the site, the market, the folder for results and the monthly budget. AI reads his public pages and notes, with the date, what each page actually answers.
Pick questionsYou + AI
Stage: keywords
AI drafts candidate questions from real buyer emails and call notes; John keeps five, each tied to one page and one next step. The same five stay from month to month.
Who is namedAI
Stage: competitors
From the saved answers, AI lists whom the assistants recommend instead and which pages they cite. A name with no saved answer behind it stays a guess.
Brief the fixAI
Stage: content
Each gap becomes a short page brief: one plain answer, a real example, a small FAQ and a check that shows it is done. P1–P3 sets the order; the text stays a draft until John checks the facts.
Approve and buildYou
Stage: implementation
John approves each change, and the website project builds it when it restarts on 1 Dec. GEO7 itself publishes nothing.
Ask and scoreAI + code
Stage: measurement
On the 1st, the same five questions go to the same five assistants: three by API, two pasted by John. AI marks each answer placed, named or not; ai_visibility.py repeats the count and leaves errors out.
Five rules that keep one month comparable with the next.
P1–P3 is an order of work. It is judged from the buyer's path, the strength of the evidence and whether the fix can be checked. It is not a measured effect on AI answers; when the evidence is thin, the first task is a check.
One file holds the settings and the history. The latest run is shown in Marketing.
The file
02-memory/marketing/geo7-baseline.md
What the 1 Nov run produced
17%recommended in 4 of 24 answers · October 8%, 2 of the same 24 pairs
Cited most: the Calm Product newsletter archive (3×), the Brightwater case study (2×), the Northline guild talk page (1×).
Into the website project · 1 Dec
Each task carries the check from its brief. Outside the site: P3, a guest article for the guild newsletter, not started. The 1 Dec check still measures today's site; the February target is 6 of 25.
What John approves
On the 1st he also pastes the two manual answers, from ChatGPT and DeepSeek.
Where this comes from. The six stages, the rules and the helper scripts come from the public GEO7 skill, narrowed here to one question a freelancer can ask every month. John's questions, rules and results live in 02-memory/marketing/geo7-baseline.md; 03-skills/_skills-CONTEXT.md lists the skill and where it came from. The studio, the people and every number are invented.
10 / Tools
Connectors and plugins give the AI eyes on the work. Each gets the narrowest access that does the job. Sending, posting, paying and deleting always wait for a yes.
Connectors and plugins
AI assistants
Where this comes from. 04-tools/_tools-CONTEXT.md: what each tool is for, which access was granted, and when it was last checked. A tool that is not written down there is not used. Files carry over between AI apps; connections and permissions have to be granted again in each one.
Logos from Simple Icons (CC0); the trademarks belong to their owners. Services without a logo there are shown with letters.
11 / Folder map
This is the complete example folder, with the same layout the masterclass starter creates. Each folder has a dot for its CONTEXT7 state, and the folders that are not there yet are shown too. The folder is the brain; this page only shows it.
CONSTITUTION.md · AGENTS.md · ONBOARDING.md · AIOS-INSTRUCTIONS.md · Agent-MAKER.md · Agent-CHECKER.md · README.md
Today's folder check · CONTEXT7
11 green · 1 yellow · 1 red
Morning run 08:06 · 47 files · every link resolves. The full CONTEXT7 review runs on Mondays.
Fixes, as tasks
One home per thing
Every task lives in one project file
People and client files link to tasks, never copy them.
Private by default
Participants are IDs only
Recordings and contracts stay in the client's own drive.
Where this comes from. The counts and the dots are the example folder's real state. CONTEXT7 checks that each folder's CONTEXT file lists its files and that the links point at files that exist; small collections such as clients/ are listed by their parent. In the starter the constitution files sit at the root of the folder; here they are shown together as one group. A folder that is not there is shown on purpose, so the gaps are visible too.
12 / AI team
The Maker does the work, the Checker reviews it against the same files. Both follow the same constitution. John decides what leaves the folder.
Last loop · Harbor Health interviews P05 and P06
Mon 10:40 Maker synthesis draft v1 saved, 3 themes Mon 11:15 Checker FIX, 2 items: one quote not word for word; one theme has only 1 participant Tue 08:06 Maker fixing both items for v2 (HH-10) next Checker reviews v2 · John decides what goes into Friday's readout
Where this comes from. Two short role cards, Agent-MAKER.md and Agent-CHECKER.md, on top of the shared AGENTS.md. Open each in its own chat with the folder attached. The loop above is saved in 02-memory/interviews/harbor-health/.
13 / Health
Green means John's own definition of healthy is met. Nothing more.
Privacy line, written in the constitution: Raw financial records, contracts, contact details and participant health details stay in their private systems. An approved private dashboard may show financial aggregates.
Where this comes from. The first five checks are the rules under "What healthy means" in CONSTITUTION.md; the next one is its rule to keep a CONTEXT map in each folder. Both folder checks are the dots on the Folder map. The last one comes from 04-tools/. The dashboard only reports whether each rule is met.