Confidential — For Investigators Only

The Bureau of Unsolved Business

Five firms. Five quiet catastrophes. The evidence is on your desk. The hours are short. The verdict is yours to render.

✦   ✧   ✦
Case FilesI — V
InvestigatorOne per Case
Per Case60 min crack · 90 min build
Preliminary Instructions

How to investigate a leaking business.

Each case is a body of evidence. No villain confesses in chapter one. Read everything. Build a timeline. Name the leak — and prove how you'd plug it with something you can ship by Monday.

I

Read First

Read every piece of evidence once before forming a theory. The first read is for understanding. The second read is where the leak shows itself. Resist the urge to diagnose on page one.

II

The Evidence

Slack threads, call logs, email chains, CRM exports, dashboards, the occasional handwritten memo. Some clues matter. Some are red herrings. Treat each artefact as testimony.

III

The Diagnosis

There is no single right answer. Multiple readings are valid. You will be judged on the strength of your reasoning — what you noticed, what you ruled out, and why.

IV

Build Time

Once the leak is named, you build. Ninety minutes. A Claude project, a Lovable app, an n8n flow, a script, a dashboard — whatever fits the leak. Technical or no-code. The Bureau is indifferent to your stack.

V

The Verdict

Five-minute presentation. Walk us through your timeline, your diagnosis, your build. The Bureau is unforgiving but fair.

VI

The Code

No googling the case. No comparing notes with someone working a different case. Claude, Lovable, and friends are welcome — but the reasoning must be yours. The Bureau notices.

The Schedule of Inquiry

Hour
Activity
Duration
00:00
Introductions. Settle in. Meet the room.
10 min
00:10
Crack the case. Read the evidence. Name the leak.
45 min
00:55
Build something that plugs the leak. Anything you can ship.
90 min
02:25
Demos. Walk the room through what you built.
30 min
Submitting Your Build

Once your project is ready, submit it at built.growthx.club. The Bureau reviews every submission.

Unlocking a Case File

Every case file is sealed. To open one, you'll need to solve a small word puzzle on the case page itself — a clue and a word with missing letters. Read the clue, fill in the word, and the file unseals.

1

Pick your case

From the sidebar on the left, click the case file you've been assigned.

2

Read the clue

You'll see a one-line clue and a word with some letters missing — for example, M _ S T _ R Y

3

Type the word

Fill in the blanks in your head. Type the full word into the box below and press Unlock. Get it right, and the file opens.

Case File No. I  ·  Marketing

The Case of the Vanishing Click

A candle company says they are making four times their ad spend back. The bank account disagrees.

The FirmHeatherwick Wax Co.
TradeD2C candles
Last reportedROAS 4× / Bank ₹5L
🔒

This case file is sealed.

Solve the word puzzle below to unlock the evidence within.

The clue

The art of figuring out which ad actually earned the sale — and which one is taking credit for someone else’s work.

_ T T R _ B _ T _ _ N
Not quite. Try again.
File Unsealed · Investigators Cleared

Heatherwick Wax Co. is a 2-year-old candle company. They sell on their own website (Shopify), Amazon, and Nykaa. Eight months ago they hired Florence as Head of Marketing. She runs Meta and Google ads.

Every Monday, Florence shares a "weekly numbers" update in Slack. Her dashboards look amazing — ROAS at 4×, growing 20% month-on-month. The team celebrates.

But Edmund, the founder, has been quietly worried. The bank balance does not match the celebration. He has asked you to take a look.

Edmund Heatherwick
Founder, CEO
Looks at the bank balance daily. Does not understand marketing dashboards. Suspicious but cannot articulate why.
Florence Vale
Head of Marketing
Hired 8 months ago. Confident, fast-talking, posts weekly wins. Reports to Edmund.
Ravi Kumar
Junior Marketing Analyst
Joined 3 months ago. Fresh out of college. Has been quietly cross-checking Florence's numbers in his own time.
Priya Shah
Bookkeeper, part-time
Tracks every rupee in and out. Has the cleanest data in the company.
Exhibit A
Slack — Florence's Monday post
A weekly celebration.
#marketing  ·  Mon 9:14 AM
Florence Vale9:14 AM🚀 Massive week team!
Spent: ₹10L
Revenue from ads: ₹40L
ROAS: 4× (industry avg is 2.5×)
We're crushing it 🔥
Exhibit B
WhatsApp — Edmund to Priya, Tuesday morning
A founder asks the bookkeeper a quiet question.
Edmund (8:42 AM): Priya, quick one — what was our actual revenue last month? Florence is showing ₹40L from ads but I don't think we made anywhere near that.

Priya (9:15 AM): Let me pull it together and get back to you in an hour.
Exhibit C
Email — Priya's reply with the numbers
The bank account, plainly stated.
Exhibit D
Google Doc — Ravi's private notes (never shared)
A junior analyst spots something.
Been looking at the Meta dashboard. Three things I don't understand:

1. Meta says we got 8,000 "purchases" last month. Shopify only shows 1,200 orders total across all channels.

2. The Meta pixel and the Google tag both seem to claim the same sales.

3. Florence's "ROAS 4×" comes from adding Meta-reported revenue + Google-reported revenue. But these are the same customers being counted twice (or three times).

Should I bring this up? She's senior to me.
— Ravi K., Google Doc
Exhibit E
Slack — Florence to Edmund, when he asked
A confident dismissal.
DM  ·  Florence → Edmund
Florence ValeWed 4:48 PMedmund the platforms are super accurate now, attribution has gotten really sophisticated. the ₹40L is what the algorithms are bringing us. bank reconciliation is messy because of refunds, payout cycles etc. trust the platforms 🙏

A few possibilities the Bureau has flagged. Treat these as starting points, not verdicts. Multiple readings are valid — and the right answer may not be on this list at all. Pen and paper for your own diagnosis.

Double-counting
Possible
Meta and Google both claim credit for the same sale. The "₹40L" is the same revenue counted twice.
Florence is making it up
Possible
She knows the numbers don't match and is hiding it because her job depends on the dashboard.
Attribution lag
Possible
The platforms are reporting revenue that hasn't shown up yet. Edmund is just being impatient.
Tracking is broken
Possible
The pixels are firing wrong and inflating the numbers. Nobody set them up properly.

Build something Edmund can use.

Ninety minutes. Build something that gives Edmund one number he can trust every Monday morning. Anything that surfaces the truth fits.

A suggested toolkit lives in the Appendix. Pick whatever you can ship in the time you have — Claude, Lovable, n8n, a Google Sheet with formulas, anything that gets the job done.

Case File No. II  ·  Hiring

The Case of the Silent Inbox

A fintech CTO says there are no good engineers in India. Twelve hundred candidates have something to say about that.

The FirmRavensworth & Co.
TradeSeries A fintech
Pass rate5% (60 of 1,200)
🔒

This case file is sealed.

Solve the word puzzle below to unlock the evidence within.

The clue

The thing every fair hiring process needs — written down, not in someone’s head.

R _ B R _ C
Not quite. Try again.
File Unsealed · Investigators Cleared

Ravensworth & Co. raised Series A nine months ago and promised investors they'd hire 15 engineers in six months. They've hired 4. The CTO, Reginald, has started telling people "the Indian engineering market is broken — there's no talent."

Beatrice, who runs People Ops, is exhausted. She set up a new "AI resume screener" three months ago to help her get through the volume. Since then, things have only gotten worse — but the dashboard says everything is fine.

You've been asked to figure out why the hiring pipeline is bleeding.

Reginald Ainsworth
CTO, Co-founder
Believes the candidate quality is bad. Hasn't personally interviewed anyone in 6 weeks. Looks at the hiring dashboard and sees green.
Beatrice Holloway
Head of People
Set up the AI resume screener 3 months ago to handle volume. Trusts it. Reports the "shortlist quality" to Reginald weekly.
Aakash Menon
Engineering Manager
Has been cold-DMing engineers on LinkedIn out of desperation. Some of the people he's reached out to did apply to Ravensworth.
Cordelia Wright
Recruiter (junior)
Joined 2 months ago. Manually reviews the shortlist the AI screener spits out. Has been quietly noticing that the same kinds of resumes always get through.
Exhibit A
Hiring Dashboard — what Reginald sees every Monday
A green-light dashboard.
Applications received
1,200
this quarter
Passed AI screen
60
5% pass rate
Interviewed
22
of 60
Hired
4
of 22

STATUS: ON TRACK PER AI SCREENER CONFIDENCE SCORING ✅

Exhibit B
LinkedIn DM — Aakash to a senior engineer
A cold outreach that reveals something cold.
Aakash → Priyanka R., Senior Backend Engineer, ex-Razorpay:
Hey Priyanka, would love to chat about a senior role at Ravensworth.

Priyanka:
Funny — I applied to your company 6 weeks ago and never heard back. Thought you weren't hiring anymore.

Aakash:
Wait what? Let me check. We're definitely hiring.

(Aakash, separately to Beatrice in Slack):
Bea, this is the THIRD person this month who told me they applied and got ghosted. What's going on?
Exhibit C
AI Resume Screener — Settings page
The rules nobody has reviewed.
Auto-reject candidates if:
· Less than 5 years of experience ❌
· No degree from a Tier-1 college (IIT/IIM/BITS) ❌
· Resume contains an employment gap of more than 3 months ❌
· Has not worked at a "top-tier" company (FAANG, unicorns, or Y Combinator-backed) ❌

Configured by: Beatrice H. · 3 months ago
Last reviewed: never
— ATS settings page
Exhibit D
Notion — Cordelia's private notes
A junior recruiter notices a pattern.
Something feels off. Every shortlist I review looks the same — IIT/IIM grads, ex-FAANG, no career breaks, all male.

We rejected 1,140 people this quarter. Beatrice says the AI is "filtering for quality" but I don't think anyone's actually checked what it's filtering OUT.

I don't want to step on toes but… should I say something?
— C.W., Notion (private)
Exhibit E
Slack — #leadership, Reginald's message
A diagnosis that may be the problem.
#leadership  ·  Thu 3:18 PM
Reginald Ainsworth3:18 PMThe candidate quality in India is genuinely shocking. Every shortlist I see is mediocre. We need to start hiring from abroad. Going to bring this up with the board.

A few possibilities the Bureau has flagged. Treat these as starting points, not verdicts. Multiple readings are valid — and the right answer may not be on this list at all. Pen and paper for your own diagnosis.

The AI screener is over-filtering
Possible
The auto-reject rules are so strict that great candidates are getting cut before any human sees them.
Reginald is the bottleneck
Possible
He hasn't interviewed in 6 weeks. Even good shortlists sit idle. The funnel is fine; the founder isn't.
The job is genuinely unattractive
Possible
Ravensworth pays badly or has a bad reputation, so good candidates self-select out.
The dashboard hides the rejected pile
Possible
Reginald only sees what got through. Nobody is looking at the 1,140 who didn't.

Build something Beatrice can use.

Ninety minutes. Build something that helps Beatrice see who's getting rejected and why — not just who's getting through.

A suggested toolkit lives in the Appendix. Pick whatever you can ship in the time you have — Claude, Lovable, n8n, a Google Sheet with formulas, anything that gets the job done.

Case File No. III  ·  Sales

The Case of the Phantom Pipeline

A B2B SaaS forecasts confidently every quarter. Every quarter they miss. Every quarter, the same explanation.

The FirmWhitford & Hale
TradeB2B workflow software
Quarterly miss40% under forecast
🔒

This case file is sealed.

Solve the word puzzle below to unlock the evidence within.

The clue

A confident number for next quarter that nobody seems able to actually hit.

F _ R _ C _ S T
Not quite. Try again.
File Unsealed · Investigators Cleared

Whitford & Hale sells workflow software to mid-market companies. They have one senior salesperson, Bea, who has been there for two years and brings in 70% of the pipeline. They have a sales lead, Hugo, who pulls up the forecast every Monday and announces a confident number.

The number is wrong. They've missed forecast by ~40% three quarters in a row. Hugo's explanation is the same every time: "Q4 will be the inflection."

Marcus, the CEO, has stopped believing the explanation. He's brought you in before the next board call.

Hugo Whitford
VP Sales
Trusts what his salespeople tell him. Has not joined a customer call in two quarters. Believes activity equals pipeline.
Beatrice "Bea" Carrington
Senior Account Executive
Top performer on paper. Every deal is "very confident, signing soon" — and has been for months.
Marcus Albright
CEO, Co-founder
Has personally jumped onto three of Bea's biggest deals "to help close." None of them have closed.
Henrietta Sloane
Sales Operations
Quietly pulled the actual deal-by-deal data two months ago. Has not shown anyone yet. Has reasons.
Exhibit A
CRM — Bea's top 5 deals (snapshot)
All "Negotiation." All for a long time.
AccountARRStageDays in StageLast reply from customer
Pemberton Logistics₹60LNegotiation150 days2 months ago
Ashford Retail₹50LNegotiation190 days3 months ago
Brookfield FMCG₹80LNegotiation200 days2 months ago
Crestmont Mfg₹40LNegotiation110 days3 weeks ago
Hartley Foods₹95LNegotiation170 days2 months ago

Average days in "Negotiation" across the rest of the team: 30. Bea's average: 164.

Exhibit B
Slack — Monday forecast call
A confident number, every week.
#sales-forecast  ·  Mon 10:00 AM
Hugo Whitford10:02 AMteam — quarterly forecast call in 30. how are we looking? bea kick us off
Bea Carrington10:04 AMstrong week 💪 pemberton is hot, expecting paper this fortnight. hartley super engaged. ashford going quiet but rohan said it's just budget cycles. brookfield — vikram is internally championing. all five closing this quarter, easy.
Hugo Whitford10:05 AMlove it. ₹3.3 cr weighted. that puts us at 88% of target.
Exhibit C
Email — Ashford Retail to Bea (2 months ago, no reply)
A customer who already moved on.
Exhibit D
Henrietta's spreadsheet (private — never shared)
The data that would change the conversation.
SalespersonDeals in "Negotiation"Of those, % that actually closed
Bea Carrington22 (last 12 months)11%
Other AEs (combined)3858%

Note: the forecast model assumes "Negotiation" stage closes at 70% across all reps.

Exhibit E
Henrietta — draft message to Hugo (never sent)
An ops lead, hesitating.
Hugo, I have something I think you should look at.

Bea's "Negotiation" stage closes at 11%. The model is weighting it at 70%. That alone explains most of our forecast miss.

Either she's moving deals to Negotiation that aren't really there, or our stages mean different things to different reps. Probably both.

I haven't sent this. Bea is your top rep. If I'm wrong, this ends my career here. If I'm right, it ends hers. Trying to figure out the right way to bring it up.
— H.S., draft, never sent

A few possibilities the Bureau has flagged. Treat these as starting points, not verdicts. Multiple readings are valid — and the right answer may not be on this list at all. Pen and paper for your own diagnosis.

Bea is gaming the stages
Possible
Moving deals into "Negotiation" because the forecast rewards it. Most of those deals are dead.
The forecast model is broken
Possible
It applies the same close-rate to every rep, regardless of who is reporting. Bea's reality is not the team average.
Enterprise sales is just slow
Possible
Big deals take time. Hugo's "Q4 inflection" might actually be true. Maybe the Bureau is being impatient.
Nobody is checking the data
Possible
Henrietta has the answer. Hasn't shared it. The org has a culture of not asking hard questions about top performers.

Build something Marcus can use.

Ninety minutes. Build something that gives Marcus a forecast he can actually trust — based on what's really happening in deals, not what salespeople say is happening.

A suggested toolkit lives in the Appendix. Pick whatever you can ship in the time you have — Claude, Lovable, n8n, a Google Sheet with formulas, anything that gets the job done.

Case File No. IV  ·  Product

The Case of the Buried Feedback

A productivity app ships features every week. Their power users love it. Everyone else is quietly leaving.

The FirmLockwood Atlas
TradeProductivity SaaS
Net retention88%, falling
🔒

This case file is sealed.

Solve the word puzzle below to unlock the evidence within.

The clue

The plan that decides what gets built — and, by extension, who gets ignored.

R _ A _ M _ P
Not quite. Try again.
File Unsealed · Investigators Cleared

Lockwood Atlas is a productivity app — task lists, calendar, notes. Three years old. Loved by a small group of "power users" who use it 5 hours a day and post about it on Twitter.

The team ships features fast. The CEO, Cassia, talks personally to a Slack group of 12 power users and turns their requests into roadmap items. The product gets more powerful every month.

And yet revenue is shrinking. Cassia thinks the answer is to ship faster. Wystan, the head of product, thinks the answer is something else entirely. He just hasn't been able to get her to listen.

Cassia Lockwood
CEO, Founder
Replies personally to her power users in Slack. Has not opened a churn report this quarter. Believes "the most engaged users see the future."
Wystan Crowe
Head of Product
Has been quietly pulling churn data. Has tried to flag it to Cassia twice. Both messages got "love it, let's talk soon."
Imogen Hart
Head of Customer Success
Talks to churning customers every day. Hears the same thing on every call. Nobody asks her about it.
"The Power 12"
Top users (Slack group with Cassia)
12 people. Use the app daily. Pay ₹500/month each. Are 0.4% of revenue. Drive 60% of the roadmap.
Exhibit A
Internal report — revenue by user type
Where the money actually comes from.
User segmentHow many% of revenueRetention (1 year)
Power users (top 1%)~150~5%95%
Regular paying users (rest)~10,000~95%62%
Exhibit B
Slack — #power-users (private, 12 members + Cassia)
Where the roadmap is actually decided.
#power-users (private)  ·  Tue 8:00 AM
Felix R.8:18 AMcassia! big idea — what if we could nest tags inside conditional filters that auto-archive based on calendar status? would be GAME CHANGING for my workflow 🔥
Cassia Lockwood8:24 AMfelix this is genius. flagging to wystan to scope. love how you're using the product 🙏
Mira K.8:31 AM+1 to felix. also could we get markdown footnote support in the comment threads?
Cassia Lockwood8:33 AMnoted! adding to the roadmap doc 📝
Exhibit C
Imogen's call notes — last 10 churn interviews
A pattern, in regular customers' words.
"I just wanted a simple to-do list. The new version has too many menus."

"I loved this 18 months ago. It feels like it's now built for someone with way more time than me."

"Onboarding has gotten harder. Took me a week to figure out where the basic features moved to."

"Three new features last month. I don't need any of them. I need search to be faster."

"My team uses it but I gave up. It became a hobby app."

9 of 10 churn interviews had the same theme: too complex, getting worse.
— Imogen H., CS notes
Exhibit D
Email — Wystan to Cassia (opened, not replied)
A head of product trying to be heard.
Exhibit E
Cassia's recent post in #all-hands
The official line.
#all-hands  ·  Cassia Lockwood
Cassia Lockwood9:00 AMteam — going to be honest, this quarter has been tough on retention. I think we need to ship FASTER, not slower. our power users are showing us the future. when the rest of the market catches up to where they are, our retention will follow. trust the vision 💪

A few possibilities the Bureau has flagged. Treat these as starting points, not verdicts. Multiple readings are valid — and the right answer may not be on this list at all. Pen and paper for your own diagnosis.

Building for the wrong customer
Possible
The roadmap serves the loud 1%. The quiet 99% pays the bills and is leaving.
The product is too complex now
Possible
Every new feature adds menus and friction. Regular users want simple. They're getting overwhelmed.
No system to hear quiet feedback
Possible
Power users have a Slack with the CEO. Regular users have no channel. Their voices never reach the room.
Cassia's vision is right, just early
Possible
Maybe she's correct that the market will catch up. The retention drop is just the cost of being early.

Build something Wystan can use.

Ninety minutes. Build something that helps Wystan hear the quiet 99% — the regular paying users whose voices never reach the roadmap.

A suggested toolkit lives in the Appendix. Pick whatever you can ship in the time you have — Claude, Lovable, n8n, a Google Sheet with formulas, anything that gets the job done.

Case File No. V  ·  Growth

The Case of the Leaking Funnel

A meditation app celebrates installs every Monday. By the next Monday, almost nobody is left.

The FirmMarlowe & Finch
TradeMindfulness mobile app
D7 retention9% (was 19%)
🔒

This case file is sealed.

Solve the word puzzle below to unlock the evidence within.

The clue

The number that lies less than installs ever will.

R _ T _ N T _ _ N
Not quite. Try again.
File Unsealed · Investigators Cleared

Marlowe & Finch is a meditation app. Two years old. Every Monday the growth team celebrates installs, which are up 40% quarter-on-quarter. The dashboard is a beautiful gradient going up and to the right.

There is another number — D7 retention. That's what percentage of users come back a week after they install. A year ago it was 19%. Now it is 9%. Of every 100 people who download, 91 are gone within a week.

The growth lead, Imelda, blames the product. The CEO, Cosmo, blames onboarding. A junior PM has been quietly tracking something different — and nobody has asked him about it.

Cosmo Marlowe
CEO, Founder
Talks about "the magic" of the product a lot. Has not personally onboarded as a new user in over a year. Believes more meditations = more retention.
Imelda Finch
Head of Growth
Buys ads on Meta and Snapchat. Reports installs every Monday. Convinced retention is "a product problem, not a growth problem."
Ezra Whitlock
Junior PM
Joined 4 months ago. Quietly built a side dashboard tracking what new users actually do in their first week. Has shown nobody.
Pippa Doyle
Lead Designer
Redesigned the onboarding 8 months ago. The redesign was beautiful. Cosmo loved it. Pippa has not looked at the metrics since.
Exhibit A
Growth Dashboard — what the team sees on Monday
The numbers everyone celebrates.
Monthly installs
140,000
▲ 40% QoQ
Cost per install
₹85
▼ 12% QoQ
D1 retention
34%
flat
D7 retention
9%
▼ from 19% a year ago

Note: the Monday standup only shows installs and cost-per-install. Retention exists but is on a tab nobody opens.

Exhibit B
Ezra's private dashboard — what new users actually do
The number nobody is looking at.
What new users did in week 1% of installsTheir D7 retention
Listened to ≥30 sec (the "activated" definition)78%9%
Completed 1 full meditation (≥5 min)12%42%
Completed 3 meditations in week 14%71%

Translation: the official "activated" metric (78%, looks great) does not predict retention at all. The real predictor is "3 meditations in week 1." Almost nobody hits it.

Exhibit C
Slack — #growth-weekly
A celebratory Monday and a quiet flag.
#growth-weekly  ·  Mon 9:30 AM
Imelda Finch9:30 AMhuge week 🎉 140k installs, up 40% qoq, cost per install at an all-time low. snap creative is crushing it 🔥
Cosmo Marlowe9:32 AMamazing!!! the magic is spreading 🌱
Ezra Whitlock9:48 AMquick flag — d7 is at 9%, down from 19% a year ago. wanted to make sure that's on someone's radar.
Imelda Finch9:51 AMretention is a product problem ezra! growth's job is to fill the bucket 🪣 cosmo's team needs to plug the leaks 🙏
Cosmo Marlowe9:53 AMwe're shipping a new "deep focus" course next month, that should help retention
[no follow-up]
Exhibit D
Pippa's design notes — onboarding redesign, 8 months ago
A change that was celebrated.
Onboarding redesign shipped today! Really proud — the new welcome flow is gorgeous, the new permission flow feels much cleaner. Cosmo loved the polish.

One small change: moved the push-notification permission ask from screen 2 (felt aggressive) to screen 8 (after the user has had their first calm experience).

Should be a much better user experience.
— Pippa Doyle, design notes
Exhibit E
Push notification opt-in (graph, never shown in standup)
A line that fell off a cliff.
Push opt-in (1 year ago)
51%
stable for years
Push opt-in (after redesign 8mo ago)
22%
▼ 29 points
Users who never return after D1
66%
▲ from 48% a year ago

Without push permission, the app cannot remind users to come back. The drop in opt-in lined up exactly with the onboarding redesign.

A few possibilities the Bureau has flagged. Treat these as starting points, not verdicts. Multiple readings are valid — and the right answer may not be on this list at all. Pen and paper for your own diagnosis.

The "activated" metric is wrong
Possible
"30 second listen" predicts nothing. The team is celebrating the wrong number every week.
The redesign broke the loop
Possible
Push opt-in halved when Pippa moved the ask. Without push, no reminders. Without reminders, no return.
Acquisition is bringing the wrong audience
Possible
Snap ads are pulling people who want a 30-second hit, not 5-minute meditations. The audience and product don't match.
Org structure failure
Possible
"Growth fills the bucket, product plugs the leaks." Classic broken org. Nobody owns end-to-end retention.

Build something Cosmo can use.

Ninety minutes. Build something that gives the team a clearer view of what actually predicts retention — not just installs.

A suggested toolkit lives in the Appendix. Pick whatever you can ship in the time you have — Claude, Lovable, n8n, a Google Sheet with formulas, anything that gets the job done.

The Verdict

How the Bureau judges.

Five minutes per investigator. The rubric below is shared with all judges. Reasoning beats polish; honesty beats certainty; a working build beats an elegant deck.

#
Criterion
Weight
i.
Timeline Reconstruction
Did you order the events correctly? Did you spot the moment the leak began?
15 pts
ii.
Quality of Diagnosis
Is the leak named precisely? Does the reasoning hold? Does it survive the counter-argument?
25 pts
iii.
Evidence Handling
Did you cite specific exhibits? Did you spot what others would miss? Did you ignore the red herrings?
15 pts
iv.
Build — Design
Does the build address the actual leak? Is it scoped tightly? Does the design follow from the diagnosis?
25 pts
v.
Build — Demo
Does the build work? Does it produce a real output someone at the firm could actually use?
15 pts
vi.
Presentation
Five clear minutes. No throat-clearing. Reasoning legible to a non-expert.
5 pts

"It is a capital mistake to theorise before one has data.
Insensibly one begins to twist facts to suit theories, instead of theories to suit facts."

— A. C. D., 1891

Appendix

Instruments of investigation.

A suggested toolkit. Pick one or two — do not pick six. The Bureau favours focus.

Reasoning Engine

Claude Project

For ingesting evidence, drafting prompts, generating briefs. Pin your case files as Project knowledge. Claude does the heavy thinking; you do the framing.

Front-End

Lovable

Spin up a dashboard, a triage tool, a daily-brief inbox in under an hour. Best for anything a human will look at every day.

Workflow Glue

n8n / Make

Stitches APIs without code. Pull from Sheets, fan out to Slack, run on a schedule. The unsexy backbone of every shipped automation.

Data Layer

Google Sheets / Airtable

Where structured data lives. Connect via API or webhook. The boring, dependable filing cabinet. Do not overlook.

Notification

Slack Webhooks

Where the brief lands. A two-paragraph daily digest in #leadership beats a perfect dashboard nobody opens.

Voice / Calls

Fireflies / Otter

Transcribes calls. Feed transcripts into Claude. Sales and CS cases come alive when the actual words are on the table.

Emergency

Cursor / Claude Code

If the workflow needs custom logic, drop into a code editor. Claude can write the script; you orchestrate. Use sparingly.

Storage

Supabase / Vercel Postgres

If your build needs persistence beyond a Sheet. Free tiers are generous. Don't reach for it unless you must.

A Closing Word from the Bureau

You are not here to ship a perfect product. You are here to render a verdict. The best detectives in this room will not be the ones who built the prettiest dashboard — they will be the ones who, when asked, can defend with a straight face why the leak is what they say it is, and why their workflow plugs it. Build less. Notice more. Be specific.

The evidence is on your desk. The clock starts when you open the file.