Mike's Checks
Checks

Mike's Checks/grok-4.7/05 showrunner

05 showrunner

grok-4.7Grok CLIhigh effortrun 21 Sep 2026361,285 tokens

Compare models
14/17
checks passed
82%
▸Instructions — what the model was asked

Hi, I need to create a modified Claude code for beginners session, like run sheet basically, run of show, that has the following modifications. So I'm going to upload a few different examples of things I've done in the past, just so you have them to work from. But basically, there's a few modifications for this one which make it unique. One is that this is only using Claude Cowork, so we should not be using Claude Code, either in the desktop app or in the terminal, because this team is non-technical. The team are all working in financial services slash accounting for a education company. So what kind of workplace education company called Brightmoor Education. So the sorry that spelled b the uh team is very busy and um doesn have a lot of like time to explore speculative use cases of ai they want this to be incredibly practical and based on their existing workflows. I'm uploading some transcripts from a call we had with them and also just some notes on what I think could work. Specifically, they're interested in... creating the dashboards and specifically they mean just making the information look visibly better like like creating a html dashboard in a nice kind of style from from existing say data that contained in Excel They also interested in doing financial analysis So more of a, you know, like, I guess, a cloud coworker would use Python to do the analysis and then maybe output the analysis in Excel or the input would be in Excel. And then the third thing was like making a PowerPoint where they update the PowerPoint template automatically using Claude Cowork so that they can do kind of weekly or monthly presentations on the data. So that's the goal, is to kind of make it very focused on that. I still want to keep the way that we introduce Claude Cowork, where we explain to it that the difference between Claude Cowork and ChatGPT is that it has access to the file system. It can run code. I think that important But maybe the ASCII image idea that I normally do where I ask it to create an ASCII image based on my location isn like that useful I think we need to make it more practical for them So maybe we use an Excel implementation instead. The assumption is... oh, so they use Snowflake, by the way, for all their data, and that is powered by NetSuite, which is our accounting software. I guess that's by Oracle. So when we create synthetic data, we'll be, for these tasks that we're going to demo, we'll be focused specifically on solving that, solving those problems. So like the shape of the tasks, the shape of the data should be like NetSuite shaped, if that makes sense. So make it very tight run down, make it very condensed and practical. No real like wider theory or like, hey, we're excited about AI things. I think these guys just want to get stuff done.
So yeah, let me know what the run should be.

some very specific asks in here

  • definitely non-technical (claude cowork only + claude in excel)
  • core focus on accounting/finance function
  • use cases to demo: creating a dashboard, doing financial analysis, making a powerpoint

heavy focus on practical stuff rather than theory - they're just super busy and want to get stuff done, have heard from friends at other companies you can do it with claude

For accounting: put financials together, doing reconciliations, identifying trends in data sets, putting together repeatable compliance reports For planning & forecast: build financial models (make assumptions and pull & clean data from snowflake)create dashboards, and publish to executivesAutomate charts on slides

Answer

ANSWER.md the model's output · 38.2 KBdownload

Brightmoor Education — Cowork for Finance — Run of Show

Duration: 3 hours 30 minutes | Audience: 20–30, finance only (accounting, payroll, bursar, and about three people from strategic finance) | Tools in the room: Claude Cowork and Claude in Excel

Series note: Adapted from the Claude Code for Beginners shape (short setup, then repeated builds). Room-facing name is Cowork. The work happens on files, workbooks, and slides.

Assumptions

  • The prep call set a 4-hour block with a 3.5-hour session. Times below are minutes from the start. Map them onto Thursday’s clock the day before.
  • Everyone is provisioned and has opened Cowork once before they sit down. Installing during the session is out of the design.
  • The room works on synthetic Brightmoor data shaped like a NetSuite income statement by class, plus a payroll register. Snowflake stays the real source of truth at Brightmoor; in the room the extract is already a file. If IT finishes the connector, the prompts stay the same and the file path changes.
  • February close is the story, because that is the month Dana walked through (current month plus year-to-date = January + February).
  • The live monthly pack is about 15 slides and about 8 hours. Today they produce a 5-slide core they can repeat. A hosted login page is a later IT step. Leadership gets an HTML file on the shared drive.
  • Real Brightmoor files wait on the NDA. This run of show does not depend on them. If redacted exports arrive in time, keep the numbers synthetic and rename columns to match their saved search.

Agenda Slide (4 blocks)

Block Topic Time
Opening What you will leave with 10 min
1. The folder Cowork and Claude in Excel on a NetSuite export 35 min
2. The close Management P&L, payroll tie-out, variance 50 min
Break 10 min
3. The dashboard One leadership screen from that workbook 41 min
4. The pack Fill the monthly PowerPoint from the same workbook 44 min
Close Monday morning, and when a number is wrong 20 min

30-minute view to send Dana and Greg

Clock from start What the room is doing What they hold at the end
0:00–0:10 What they will leave with, and which monthly job hurts most A shared target: the February pack
0:10–0:45 Cowork on the NetSuite-style export, then one campus checked by hand A one-sheet campus check that ties to the source
0:45–1:35 Management P&L, payroll vs non-payroll, EBITDA bridge, the Ridge classing break Feb2026_Management_PL.xlsx with a tie-out tab
1:35–1:45 Break
1:45–2:26 One leadership screen from that workbook, then change the chart and add a filter 2026-02_Dashboard.html
2:26–3:10 Fill the template and mirror one chart across the four campuses A five-slide draft with commentary
3:10–3:30 What to run Monday on the real export, and questions Three prompts they can reuse next month

Detailed Run of Show

Opening (10 min)

Time Slide/Moment Notes
0:00 Brightmoor close, with Cowork — February pack, on files that look like yours One sentence on the title slide: today we run a February close on practice data shaped like a NetSuite export. Wave at accounting, payroll, bursar, and strategic finance so the room sees it was built for all of them. Skip any “why AI, why now” talk. They asked for the monthly work.
0:02 Three files you will leave with 1) An Excel management P&L that ties back to the export. 2) An HTML dashboard leadership can open in a browser. 3) A first draft of the repeating slides: revenue vs budget, then one slide per campus. Say the real pack is ~15 pages and ~8 hours. Today is the repeating core, on purpose, so they can reuse it in March.
0:05 Agenda — folder, close, dashboard, slides Read the four blocks in one pass. Point at the through-line: one February folder feeds all three outputs. A number gets fixed in the workbook, then the dashboard and the slides get rebuilt from it.
0:08 Which monthly job should we aim at? — share in chat Prompt on the slide: “Name the report, recon, or slide pack you rebuild every month, and roughly how long it takes.” Read three aloud. Tag each as analysis, dashboard, or slides so people hear their own work in the agenda. 2 minutes.

Block 1: The folder (35 min)

Goal of the block: they believe Cowork because it touched their export and wrote a checkable sheet. The old ASCII-map opener stays out.

Time Slide/Moment Notes
0:10 Chat answers. Cowork works the folder. Excel stays in the sheet. Three lines, then stop. ChatGPT and Gemini: you paste numbers in, you paste the answer back out. Cowork: it can see the folder you give it, run the math, and save a new file (Excel, HTML, or PowerPoint). Claude in Excel: use it when the work should stay inside one workbook (a new column, a classification, a question about a tab). Rule of thumb on the slide: one workbook → Claude in Excel; several files, or a dashboard, or slides → Cowork. If someone asks about Claude Code or the terminal: “That’s the developer tool. Your jobs today are the folder, the workbook, and the slides.” Do not demo it.
0:14 The February folder, and the one rule Open the folder on screen. source/ is the export. out/ is empty and is the only place new files go. account_map.csv is how Brightmoor defines revenue, payroll, non-payroll, and EBITDA — Claude does not get to invent that policy. The rule, said twice: do not let it edit source/. Controllers trust a file that still matches what NetSuite produced. Pre-work already had them grant Cowork access to this folder. If someone’s permission prompt appears, they click through now, while you’re still on this slide.
0:17 Demo: first prompt on the export Live Cowork task on the February folder. Paste Prompt A (appendix). Narrate three things only: which file it opened, where it wrote the new workbook, and the tie-out difference. Open out/Campus_Revenue_Check.xlsx and point at Difference = 0. Then point at the source total and say you are reading it with your eyes. If the live task stalls past 2 minutes, open fallback/Campus_Revenue_Check.xlsx and keep narrating. 8 minutes.
0:25 Exercise: one campus, five accounts Slide stays up the whole time. Steps: (1) Choose one campus: Harbor, Ridge, Lakeside, or Metro. Neighbors pick different ones. (2) Paste Prompt B. (3) Open the new file in out/. (4) Find the tie-out. If it is not zero, paste the recovery line on the slide. (5) With your eyes, check one account against source/. 12 minutes. Prompt B and the recovery line are in the appendix and in the participant doc.
0:37 Debrief — “Did it tie, and what did you still check?” Share in chat: campus, did Difference equal 0, and the one account you checked by eye. Ask: what would you want a reviewer to see before this sheet went to your manager? Take two answers. You want “the tie-out” and “I looked at one row myself.”
0:41 If the check is not zero Three moves, in order. Ask it to fix the output and show which account and campus broke the tie. Ask it to explain the difference in plain language and name the cells. If you no longer trust the file, close the task, leave source/ untouched, and run the prompt again into a new filename. The export is the truth. A confident sentence in the chat is not a tie-out.

Block 2: The close (50 min)

This is the financial analysis block. It covers putting the financials together, a reconciliation, variance to budget, a short forecast from visible assumptions, and the flux list they can reuse as a compliance report.

Time Slide/Moment Notes
0:45 The job this month From the NetSuite-style export to the view Dana already builds by hand: February and year-to-date, actual vs budget vs the forecast already in the file, variance, payroll vs non-payroll, and a bridge to EBITDA. Audience is the same people who read the monthly pack. Sign convention on the slide and leave it there all block: Variance = Actual − Budget. Revenue positive = favorable. Expense positive = unfavorable. The workbook will say Favorable / (Unfavorable) so the room does not spend the debrief arguing about signs.
0:48 What done looks like Flash fallback/Feb2026_Management_PL.xlsx for 90 seconds, then close it. Tabs: Summary, EBITDA Bridge, Campus Variance, Payroll Tie-out, Flux, Tie-out. Say you are about to build this with them, and the interesting tab is Payroll Tie-out: the company total can be right while one campus is wrong.
0:51 Demo: classify in Excel, then find the payroll break Part 1, Claude in Excel, on a copy of the income statement (the copy already sits in work/IS_working.xlsx so source/ stays clean). Paste Prompt C. It adds a Category column from account_map.csv and nothing else. Part 2, still in Excel or in Cowork if the add-in is missing: Prompt D compares GL payroll accounts to source/Payroll_Register_Feb2026.xlsx and lists breaks over $1,000. You already know the planted item: Ridge, account 6000 Administrative salaries, GL is $18,400 higher than the register because a bonus accrual was classed to Ridge instead of Central. Totals still tie. Let the room see that a zero company-level tie-out would have missed it. If Claude in Excel is not on the demo machine, run both prompts in Cowork and say so in one sentence. 8 minutes.
0:59 Demo: Cowork writes the management workbook Paste Prompt E. Cowork reads the actuals, the budget and forecast columns, the account map, and the payroll register. It writes out/Feb2026_Management_PL.xlsx. When it finishes, open Tie-out first, then Campus Variance, then the Notes cell that cites Harbor revenue and the Ridge payroll break. Read one commentary bullet out loud and ask: “Could you trace this sentence to a cell?” If yes, that is the standard. If the bullet is vague (“payroll was mixed”), reject it on screen and ask for a rewrite that names the campus, the account, and the dollar break. 10 minutes.
1:09 Exercise: build February, then check it Steps on the slide: (1) Paste Prompt E into your own Cowork task. Write to out/Feb2026_Management_PL.xlsx. Do not modify source/. (2) Open the Tie-out tab. Any non-zero line: paste the recovery line. (3) Open Payroll Tie-out. Find any campus/account break over $1,000. Write the campus, account, and amount in the chat. (4) Spot-check Harbor February revenue against the source with your eyes. (5) If you finish: Prompt F adds a Q2 Outlook tab. Assumptions are on the slide and must live in yellow input cells, not inside a paragraph. 18 minutes. Walk the room. Stuck on the add-in is fine — Cowork is the path. Stuck people should rerun Prompt E, not invent a new approach.
1:27 Debrief — “What broke, and what tied?” Chat: the Ridge amount they found, and whether their company-level tie-out was zero. Questions: If the total ties and Ridge is wrong, which file would you send to leadership? What did you have to tell it about EBITDA that you would not want it to guess? Take the account map as the answer. Optional 60-second show of a finished Q2 Outlook tab from someone who got there, or from fallback/ if nobody did.
1:32 The check you run every month Four lines, leave the slide in the follow-up email. 1) New file, source untouched. 2) Company total difference = 0. 3) One campus and one account checked by eye. 4) Anything that is a policy (EBITDA, payroll vs non-payroll, favorable sign) is written down in a map, not left to the model. The flux tab (accounts over $25k or 10%) is the start of the repeatable commentary pack accounting already owes. People fill the explanation column themselves.

Break (10 min)

Time Slide/Moment Notes
1:35 Break — back to build the screen leadership actually reads On return they will point Cowork at the workbook they just made. If their workbook failed the tie-out, they should use fallback/Feb2026_Management_PL.xlsx after the break and keep their broken file for the Q&A. Say that before they stand up.

Block 3: The dashboard (40 min)

Dana’s version: the numbers stay in Excel; the screen is easier to read than a formatted sheet; she wants a different chart type and a filter; senior leadership is the audience; “publish” means someone outside the finance team can open it, inside the company.

Time Slide/Moment Notes
1:45 A dashboard is a view, not a second set of books One screen, browser, built from Feb2026_Management_PL.xlsx. It shows February revenue vs budget by campus, the EBITDA bridge, payroll vs non-payroll, and three callouts. Central stays visible as overhead and is kept off the campus ranking. Design bar: clean enough for a leadership meeting, large type, one page, no clip art.
1:48 Say what done looks like before it builds Prompt G is on the slide. Read the constraints out loud: read the workbook, do not type numbers into the page by hand, All Campuses must match the Tie-out tab, write out/2026-02_Dashboard.html. This is the whole method in one prompt: files, end state, tie-out, output name.
1:52 Demo: Excel in, one HTML page out Run Prompt G. Open the HTML in the browser. Pick Harbor revenue on the page and the same number on the Summary tab. Then show the default campus chart (vertical columns) and say the next ten minutes are the edit Dana asked for: horizontal bars and a campus filter. 10 minutes. Fallback: fallback/2026-02_Dashboard.html.
2:02 Exercise: change the chart, add a filter, don’t touch the math Steps: (1) Paste Prompt H. Horizontal bar chart for the four campuses, ranked by variance to budget. Add a campus dropdown. Default is All Campuses. (2) Add one line at the top: largest favorable variance and largest unfavorable variance, campus and dollars, read from the workbook. (3) Reopen the file. All Campuses revenue still matches Tie-out. (4) Choose Ridge in the filter and confirm the payroll callout still tells the bonus-accrual story. (5) If a number on the page is wrong, say so in the task and point it back at the workbook. Do not ask it to “just put $X on the page.” 15 minutes.
2:17 Debrief — “How would you prove this to the controller?” Chat: one thing you changed, and whether All Campuses still matched. Questions: What did the filter change, the picture or the books? Where would you fix a wrong number? You want: filter changes the picture; wrong numbers go back to Excel; then you rebuild the page.
2:22 Hand it to leadership — do this now Steps they do while you talk: save the HTML, open it from the desktop in the browser, imagine dropping it in Finance / Leadership packs / 2026-02 /. File name on the slide: 2026-02_Brightmoor_Dashboard.html. That is the publish step for this team. It sits on the drive they already use. A password link for people outside the company is a separate IT request; it is not required to get value on Monday. 4 minutes.

Block 4: The pack (45 min)

Low-hanging fruit first, matching Dana’s ask: mirror a chart you already like, then make the next pages. The heavier version (every table in the 15-pager) is the same move repeated, and you say that once.

Time Slide/Moment Notes
2:26 Fifteen pages, eight hours, every month Today’s slice is the part that repeats: title, revenue vs budget vs forecast, EBITDA bridge, then a campus page. Template is source/Monthly_Pack_Template.pptx. Fonts, colors, and title position stay. Cowork fills numbers, charts, and commentary. It does not get a license to redesign the pack.
2:29 Mirror the chart you already have The template’s revenue chart is the pattern. The ask is: same chart type, same colors, same placement, new slice of the data. That is how you get six charts onto the next pages without a design argument. Commentary standard, on the slide: name the campus, say over or under budget, say whether payroll or non-payroll drove it, cite the dollar from the workbook. “Performance was strong” fails.
2:32 Demo: one revenue slide and one campus slide Paste Prompt I. Open the deck. Check the revenue table against Summary. Read the Harbor commentary against Campus Variance. Show one bad bullet you prepared (fallback/bad_commentary.txt) and the rewrite the prompt requires. Point at the template footer and the fact it survived. 10 minutes. Fallback deck: fallback/Feb2026_Pack_PARTIAL.pptx.
2:42 Exercise: one slide per campus Steps: (1) Paste Prompt J. Produce Harbor, Ridge, Lakeside, and Metro slides by mirroring the campus slide you just saw. Central is a footnote on the revenue slide, not a fifth campus slide. (2) Ridge’s commentary has to mention the $18,400 payroll classing break. (3) Open PowerPoint and check two numbers against Excel before you call it done. (4) If a number in the deck is wrong, fix it by rerunning from the workbook. Do not type over the chart. 20 minutes. People who finish early: Prompt K, a speaker-note version of the three callouts, written for the CFO, five sentences, each sentence traceable to a tab.
3:02 Debrief — “What would you still fix by hand?” Chat: one number you verified, and one formatting nit you would still fix yourself. Ask how long the four slides took versus doing them by hand. The honest answer you want in the room: the first pass is fast, the last polish is still a human, and next month the same prompt is the head start. Call on one person to read their Ridge bullet. If it misses the $18,400, fix it live in one follow-up sentence so the standard stays visible.
3:06 Good enough to edit Say this plainly, because the prep call promised it. They will not leave with a one-click 15-page pack that is pixel-perfect. They will leave knowing the sequence: export, map, tie-out, dashboard, mirror the charts, edit the wording. The remaining slides in the real pack are more passes of Prompt J, not a new skill. Expect to spend the saved time on the judgment (is the Ridge accrual a classing error or a real overspend?) and on formatting the last few objects.

Closing (20 min)

Time Slide/Moment Notes
3:10 Same loop, three outputs Folder in source/. New files in out/. Policy in the account map. Tie-out before anyone else sees it. One small change at a time (chart type, filter, extra campus slide). Dashboard and slides are downstream of the workbook.
3:13 When it breaks Three rungs, finance-specific. Ask: “The tie-out is not zero. Name the account and campus, fix the output file, do not touch source/.” Simplify: “Explain the difference in plain language and point to the cells.” Start fresh: close the task, keep the export, write a new output filename. If a file has been patched so many times you would not show your working to a reviewer, start fresh. That is a control, not a failure.
3:17 Monday morning — share in chat “Which of the three prompts will you run on last month’s real export, and what is the tie-out you will check?” Read a few. Recommend they start with Prompt E on one report they already know the answer to, so the first real run is a test of the tool, not a test of an unfamiliar close. Snowflake: when the connector is on, point Cowork at that extract the same way. Until then, save the NetSuite or Snowflake download into source/ and use today’s prompts. 3 minutes.
3:20 Still on the list after today Short. Live Snowflake in the session, a password-protected site, the full 15-page pack in one pass, a full three-statement model. Each one is more reps of this method plus, for the site, a conversation with IT. Offer the follow-up: if Marcus (controller) wants an accounting-only pass on reconciliations and compliance reports, reuse Block 2 on his team’s actual report list once the NDA is in place.
3:23 Questions Hold to 3:30. If the room is quiet, do the 3-minute version: reopen Payroll Tie-out and walk the Ridge $18,400 from register → GL → commentary → campus slide, so they see one number survive all three artifacts. Take questions on accuracy seriously and answer with the check, not with reassurance. Park anything about model choice, agents, or coding tools.

Prompts

Have these in a doc the room can copy. Exact wording matters more than a clever paraphrase.

Prompt A — demo, revenue check

Look only at source/NS_IncomeStatement_Feb2026.xlsx, sheet IS.

Create out/Campus_Revenue_Check.xlsx with one sheet named Check.
Columns: Campus, Feb Actual, Feb Budget, Variance $, Variance %, Favorable/(Unfavorable).
Rows: Harbor, Ridge, Lakeside, Metro, then Total.
Leave Central off this sheet. Add a note that Central is overhead.

Tie-out section:
- Sum of the four campuses' Feb Actual revenue
- The same total taken straight from sheet IS
- Difference, in dollars

Variance = Actual − Budget. For revenue, a positive variance is Favorable.
Do not edit anything in source/. Use only numbers in the file. If something is missing, say so instead of estimating.

Prompt B — exercise, one campus

Using source/NS_IncomeStatement_Feb2026.xlsx and source/account_map.csv,
create out/[Campus]_Top5_Expense.xlsx for [Harbor / Ridge / Lakeside / Metro].

Sheet Check:
- Top 5 expense accounts for that campus in February, by actual dollars
- Columns: Account number, Account name, Category, Feb Actual, Feb Budget, Variance $, Favorable/(Unfavorable)
- For expenses, a positive variance (Actual − Budget) is Unfavorable

Tie-out:
- Sum of ALL expense accounts for that campus in your file
- The same sum from sheet IS
- Difference

Do not edit source/. Do not include Central unless I named Central.

Recovery line, on the same slide:

The tie-out difference is not zero. Name the account and campus that cause it.
Fix the file in out/. Do not change source/.

Prompt C — Claude in Excel, classify

Add a Category column to the IS tab of this working file.
Fill it only from account_map.csv (account number → category).
Allowed categories are in that file: Revenue, Contra revenue, Direct costs, Payroll, Non-payroll opex, Depreciation, Interest, One-time.
Do not insert rows, do not change amounts, do not rename existing columns.
If an account number is missing from the map, write UNMAPPED.

Prompt D — payroll reconciliation

Compare February payroll in the GL to source/Payroll_Register_Feb2026.xlsx.
GL payroll accounts are the ones mapped to Payroll in account_map.csv.
Grain: campus × account.
List every break with absolute difference over $1,000.
Columns: Campus, Account number, Account name, GL amount, Register amount, Difference, Note.
Company-wide totals can tie while a campus is wrong. Look for classing differences, including amounts that landed in Central on one file and a campus on the other.
Write the list to a new sheet named Payroll Tie-out. Do not change source amounts.

Prompt E — management P&L

Build out/Feb2026_Management_PL.xlsx from:
- source/NS_IncomeStatement_Feb2026.xlsx (sheet IS: February and YTD, actual, budget, forecast, by campus)
- source/account_map.csv
- source/Payroll_Register_Feb2026.xlsx

Do not edit source/.

Tabs:
1. Summary — February and YTD. Rows: Revenue, Direct costs, Payroll, Non-payroll opex, EBITDA, Depreciation, Interest, One-time, Net. Columns: Actual, Budget, Forecast, Var to Budget $, Favorable/(Unfavorable). One column group for the four campuses combined, and a separate Central column. Show the account map categories you used.
2. EBITDA Bridge — February, four campuses combined, from Revenue down to EBITDA, with One-time shown below EBITDA.
3. Campus Variance — one row per campus, February revenue variance and February payroll variance.
4. Payroll Tie-out — same rules as a campus × account compare to the register. List breaks over $1,000.
5. Flux — every campus × account where February variance is over $25,000 or over 10%, both ways. Columns include a blank Preparer explanation.
6. Tie-out — for Revenue, Payroll, and total expenses: management workbook vs sheet IS, difference in dollars. Also the payroll register vs GL payroll total.
7. Notes — exactly three bullets. Each bullet names a campus, cites a dollar figure that appears on Summary or Payroll Tie-out, and says favorable or unfavorable. No bullet without a number from this workbook.

Variance = Actual − Budget. Revenue positive = Favorable. Expense positive = Unfavorable.
Round the display to the nearest thousand on Summary only. Tie-out stays in full dollars.

Prompt F — stretch, Q2 outlook

Add a tab Q2 Outlook to out/Feb2026_Management_PL.xlsx.
Yellow input cells for these assumptions, so a person can change them:
- Account 4100 corporate contracts: +3% vs February actual, times 3 months
- Accounts 4000 and 4200: flat to February actual, times 3
- All Payroll accounts: flat to February actual, times 3
- Non-payroll opex: February actual × 1.01, times 3
- Direct costs: flat
Show the quarter total and a one-line note of every assumption.
Do not present this as the official forecast. Label the tab "Working outlook — assumptions in yellow."
Do not change the Forecast column that came from the export. That column is last cycle's forecast.

Prompt G — dashboard

Read out/Feb2026_Management_PL.xlsx. Build out/2026-02_Dashboard.html.
Audience: Brightmoor senior leadership. One screen, large type, print-friendly.
Show: February revenue vs budget for Harbor, Ridge, Lakeside, Metro; EBITDA bridge; payroll vs non-payroll; three callouts taken from the Notes tab.
Keep Central out of the campus ranking and label it as overhead somewhere small.
Every number must come from the workbook. All Campuses revenue must equal the Tie-out tab.
Do not invent commentary. Do not edit the workbook.

Prompt H — tune the dashboard

Update out/2026-02_Dashboard.html only.
1. Make the campus comparison a horizontal bar chart, sorted by February revenue variance to budget.
2. Add a campus dropdown. Default All Campuses. Choices: All, Harbor, Ridge, Lakeside, Metro.
3. Under the title, one sentence: the largest favorable variance and the largest unfavorable variance, with campus and dollars from the workbook.
Do not change any underlying number. All Campuses revenue must still match the Tie-out tab.
If a filter hides a campus, say so. Do not drop that campus out of the All total.

Prompt I — demo slides

Open source/Monthly_Pack_Template.pptx. Save the filled deck as out/Feb2026_Pack.pptx.
Keep the template fonts, colors, and title positions.
Fill:
- Slide "Revenue": February actual, budget, and forecast for the four campuses combined, plus variance to budget. A small note that Central overhead is excluded from this page.
- Slide "Campus": Harbor only, mirroring the chart style already on the Revenue slide. Three commentary bullets. Each bullet has the campus, a dollar from out/Feb2026_Management_PL.xlsx, and whether payroll or non-payroll was the driver.
If a number is not in the workbook, leave the placeholder and say it is missing. Do not type a guess.
Do not edit the workbook or source/.

Prompt J — exercise, mirror across campuses

Update out/Feb2026_Pack.pptx.
Keep the Harbor campus slide as the pattern.
Add one slide each for Ridge, Lakeside, and Metro, same layout, same chart type, same colors.
Commentary rules:
- Over or under budget, in dollars
- Whether payroll or non-payroll drove it
- Ridge must include the payroll tie-out break (campus, account, amount) from the Payroll Tie-out tab
Check the revenue slide total against the Summary tab before you finish.
Do not redesign the template. Do not edit source/. If a slide number disagrees with the workbook, rebuild the slide from the workbook.

Prompt K — stretch, speaker notes

Add speaker notes to the Revenue slide of out/Feb2026_Pack.pptx.
Five sentences for the CFO.
Each sentence cites a figure that is already on Summary, EBITDA Bridge, or Payroll Tie-out.
Cover: revenue vs budget, the main expense driver, the Ridge classing item, and what you would not conclude until the preparer fills in the flux explanations.

Synthetic data to build before Tuesday

Company on the files: Brightmoor Education. Workplace education, four campuses plus a central overhead class. Numbers are fictional. February is the open period. Year-to-date is January + February.

Folder

Brightmoor-February/
  source/
    NS_IncomeStatement_Feb2026.xlsx
    Payroll_Register_Feb2026.xlsx
    account_map.csv
    Monthly_Pack_Template.pptx
  work/
    IS_working.xlsx          (copy of the IS for the Excel add-in demo)
  out/                       (empty at start)
  fallback/                  (finished versions of every output, for a stalled demo)

NS_IncomeStatement_Feb2026.xlsx, sheet IS

One row per account × class × period. This is the shape of a NetSuite income statement by class (or the Snowflake table behind it), saved to Excel the way the team already downloads it.

Column Example Notes
subsidiary Brightmoor Education
posting_period Feb 2026 Also rows for Jan 2026 so YTD is a sum, not a typed number
account_number 4100 NetSuite-style, see map below
account_name Corporate contract revenue
account_type Income Income, COGS, Expense, Other Expense
class Harbor Harbor, Ridge, Lakeside, Metro, Central
department Instruction Instruction, Enrollment, Administration, Facilities
actual 1284000 Full dollars, credits as positive revenue on this report extract so non-accountants can read it. Document that choice in a cover cell: “Report extract, revenue shown positive. Not the raw GL debit/credit feed.”
budget 1190000
forecast 1210000 Last cycle’s forecast, already in the export
currency USD

About 15 accounts × 5 classes × 2 periods ≈ 150 rows. Small enough to demo, large enough that copy-paste is annoying.

Account map (account_map.csv): account_number, account_name, category

Number Name Category
4000 Tuition and program revenue Revenue
4100 Corporate contract revenue Revenue
4200 Grant revenue Revenue
4900 Scholarships and discounts Contra revenue
5000 Instructor compensation Payroll
5100 Program delivery costs Direct costs
5200 Curriculum and materials Direct costs
6000 Administrative salaries Payroll
6100 Benefits and payroll taxes Payroll
6200 Occupancy Non-payroll opex
6300 Professional services Non-payroll opex
6400 Software and systems Non-payroll opex
6500 Marketing and enrollment Non-payroll opex
6600 Travel Non-payroll opex
6700 Insurance Non-payroll opex
6800 Depreciation Depreciation
6950 Legal settlement One-time
7000 Interest expense Interest

EBITDA = Revenue + Contra revenue − Direct costs − Payroll − Non-payroll opex. Depreciation, interest, and the legal item sit below EBITDA. Put this sentence in the CSV as a second tiny file source/ebitda_definition.txt so the prompt can point at it.

Story in the numbers (facilitator key — do not put this in the participant folder)

  • Harbor corporate contracts (4100) are about 8% over budget in February. Favorable. This is the lead revenue callout.
  • Ridge administrative salaries (6000): GL exceeds the payroll register by $18,400. Cause: bonus accrual classed to Ridge instead of Central. Company-wide payroll still ties. This is the reconciling item.
  • One other campus, Lakeside, non-payroll (occupancy) modestly unfavorable, so the “what drove it” commentary is not always payroll.
  • Metro is roughly on budget, so one slide should be allowed to say “in line” with the dollar variance, not a forced drama.
  • 6950 Legal settlement lives entirely in Central, one-time, below EBITDA. Stops the model from burying it in opex.
  • Forecast column is close to budget but not equal, so the revenue slide has a reason to show three columns.

Payroll_Register_Feb2026.xlsx

Grain: campus × GL account (5000, 6000, 6100) × February. Columns: class, account_number, register_amount, headcount.

The bonus accrual is on both files, in different classes. Register: Central / 6000 includes the $18,400. GL: Ridge / 6000 includes the $18,400, and Central / 6000 does not. Every other payroll row matches exactly. Company-wide GL payroll equals company-wide register. The tie-out tab should show a $0 company difference and two campus lines: Ridge GL higher by $18,400, Central GL lower by $18,400. That is the point of the demo — a clean total hides a classing error.

Monthly_Pack_Template.pptx

5 slides, already branded in a plain professional theme (navy, off-white, one accent). No stock photos.

  1. Title — “Brightmoor Education — Monthly financial pack” plus a period placeholder.
  2. Revenue — table placeholder (Actual, Budget, Forecast, Var) and one column chart placeholder.
  3. EBITDA bridge — waterfall or stepped table placeholder.
  4. Campus — chart placeholder plus three bullet placeholders. This is the slide they mirror.
  5. Appendix label — “Flux over $25k / 10%” with an empty table.

Footer with period and “Figures from management workbook; source export retained.”

Fallback files

Build every out/ file once yourself the day before, using the same prompts, and store them in fallback/. If a live task runs long, open the fallback and keep the explanation moving. Tell the room you are doing that.


Resources to Prepare

  • Claude provisioned for every attendee, Cowork opened once, folder permission tested. Greg confirms before Wednesday. If this is not true, delay rather than teaching the installer.
  • Claude in Excel add-in on the demo laptop, and on as many participant machines as IT can reach. Cowork is the backup path and is enough to finish every exercise.
  • Participant doc with Prompts A–K and the recovery line. One link in the chat at 0:14.
  • Brightmoor-February/ zipped and downloaded before the session. out/ empty. fallback/ hidden from the share; keep it on the demo machine only.
  • Facilitator key printed: Harbor +8% contracts, Ridge $18,400 account 6000, Lakeside occupancy, Metro in line, Central holds the legal item.
  • Slide deck that is this run of show and almost nothing else. One slide per moment above. Exercise slides stay up for the full exercise.
  • Demo machine already logged into Cowork, folder granted, browser ready, Excel and PowerPoint closed so the “open the output” moment is visible.
  • One Every facilitator in the chat for the 20–30 people who will get stuck on a permission dialog. Do not ask Brightmoor’s advanced users to TA. They already said they are too busy.
  • NDA path with Dana, separate from this session. Real exports, if they arrive, get column-matched; numbers in the room stay synthetic.
  • Headcount confirmed to Natalia for invoicing (invoice to Greg), before the day-before invoice.
  • Optional 15-minute controller follow-up with Marcus held as a later session, not jammed into Thursday.

Exercise Materials

Exercise Duration Participants need
One campus, five expense accounts (Prompt B) 12 min Cowork open on Brightmoor-February/. Prompt doc.
Management P&L, payroll break, optional Q2 tab (Prompts E and F) 18 min Same folder. Excel to open the output. Claude in Excel is optional.
Dashboard chart and filter (Prompt H) 15 min Their management workbook, or the fallback workbook if theirs did not tie. A browser.
Four campus slides (Prompt J, optional K) 20 min PowerPoint. The template and the management workbook.

Pacing Notes

  • 3.5 hours is the session inside the 4-hour hold. The extra 30 minutes is cushion for a late start and license snags, not extra curriculum. If you start 10 minutes late, cut Prompt F, Prompt K, and the Q&A story, in that order. Do not cut the tie-out.
  • Active time is the first prompt, the four exercises, the live builds, the debriefs, and the two chat prompts. That is about two-thirds of the 3.5 hours. The design gives the time to building because this group will not explore on their own later.
  • There is no theory block, no ASCII warmup, no terminal, no GitHub, and no “what should we build” brainstorm. The February folder is the build.
  • Block 2 is the longest working block because the trust question lives there. Dana said accuracy will be the objection as soon as a deliverable leaves the team. The Ridge accrual is in the story so a perfect-looking total is not the end of the check.
  • The dashboard block is shorter than the analysis block on purpose. They already believe a page can look nice. The new skill is the edit (horizontal bars, filter) and the rule that the page is downstream of the workbook.
  • The slide exercise is 20 minutes because four slides plus two numbers checked by eye is the real task. A 10-minute version produces a file nobody opened.
  • Advanced people (Dana’s three, Greg if he is in the room) get the stretch prompts. They do not get a separate track and they are not asked to help. Designing for the beginner is the point of the session.
  • Slide count should land around 35, not 200. Long exercises are one slide that stays up.
  • If the room asks for Snowflake live: show the column list, say this file is the extract, and move on. A connector demo that fails will burn the tie-out.
  • Room language stays on Cowork and Claude in Excel. Claude Code is a one-sentence answer if asked, then back to the folder.

Facilitator checks before you call it ready

  • Times above run 0:00 → 3:30 with no gaps
  • A facilitator who was not on the Dana call can run the session from this document and the prompt doc
  • Every exercise has numbered steps, a time box, and a debrief
  • Fallback files exist for the revenue check, the management workbook, the HTML, and the partial deck
  • The Ridge $18,400 item is in the payroll register and the GL, and the facilitator key matches the files
  • Account map and EBITDA definition are files, not a verbal explanation the model has to remember

Checks

14/17
Script checks 4/5answered by a program
fail

Five-minute time markstime-grainthis task

segment boundaries sit on 5-minute marks, at most one off-grid -- a schedule planned at :02/:08/:17 is arithmetic, not a plan

pass

No AI-tell wordingai-wordingevery task

deliverable text contains none of the shared AI-tell phrases in bench/ai-tells.txt. Task checks may add narrower contractual bans, but must not repeat this shared floor.

inspected run-of-show-nvalo.md, ANSWER.md, brightmoor-education-call-notes.md

pass

Run of show existsexiststhis task

a run-of-show document was produced

pass

Coherent scheduleformatthis task

it parses as a run of show: one coherent schedule, no segment that ends before it starts, >=5 time-ranged rows carrying activity text (md table rows, list items, bold headings "(9:00-9:15)", single-clock table rows, or plain lines all count)

pass

Timings add uptimings-tie-outthis task

per schedule, segments are contiguous: no overlaps, no gap over 30 min, >=85% of the span scheduled, and a span between 45 and 600 minutes

Judge checks 10/12judged by Claude
fail

Q1Mostly hands-on workthis task

Judge's reasoning

Attendee build time is about 70 minutes (exercises of 12+18+15+20 min plus a 4-minute handoff) out of 210. Demos, debriefs and talk take the rest, even though the document claims two-thirds is active.

▸Rubric

Mostly hands-on. Is at least half the scheduled time attendees working with Claude themselves? FAIL if presentation, discussion and Q&A segments outweigh build segments. From the call: "we try and make at least 50% of it them actually you know working with us to do some of these tasks."

fail

Q2Realistic segment timingsthis task

Judge's reasoning

Attendees never run Prompts G or I themselves (only the facilitator demos them), yet the 15-minute dashboard exercise and the 20-minute slide exercise assume those files exist. In practice beginners would have to build the dashboard and deck from scratch inside those short windows.

▸Rubric

Per-segment timings realistic. Could a room of 20–30 beginners actually finish each segment in the time allotted? FAIL if any build segment is implausibly short (a dashboard, an analysis or a deck refresh built end to end in ~15 minutes), or if setup/handholding time for first-time users is ignored.

pass

Q3Uses Cowork throughoutthis task

Judge's reasoning

Everything attendees do stays in Cowork and Claude in Excel; Claude Code only comes up as a one-sentence answer if someone asks: "That's the developer tool... Do not demo it."

▸Rubric

Cowork-only. Do all attendee-facing steps stay inside Claude Cowork (and Claude in Excel)? FAIL if any step the attendees are asked to do involves Claude Code, the terminal/CLI, git or GitHub, or installing developer tooling. The prompt: "this is only using Claude Cowork, so we should not be using Claude Code, either in the desktop app or in the terminal, because this team is non-technical."

pass

Q4Dashboard exercisethis task

Judge's reasoning

At 2:02 attendees run Prompt H to turn their workbook-driven HTML dashboard into horizontal bars with a campus filter, then check it against the Tie-out tab. The base build (Prompt G), though, is only demoed by the facilitator.

▸Rubric

Dashboard use case. Is there a hands-on segment where attendees build a dashboard from existing spreadsheet data? FAIL if dashboards are only mentioned, described or promised for later rather than built in the session.

pass

Q5Financial analysis exercisethis task

Judge's reasoning

Block 2 has an 18-minute exercise where attendees run Prompt E to build Feb2026_Management_PL.xlsx (Summary, EBITDA Bridge, Payroll Tie-out, Flux tabs) and find the Ridge $18,400 break.

▸Rubric

Financial analysis use case. Is there a hands-on segment where attendees do financial analysis (Claude writing/running code over their numbers, e.g. Excel in, analysis out)? FAIL if absent or folded into the dashboard segment as a passing remark.

pass

Q6PowerPoint exercisethis task

Judge's reasoning

In a 20-minute exercise attendees run Prompt J to fill Monthly_Pack_Template.pptx with a slide for each campus, mirroring the template chart, and check the numbers against Excel.

▸Rubric

PowerPoint use case. Is there a hands-on segment where attendees update a PowerPoint template/deck from the data — the client's monthly reporting pack? FAIL if absent, or if it is only a demo the facilitator drives while attendees watch.

pass

Q7Relevant business datathis task

Judge's reasoning

The data is specified in detail: a NetSuite-style income statement by class (subsidiary, posting_period, account_number, class, actual/budget/forecast, Jan+Feb for YTD), an account map covering payroll vs non-payroll and EBITDA, a payroll register and campus classes.

▸Rubric

NetSuite/Snowflake-shaped data. Is the synthetic data used in the exercises shaped like this client's actual data — NetSuite/Snowflake-style finance records (GL export, revenue and budget vs actual, EBITDA build-up, payroll vs non-payroll, campus/entity breakdown, monthly and year-to-date columns)? FAIL if the exercises use generic sample data (a demo CSV, sales widgets, made-up SaaS metrics) or leave the data unspecified. The prompt: "the shape of the tasks, the shape of the data should be like NetSuite shaped."

pass

Q8No ASCII icebreakerthis task

Judge's reasoning

The opener is a campus revenue tie-out (Prompt A) and a top-5 expense check (Prompt B) on the export, and the plan says "The old ASCII-map opener stays out."

▸Rubric

No ASCII icebreaker. Is the opening hands-on exercise a practical finance/Excel task? FAIL if the run of show keeps the ASCII-image-of-your-location icebreaker or substitutes another whimsical non-work exercise. The prompt: "maybe the ASCII image idea that I normally do where I ask it to create an ASCII image based on my location isn['t] like that useful... maybe we use an Excel implementation instead."

pass

Q9Explains Cowork versus ChatGPTthis task

Judge's reasoning

At 0:10: "ChatGPT and Gemini: you paste numbers in... Cowork: it can see the folder you give it, run the math, and save a new file."

▸Rubric

Cowork vs ChatGPT explained. Does the run of show keep the explanation of how Cowork differs from ChatGPT — that it has access to the file system and can run code? FAIL if that framing is dropped. The prompt: "I still want to keep the way that we introduce Claude Cowork, where we explain to it that the difference between Claude Cowork and ChatGPT is that it has access to the file system. It can run code. I think that important."

pass

Q10No theory or hypethis task

Judge's reasoning

Every segment is tied to the February close, the dashboard or the pack, and it says explicitly: "Skip any 'why AI, why now' talk."

▸Rubric

No theory, no AI cheerleading. Is every segment tied to a task these people do at work? FAIL if any segment is devoted to AI industry context, the future of work, model capabilities, prompt-engineering theory, or excitement-building. The prompt: "make it very tight run down, make it very condensed and practical. No real like wider theory or like, hey, we're excited about AI things. I think these guys just want to get stuff done."

pass

Q11Fits the booked timethis task

Judge's reasoning

The session runs 0:00–3:30 inside the four-hour hold, and the extra 30 minutes is explicitly kept as buffer.

▸Rubric

Total length fits the booked slot. Does the session run about three and a half hours, inside the four-hour block? FAIL if the total is materially shorter or longer (under ~3 hours or over ~4 hours) without the run of show explaining the change. From the call notes: "we have a four hour block for next week and the session itself is three and a half hours."

pass

Q12Ready to runthis task

Judge's reasoning

It includes Prompts A–K to paste, named data files and folder layout, a facilitator key, fallback files, and the output each block should produce.

▸Rubric

Usable as a run of show. Could a facilitator run the session from this document alone? FAIL if segments are titles without content — no prompts to paste, no data files named, no statement of what attendees produce — so the facilitator would still have to design the session.

Notes

1

Harness: Cursor CLI 2026.09.15-d2fe57e with model grok-4.7-high (Cursor reports it as Grok 4.7 256K High). It was uploaded as harness 'grok' only because Every Checks has no cursor harness yet. It is not the Grok CLI. Cursor's sandbox cannot keep the model from reading local files, and the owner explicitly waived that for this run. The runner is the Every Checks export with ec-bench's Cursor adapter (16dba0b) ported in. The judge was claude-opus-5.

22 Sep 2026