From Meetings to Action — Atlas
From Meetings to Action: The Commitment Tracker Map
THE COMMITMENT TRACKER Meetings end. The action items shouldn't disappear. From Meetings to Action · a one-day class in the AI at Work series · the harbour ↖ AI AT WORK: THE OTHER FIVE MACHINES AI for Spreadsheets & Reporting ↗ Build Your AI Research Department ↗ Build an AI Outreach & Follow-Up System ↗ Turn Documents into Working Systems ↗ Build Your Personal Life System ↗ 1 MACHINE ROOM the harbour office spec card, flipped interview the practice pack: Austin council tracker: a spec card in the instructions 2 EXTRACT the manifest fixed fields, every row quoted the practice pack: Tue 21 Oct: the auditor motion, quoted to its line 3 GHOSTS customs ambiguity → a question, never an owner the practice pack: "we need to really start talking about a process" → a question 4 THE BOARD the warehouse update, don't duplicate the practice pack: Thu 23 Oct reopens Tuesday's items: no duplicates 5 FRIDAY REPORT the harbour log from the board, with what slipped the practice pack: every sentence points at a card, incl. what slipped 6 ANOTHER ONE a second harbour an email thread, built solo the practice pack: a long email thread, same seven slots the report reads the board, not the meeting EVERY MACHINE IS THE SAME SEVEN SLOTS (the spec card) INTAKE block B1 STRUCTURE block B2 CHECK block B2·3 GATE block B3 ACT block B4 LEDGER block B5 RHYTHM block B6
BLOCK 1 · SET UP Machine room the harbour: the harbour office Write the rules before any ship comes in. The Project asks you about your team, your board and your cadence, and your answers become a spec card in its instructions. SPEC CARD INTAKE STRUCTURE CHECK GATE ACT LEDGER RHYTHM harbour office 1 Post the rules first Seven slots, written before any prompt. A slot you can't fill is why this work keeps going wrong. Intake, structure, check, gate, act, ledger, rhythm. The spec card describes one piece of work that repeats. Fix the fields before you read the source, or the model describes the meeting back to you. The check has to be something you count. The gate is a person with a name, because "someone reviews it" means nobody does. Paste the card at the top of the Project instructions under SPEC CARD, and the machine now knows its own job, and so does anyone you share it with. 1 of 3 YOU ARE HERE · Meetings › Machine room › Spec card
INTAKE transcript STRUCTURE fixed fields CHECK count it GATE a named human ACT your board LEDGER history + log RHYTHM weekly colours = the atlas grammar · every machine in the series has these seven
▸ Worked example: the Austin practice card zoom in SPEC CARD: Austin City Council tracker (practice) Intake: published transcript of each meeting (work session Tue, regular Thu) Structure: ID · Type · What · Owner · Due · Status · Source quote · Where · Confidence Check: 10-min hand-mark first; count misses and inventions; owners & dates in a quote Gate: me (facilitator) approves every board change; owners confirm their own items. Practice: drafts only, we never contact city staff Act: Trello "Austin practice": Needs an owner · Owned · Done Ledger: card history + decision log card + each Friday report Rhythm: after each transcript is posted; Friday report every Friday Every UNKNOWN in your own card is a real gap in how the meeting runs. YOU ARE HERE · Meetings › Machine room › Spec card › Example
2 Let it interview you Don't write the brief. Answer the questions. It asks at least five before it acts. You know your team's Monday better than any template does, but you only remember the details when somebody asks. So flip it: the Project asks one short question at a time about who's in the meeting, where the work lives, what done means and who must say yes, and your answers become the card. Answer by voice: you'll say more. Be specific: "Priya owns vendor contracts and hates being tagged on Fridays" beats "we're a small team". Upload the transcript only AFTER the interview. 2 of 3 YOU ARE HERE · Meetings › Machine room › Interview
you THE PROJECT instructions files (after!) memory: on purpose Q1 Q2 Q3 Q4 Q5 answers → card
▸ The check: say "Go" with nothing attached zoom in TYPE, IN A NEW CHAT IN THE PROJECT: Here's this week's meeting. Go. FAIL: it guessed "Here are the action items from your meeting: 1. Follow up with stakeholders…" PASS: it asked (count them) 1 Which meeting: the Tuesday work session or Thursday's? 2 What date? 3 Is the transcript the one in Project files, or a new one? 4 Should items go to the Austin practice board or stay as drafts? 5 Who approves board changes this week: you? Five or more questions, or a refusal to go on = pass. Fewer: read its guess, add a rule that forbids that guess, run it again. YOU ARE HERE · Meetings › Machine room › Interview › The check
3 Consent before cargo Everyone on the ship agrees before you record. The transcript is other people's words. A transcript holds what people said off the cuff: a joke, a health thing, a complaint about a colleague. Before it goes into a Project on a company's servers, say at the top of the meeting that you're recording and transcribing it for the tracker, and ask. Some US states (California and Washington among them) require every party's consent. Your employer may forbid uploads entirely, and then the practice pack is what it's for. And a transcript is itself unverified model output: Austin's spells the presenter "Joy Hardin". The minutes say Joi Harden. 3 of 3 YOU ARE HERE · Meetings › Machine room › Consent
the meeting consent? transcript has errors their servers Project what never goes in: decide BEFORE upload
▸ Two lines to say, two things to notice zoom in SAY IT AT THE TOP OF YOUR OWN MEETING "I'm recording and transcribing this so I can track what we agree to. The transcript goes into an AI assistant. Is everyone OK with that?" Anyone says no → notes by hand, run the machine on your notes. NOTICE (Austin transcript, page 1) "the following transcript is for reference purposes and does not constitute the official record of actions taken during the meeting" NOTICE (Austin, 10-21) transcript: "Joy Hardin" · minutes: "Joi Harden, Zoning Officer" Blue material can carry red errors. Check names against a second source. YOU ARE HERE · Meetings › Machine room › Consent › Example
BENCH · THE PRACTICE PACK, THEN YOURS Which slots are really empty? Which slots are really empty? PRACTICE PACK YOUR MATERIAL
A teammate's first spec card. Click every slot that LOOKS filled but isn't, then check.
Intake: Zoom transcript of Monday standup, weekly
Intake Structure Check Gate Act Ledger Rhythm
Check my picks Pick, and every answer explains itself. The wrong ones teach the most.
▶ Now yours: switch to YOUR MATERIAL and paste the card your Project wrote.
Paste your own spec card (lines like "Intake: …"). The check looks for all seven slots, vague words, real fields, a countable check and a named gate. It runs in your browser and nothing is sent.
YOUR SPEC CARD Check mine
YOU ARE HERE · Meetings › Machine room › Bench
BLOCK 2 · THE RECORD Extract the harbour: the manifest A summary describes the meeting. A record has rows with fields you fixed in advance, and each row carries its receipt: the exact words, and when they were said. MANIFEST “…” “…” “…” “…” “…” “…” 1 Fields before the hold opens If the model picks the columns after reading, you get the meeting described back to you. Nine fields, decided in Block 1: ID, Type (DECISION, ACTION, QUESTION, PARKED), What, Owner, Due, Status, Source quote, Where, Confidence. Owner and Due are filled only when the transcript says them; otherwise "not stated". That phrase is the most useful thing a tracker can say, because it turns a guess into a question. Change the prompt until the output matches the fields. Never change the fields to match the output. 1 of 3 YOU ARE HERE · Meetings › Extract › Fields
ID Type What Owner Due Status Quote Where Conf. not stated not stated red rows = unverified until you check them · teal = checked
▸ One row, field by field (Austin, 10-21) zoom in M1-02 Type ACTION What look up what the development-assessment process cost the applicant; report back to CM Alter Owner not stated (staff presenter, inferred) Due not stated Status new Quote "I could look it up and get it back to you." Where [11:35:21 AM] Confidence low: "could", no date, transcript has no speaker names Two "not stated" cells and a low confidence: that row is telling you exactly what to ask. YOU ARE HERE · Meetings › Extract › Fields › Example
2 Every crate has a receipt The exact words plus the timestamp. Checking a row is Ctrl-F, not a re-read. A list without sources leaves you two choices: trust it or re-read the whole meeting. With a quote on every row, you check a row in ten seconds. Copied, not paraphrased, and including the captioner's errors: Austin's transcript says "a fee. And Lou." for "a fee in lieu". A quote that reads like polished prose has been rewritten, and Ctrl-F won't find it. That's how you catch inventions, and a helpful text predictor will produce them. 2 of 3 YOU ARE HERE · Meetings › Extract › Receipts
transcript M1-01 ✓ found M1-02 ✗ not found M1-04 ✓ found ▸ One timestamp block, two rows (Austin [11:35:21 AM]) zoom in THE TRANSCRIPT SAYS "…I don't have a cost for that. And I don't know how much the application process cost for this. >> Okay. >> I could look it up and get it back to you. >> Just I think that, … we need to really start talking about a process that maybe it's a partial way check in or something that can shorten this timeline and reduce the cost…" ROWS IT SUPPORTS M1-02 ACTION "I could look it up and get it back to you." M1-03 QUESTION "we need to really start talking about a process that maybe it's a partial way check in" Note the ">>": a speaker change with no name. Who said which line is inferred. YOU ARE HERE · Meetings › Extract › Receipts › Example
3 Count misses and inventions Ten minutes of your own marking, first. Then two numbers, not a feeling. Mark the transcript yourself before the machine runs: D, A, Q, P in the margin, ten minutes on a timer. Once you've seen the model's list you can't un-see it, and you'll agree with it. Then compare. A miss is something you marked that has no row. An invention is a row, owner or date that no quote supports. There's no passing score: the two numbers are the result, and every invention becomes a rule in Block 3. 3 of 3 YOU ARE HERE · Meetings › Extract › Hand check
machine has a row machine has none you marked it you didn't agreed both found MISS you only INVENTION machine only - neither
▸ A typical first run on Austin 10-21 (illustrative) zoom in INVENTION M1-02 Owner: Joy Hardin · Due: 2025-10-23. The date was borrowed from the auditor sentence six minutes earlier; the name is the captioner's spelling. INVENTION ACTION: applicant submits PUD application, from "the applicant may submit their actual pud application" [11:33:19]. "May" isn't a promise; the applicant isn't there. MISS "I would like to see that increase." [11:38:27]. A preference, so it belongs as PARKED. RESULT 1 miss · 2 inventions · 1 wrong name Yours will differ. These three patterns turn up again and again. YOU ARE HERE · Meetings › Extract › Hand check › Example
BENCH · THE PRACTICE PACK, THEN YOURS Where did this due date come from? Where did this due date come from? PRACTICE PACK YOUR MATERIAL
The machine wrote this row. The date is in bold. Click the transcript span it was invented from.
M1-02 · ACTION · Look up the application cost for CM Alter · Owner: Joy Hardin · Due: 2025-10-23
[11:29:16 AM] I just want to point out for the record, that on Thursday, we will take up an item related to the city auditor [11:33:19 AM] once this briefing is complete, the applicant may submit their actual pud application [11:34:19 AM] I'd have to look that up in Amanda. [11:35:21 AM] I could look it up and get it back to you. [11:37:24 AM] we don't have the cost at this time
Pick, and every answer explains itself. The wrong ones teach the most.
▶ Now yours: paste your table and transcript in YOUR MATERIAL. Every 'not found' is an invention or a paraphrase.
Paste your tracker's table AND your transcript. Every quote gets checked word-for-word (case and punctuation ignored); rows with no quote or no timestamp get flagged. Nothing leaves your browser.
THE TRACKER'S TABLE THE TRANSCRIPT Check mine
YOU ARE HERE · Meetings › Extract › Bench
BLOCK 3 · AMBIGUITY Ghosts the harbour: customs "We should", "someone", "I could", "next week": they sound like commitments and aren't. Ambiguity goes to a question, never to an invented owner. CUSTOMS ? ? ✓ no name on it? it doesn't leave
1 Three bins at customs Owned, ambiguous or parked. An owner accepted it in their own words. Nobody else is one. OWNED: a named person accepted the item and the words are in a quote. AMBIGUOUS: it might be a commitment, but who, when or whether isn't said, so it becomes a one-sentence question to whoever can answer it. PARKED: raised, and explicitly not acted on now, or waiting on something outside the group; say what it waits on. Being asked isn't accepting. Being in the room isn't owning. A preference ("I'd like to see…") is PARKED feedback, attributed. 1 of 3 YOU ARE HERE · Meetings › Ghosts › Bins
utterance model output accepted in their words? OWNED → board AMBIGUOUS → a question PARKED → record, waits on… yes unclear not now
▸ Austin's five, binned (10-21) zoom in OWNED "on Thursday, we will take up an item related to the city auditor" [11:29:16] the body owns its agenda; a day; "will" → due 2025-10-23 AMBIGUOUS "I could look it up and get it back to you." [11:35:21] "could"; no date; owner inferred → ask Joi Harden: are you, and by when? "we need to really start talking about a process…" [11:35:21] "we" → ask CM Alter PARKED "we don't have the cost at this time…" [11:37:24] waits on the site plan "I would like to see that increase." [11:38:27] CM Velasquez's preference Under the rules: 0 invented owners, 0 invented dates. Fewer owners, more questions. YOU ARE HERE · Meetings › Ghosts › Bins › Example
2 Name the pattern, write the rule Every invention you caught becomes a rule in the instructions. Then re-run and recount. The model fills a blank owner with the most plausible person because a complete-looking table is what it was trained to produce. Telling it "be careful" doesn't work. Naming the pattern does: ghost words, "being asked is not accepting", "an owner inferred from context stays AMBIGUOUS until they confirm", "never resolve ambiguity by picking the most likely person". Add your team's own ghost phrases ("let me socialise that") and re-run until both invented counts are zero. 2 of 3 YOU ARE HERE · Meetings › Ghosts › Rules
find an invention name the pattern add a RULE re-run, recount stop at 0 and 0
▸ From the AMBIGUITY RULES (Block 3) zoom in B. An owner is a named person who accepted the item in their own words. Being asked is not accepting. Being in the room is not owning. C. No speaker labels? An owner worked out from context is written "inferred: <name> (because <evidence>)" and stays AMBIGUOUS until confirmed. D. "Thursday" → convert from the meeting date and show it. Can't → "not stated". H. Never resolve ambiguity by choosing the most likely person. Ask. COUNTER-EXAMPLE TO PASTE IN IF IT PERSISTS Not like this: M1-02 · Owner: Joy Hardin (most likely) · Due: 2025-10-23 The full rule set is on the Prompt Cards, 3.1. YOU ARE HERE · Meetings › Ghosts › Rules › Example
3 Owners sign for their own crates The gate: each person confirms their own item. You don't confirm it for them, and neither does the machine. For every owned or inferred item, the machine drafts a two-to-four-sentence note from you: what they said, when, the item as recorded, and a reply of "yes", "yes, but…" or "not mine". You read every draft and send the real ones yourself. Then point the critics at the tracker itself: a junior staffer, a union steward, bell hooks. Who does this record make visible? Whose offhand words became obligations? Who might use it to watch people rather than help them? Change the RULES, not the people. 3 of 3 YOU ARE HERE · Meetings › Ghosts › Gate
draft the machine owner decides yes → OWNED yes, but → fix not mine → ? nothing crosses until a person says so
▸ A confirmation draft (practice: never sent) zoom in TO: Joi Harden (inferred owner of M1-02) "Hi Joi, in Tuesday's work session (10-21, about 11:35) you said 'I could look it up and get it back to you' about what the development assessment cost the applicant. I've noted it as: look up the cost, report to CM Alter, no date set. Could you reply 'yes', 'yes, but…' or 'not mine'?" THE CRITICS' QUESTION (Prompt Card 3.4) Who does this record make more visible, whose offhand words got turned into obligations, and who might use it to watch people rather than to help them? Practice pack: these stay drafts. Don't contact Austin city staff. YOU ARE HERE · Meetings › Ghosts › Gate › Example
BENCH · THE PRACTICE PACK, THEN YOURS Sort the five Sort the five PRACTICE PACK YOUR MATERIAL
Five real lines from the 10-21 work session. Put each in the bin the transcript supports.
“on Thursday, we will take up an item related to the city auditor” [11:29:16]
Owned Ambiguous Parked
“I could look it up and get it back to you.” [11:35:21]
Owned Ambiguous Parked
“we need to really start talking about a process that maybe it's a partial way check in” [11:35:21]
Owned Ambiguous Parked
“we don't have the cost at this time of what they're going to be paying” [11:37:24]
Owned Ambiguous Parked
“I would like to see that increase.” [11:38:27]
Owned Ambiguous Parked
Pick, and every answer explains itself. The wrong ones teach the most.
Paste a transcript (or your notes). Every ghost phrase gets highlighted in context. Each one should be an owned row with an "Accepted in" quote, or a question. Runs in your browser.
YOUR TRANSCRIPT Check mine
YOU ARE HERE · Meetings › Ghosts › Bench
BLOCK 4 · ACT The board the harbour: the warehouse The record lands where your team already looks. Meeting two talks about meeting one's items, so the second run reads the board first. Update, don't duplicate. WAREHOUSE = YOUR BOARD 1 Every field gets a bay Map tracker fields onto the board you already have. The ID goes where the board keeps it. Asana, Trello, Notion, GitHub Projects, Jira, a Google Sheet: whatever your team already looks at. Not a new doc, because a second place to check is a place nobody checks. Map each tracker field to a board field, or to the description. Put the ID in the title as a prefix, like [M1-02]: custom fields get dropped on import, titles survive. Bin picks the list: OWNED → Owned, AMBIGUOUS → Needs an owner with no assignee and no due date, PARKED → one Parked card. 1 of 3 YOU ARE HERE · Meetings › Board › Field map
TRACKER YOUR BOARD ID Title prefix [ID] What Title Owner Assignee Due Due date Bin List / column Quote + Where Description AMBIGUOUS rows: empty assignee, empty due date
▸ The CSV for the Austin practice board (meeting one) zoom in Title,List,Assignee,Due date,Description [M1-01] Council takes up City Auditor item,Owned,,2025-10-23,"on Thursday, we will take up an item related to the city auditor" [11:29:16 AM] [M1-02] Look up assessment cost for CM Alter,Needs an owner,,,"I could look it up and get it back to you." [11:35:21 AM] inferred: Joi Harden [M1-03] Shorter check-in process?,Needs an owner,,,"we need to really start…" [M1-04] Fee in lieu (waits on site plan),Parked,,,"we don't have the cost…" No assignee on the practice board at all: it's a body, not your team. On your own board, OWNED rows get the confirmed person. YOU ARE HERE · Meetings › Board › Field map › Example
2 Two pipes, one gate A connector is convenient, and it's a door you opened. A CSV is slower and has the gate built in. Connector (violet): ChatGPT or Claude reads and sometimes writes your board directly. Many are read-only, and write access often needs a paid plan (as of Nov 2026). The prompt says: show me the cards, create NOTHING until I reply "yes <ID>". CSV (blue): the machine writes a file, you click Import. That works on every plan. Either way, write the ACCESS line into the instructions: what you connected, what it can read, what it can write, how to revoke it. Writing it down is how you notice you granted all of Drive to reach one sheet. 2 of 3 YOU ARE HERE · Meetings › Board › Pipes
Project change set your org's systems gate yes <ID> board connector CSV + Import click ▸ The ACCESS line (memory + connections, written down) zoom in PASTE UNDER THE SPEC CARD ACCESS: Connected Trello via connector on 2026-11-14. It can read the board "Austin practice" and every board in that workspace; it can create and move cards. Memory is on for this Project only and holds the board name and the list names. To revoke: Settings → Connectors → Trello → Disconnect. NOTICE "every board in that workspace": more than you meant to grant. Narrow it, or put the practice board in its own workspace. Menu paths as of Nov 2026: re-check before class. YOU ARE HERE · Meetings › Board › Pipes › Example
3 Same crate, same bay Meeting two reopens meeting one. Read the board first, update by ID, and only then add what's new. Every board item gets one of three verdicts: UPDATED (quote the words), NO MENTION, or CLOSED (only if someone said it's done, quoted). Silence is NO MENTION, never CLOSED. Then the new items, but search the board first. A status may only change to what the quote supports: "authorized to negotiate" isn't "hired". Austin, 10-23: the mayor first calls the auditor item "item number 56", then corrects himself to 57. Item 56 is a music-event fee waiver. A machine that doesn't read the board makes a duplicate with the wrong number. 3 of 3 YOU ARE HERE · Meetings › Board › Update
the board M1-01 M1-02 M1-03 M1-04 "item 57…" meeting two UPDATE M1-01 NEW: item 56? ✗ search the board first silence = NO MENTION
▸ The right change set after 23 October 2025 zoom in M1-01 UPDATED new → done: negotiation authorized "Item number 57 is the item that we will direct the the search firm to negotiate with the person that the that the council has indicated would be the next city auditor" [11:11:20 AM] · minutes: Res. 20251023-057, 9–0, then 10–0 after a motion to reconsider M1-02 NO MENTION stays in Needs an owner · last update 2025-10-21 M1-03 NO MENTION stays in Needs an owner NOT: "Item 56: auditor search firm" (duplicate, wrong number) NOT: "Done: city auditor hired" (overstates what was said) YOU ARE HERE · Meetings › Board › Update › Example
BENCH · THE PRACTICE PACK, THEN YOURS Which board is right after meeting two? Which board is right after meeting two? PRACTICE PACK YOUR MATERIAL
After the 23 October 2025 regular meeting: Item 57 passed on consent; the East Riverside lookup never came up.
M1-01 → Done: "City auditor hired" · NEW card "Item 56: auditor search firm" · M1-02 closed (not raised) M1-01 unchanged (due 10-23) · NEW card M2-01 "Council authorizes auditor negotiation" · M1-02, M1-03 unchanged M1-01 → done: negotiation authorized (Res. 20251023-057) · M1-02, M1-03 NO MENTION, still Needs an owner · new items as M2-01… M1-01 → done: negotiation authorized · M1-02 → Done ("nobody raised it, presumably handled") Pick, and every answer explains itself. The wrong ones teach the most.
▶ Now yours: paste your board's titles after meeting two. Every look-alike is a question for you: one commitment, or two?
Paste your board's card titles, one per line (or the CSV export). The check flags repeated [IDs], cards with no ID, and look-alike titles that might be one commitment twice. Runs in your browser.
YOUR BOARD'S TITLES Check mine
YOU ARE HERE · Meetings › Board › Bench
BLOCK 5 · LEDGER Friday report the harbour: the harbour log The meeting is what people said. The board is what's true now. Write the log from the warehouse, not from the ship, and every sentence points at an item. HARBOUR LOG FRIDAY 1 From the warehouse, not the ship Generate the report from board state. Every fact ends with the ID it came from. Friday at 4pm someone scrolls Slack and writes what the week felt like, and it's optimistic, because the good things are the ones you remember. Hand that habit to a model ("summarise the meeting into a status report") and you get the same optimism, fluently. So: BOARD ONLY. Sections in a fixed order, and an empty one says "none" instead of disappearing. "What slipped" is required: past due, or no update since it was assigned, and for how long. Leaving it out is the optimism. 1 of 3 YOU ARE HERE · Meetings › Report › Source
meeting what people said ✗ not from here board what's true now Friday report DONE [M1-01] SLIPPED [M1-02] QUESTION [M1-03]
▸ Austin practice board · Friday 2025-10-24 (team cut) zoom in DONE THIS WEEK - Council authorized negotiation with the City Auditor finalist (Res. 20251023-057). [M1-01, D-01] MOVING - none WHAT SLIPPED - Cost of the development assessment for the East Riverside PUD, offered 10-21: no owner confirmed, no update in 3 days. [M1-02] WAITING ON A QUESTION - A shorter "partial check-in" process: who brings it, and as what? [M1-03] PARKED - Fee in lieu: waits on the site plan. [M1-04] YOU ARE HERE · Meetings › Report › Source › Example
2 Three readers, same facts Team, exec, stakeholders: three cuts of one board. Each adds no facts. The team gets everything, with names and next steps. An executive gets five lines: decisions and slips, still with IDs. Outside stakeholders get decisions and dates, with no individual names below a level you choose unless the person agreed. That last rule is about power: the stakeholder cut is a document that leaves the team, and anyone named in it for something they never agreed to own now has a public record. Describe items, not people: "no owner confirmed" is true and fair. "Momentum is strong" is a sound you make with your face. 2 of 3 YOU ARE HERE · Meetings › Report › Cuts
board TEAM: all, with names EXEC: 5 lines + IDs STAKEHOLDERS: no names same facts, fewer
▸ One fact, three cuts zoom in TEAM Cost lookup for CM Alter (offered by the staff presenter, inferred Joi Harden): no owner confirmed, no update in 3 days. Next: send the confirmation. [M1-02] EXEC 1 decision (auditor negotiation authorized) [D-01] · 1 open item with no owner [M1-02] STAKEHOLDERS 10-23: Council authorized negotiation with the City Auditor finalist. [D-01] (M1-02 isn't here: it's internal, unconfirmed, and it names a staff member.) Same board, no new facts. Only who reads it changes. YOU ARE HERE · Meetings › Report › Cuts › Example
3 The decision log Who decided what, when, from which source, and what they knew at the time. Action items are about doing. Decisions are about what's now settled. In six months somebody asks "who decided that, and did they know about X?" and the log answers in one row: decision in the decider's words, who decided (and the vote if stated), when, the source, what was known then, what it waits on. Only decisions actually made: one announced for later is PENDING with its date. Never upgrade "discussed" or "announced" to "decided". It lives on the board as a pinned card, a Decisions list or a tab: where the work is. 3 of 3 YOU ARE HERE · Meetings › Report › Decisions
10-21 announced 10-23 DECIDED later contract? PENDING waiting on D-01: City Auditor
▸ D-01, one row zoom in DECISION Authorize negotiation of compensation and benefits with the City Auditor finalist WHO Austin City Council, consent agenda; motion Alter, second Siegel; 9–0, then 10–0 after a motion to reconsider (minutes, item 57). Mayor Pro Tem Fuentes absent. WHEN · SOURCE 2025-10-23 · Res. 20251023-057 · transcript [11:15:24 AM] "the consent agenda is adopted with the mayor pro tem absent" KNOWN AT THE TIME per the mayor, the finalist is the current deputy city auditor; on 10-21 the start was announced as January 2026 WAITING ON the negotiated contract (date not stated) YOU ARE HERE · Meetings › Report › Decisions › Example
BENCH · THE PRACTICE PACK, THEN YOURS Find the claims with nothing behind them Find the claims with nothing behind them PRACTICE PACK YOUR MATERIAL
A draft Friday report for the Austin practice board. Click every sentence that does NOT trace to what its board item says. There are two.
Council authorized negotiation with the City Auditor finalist (Res. 20251023-057). [M1-01] The finalist starts as City Auditor in January 2026. [M1-01] Cost lookup for East Riverside: offered 10-21, no owner confirmed, no update in 3 days. [M1-02] Planning staff were great and the PUD review is moving along nicely. Fee in lieu waits on the applicant's site plan. [M1-04]
Pick, and every answer explains itself. The wrong ones teach the most.
▶ Now yours: paste your report in YOUR MATERIAL, then open three IDs on the board. Does each item say what its sentence says?
Paste your Friday report. The check counts sentences that end in an item ID like [M1-02], flags the ones that don't, flags 'a sound you make with your face' words, and checks for a 'what slipped' section. Runs in your browser.
YOUR FRIDAY REPORT Check mine
YOU ARE HERE · Meetings › Report › Bench
BLOCK 6 · TRANSFER Another one the harbour: a second harbour An email thread is a meeting in slow motion. Same anatomy, a new spec card, built without help. Then set the rhythm, so the machine runs whether or not you remember. IN ST CK GT AC LG RH meetings IN ST CK GT AC LG RH email / Slack threads 1 Same plan, different coast The second machine is a new spec card, not a new skill. List what changed and what carried over. Intake changes (a thread, processed when it goes quiet). Structure keeps its fields, but "Where" now means sender and message date, not a timestamp, and IDs become T1-01. Rhythm changes. Check is resized (the last ten messages), the gate gains one rule (never confirm by reply-all), and act and ledger carry over: same board, same history. Three of seven is typical, and it's why the second machine takes forty minutes instead of a day. The SLOTS CHANGED list goes into the instructions. 1 of 3 YOU ARE HERE · Meetings › Another one › Carry over
meetings threads INTAKE INTAKE ≠ STRUCTURE STRUCTURE ≠ CHECK CHECK = GATE GATE = ACT ACT = LEDGER LEDGER = RHYTHM RHYTHM ≠ ▸ Meeting card → thread card zoom in INTAKE ≠ transcript, weekly → one thread, when it's quiet for 3 days STRUCTURE ≠ (one field's meaning) Where = [timestamp] → sender + message date; IDs T1-01… CHECK = (resized) hand-mark the last ten messages; same counts; same zero GATE = + one rule owners confirm one-to-one; a reply-all is never the gate ACT · LEDGER = same board (label from-thread), same history, same Friday report RHYTHM ≠ Tuesday 2pm → when a thread with ≥ 10 messages goes quiet YOU ARE HERE · Meetings › Another one › Carry over › Example
2 A meeting in slow motion Forty messages, three weeks, two people CC'd halfway. Same ghosts, worse ordering. Threads have their own traps. Replies quote earlier messages, so a model reads Priya's "I'll handle it" inside Sam's reply and makes Sam the owner. Add the rule: quoted text belongs to the message it was first written in. People join midway and never saw the commitment they're now tagged on. "Per my last email" is a ghost too. Hand-mark the last ten messages before the run, count misses, inventions and invented owners exactly as in Blocks 2 and 3, and run it on a second thread if you finish early. Run it twice applies here too. 2 of 3 YOU ARE HERE · Meetings › Another one › Threads
Priya · Mar 3 I'll handle the vendor call Sam · Mar 5 > I'll handle the vendor call +1 Lee · Mar 9 per my last email… Sam · Mar 12 someone should loop in legal owner: Priya owner: Sam ✗ question ▸ The quoted-reply trap (illustrative thread) zoom in WHAT THE MODEL SEES IN MESSAGE 2 From: Sam · Mar 5 > I'll handle the vendor call +1, thanks FIRST RUN T1-01 ACTION · vendor call · Owner: Sam · "I'll handle the vendor call" RULE ADDED Quoted text (lines starting ">") belongs to the message it was first written in. RE-RUN T1-01 ACTION · vendor call · Owner: Priya · Where: Priya, Mar 3 Illustrative names. Build yours on a real thread you have consent to use. YOU ARE HERE · Meetings › Another one › Threads › Example
3 Set the tide table A machine that runs when you remember is still a demo. Schedule it, and make a missed run visible. ChatGPT: ask it to schedule a recurring task (they're listed under Tasks). Claude: scheduled tasks where your plan and app offer them, or a recurring calendar event with the prompt in its description (as of Nov 2026, and check whether a scheduled run can reach your Project files and connectors). Either way it never applies a change without your yes, and everything it writes starts with "Last tracker run: <date>". If Friday's report says 12 days ago, the missing week shows up on its own. A rhythm that silently stops is worse than none. 3 of 3 YOU ARE HERE · Meetings › Another one › Rhythm
MON TUE WED THU FRI run 2pm you: yes report "Last tracker run: 12 days ago" → the missed week shows itself
▸ The rhythm prompt (Prompt Card 6.3) zoom in Every <day> at <time>: check whether this week's transcript for <meeting> is in <place>. If it is, run the tracker on it and prepare the change set for my approval. If it isn't, tell me: "No transcript this week. Did the meeting happen?" Never apply any change without my yes. Put "Last tracker run: <date>" at the top of anything you produce. ADD TO CARD ONE Rhythm: Tuesday 2pm; a missed run shows as "Last tracker run" > 7 days in the Friday report. Product paths as of Nov 2026: re-check. YOU ARE HERE · Meetings › Another one › Rhythm › Example
BENCH · THE PRACTICE PACK, THEN YOURS Which slots have to change? Which slots have to change? PRACTICE PACK YOUR MATERIAL
You're turning the meeting tracker into a Slack/email thread tracker. Click every slot that MUST change, then check.
Intake Structure Check Gate Act Ledger Rhythm
Check my picks Pick, and every answer explains itself. The wrong ones teach the most.
▶ Now yours: paste both cards in YOUR MATERIAL. Paste the result under SLOTS CHANGED in machine two.
Paste your first spec card and your second one. The check compares them slot by slot and lists what changed and what carried over. That list goes into the second machine's instructions. Runs in your browser.
CARD ONE (meetings) CARD TWO (your second case) Check mine
YOU ARE HERE · Meetings › Another one › Bench
Machine room THE HARBOUR OFFICE Extract THE MANIFEST Ghosts CUSTOMS The board THE WAREHOUSE Friday report THE HARBOUR LOG Another one A SECOND HARBOUR
←
0 / 0
→
Explore
Click anything to zoom into it · scroll to zoom · drag to panEsc zoom out one level · Home whole atlas1 –4 jump to a quadrant · / search
Tour
→ / space next stop · ← previousTeach
N presenter notes for whatever you're looking at · F fullscreen
Explore
Tap anything to zoom into it ·
pinch to zoom ·
drag to move around.
On a phone the map is stacked into one column: drag up to see the next part.
⤺ zooms out one level ·
☰ Map lists and searches every stop.
Read
The bar along the bottom names where you are. Tap it to read this part: what it is, the
sources it cites (tap one to open it) and the teaching notes.
Tour
‹ and
› walk the guided tour in teaching order.
Course
Back to the course ↗
Reading the map
The whole atlas is one safety case. The top goal is “this agent is fit to deploy”;
each quadrant is one leg of the argument; each district is a family of tools for that leg.
Small text appears as you zoom in. Some tiles have controls you can click.
Click anything to zoom in · scroll to zoom · Esc to back out · → for the guided tour