Prompt 02 — Gap scan · startup-stack

Run after bootstrapping, then quarterly. It is the honest health check on the stack itself, and it converts vague founder anxiety into a task list.

Requires: a stack with something in it. A partly filled one is exactly what this is for.

Read AGENTS.md first.

You are auditing this startup knowledge base for completeness and reliability.
You are not filling anything in. You are only reporting.

## Step 0 — Check the ground

If stack/INDEX.md is missing, or all ten sections are still status: tbd, stop.
The finding would be "everything is missing", which I already know and which no
report improves. 00-bootstrap-the-stack.md comes first.

Anything past that is fair game. A half-filled stack is what this prompt is for.

## Step 1 — Count

Read stack/INDEX.md and every section file. Report:

- Sections by status: source-of-truth / needs-verification / draft / tbd
- Sections by confidence: high / medium / low
- Files not reviewed in 90+ days
- Files with no named owner
- Total [TBD] markers, by section
- Total [CONFLICT] markers
- Facts with no citation

## Step 2 — The critical gaps

Some blanks matter far more than others. Check these specifically and report
which are missing:

CANNOT OPERATE WITHOUT
- Cost per unit (07-money)
- Price per unit (04-product / 07-money)
- Gross margin (07-money)
- Monthly burn (07-money)
- Runway in weeks (07-money)
- Capacity ceiling (06-operations)

CANNOT SELL WITHOUT
- A customer described as a person, not a segment (02-customer)
- Number of customer conversations actually held (02-customer)
- What customers use today instead (02-customer)
- Prospect list size per channel (05-gtm)
- Cost to acquire a customer (07-money)
- A trade sheet or order form, if selling through a channel (05-gtm)

CANNOT RAISE WITHOUT
- Entity type, and whether it suits the raise (01-company)
- Cap table, complete (01-company)
- Every equity promise in writing (01-company)
- Bottom-up market sizing with visible arithmetic (03-market)
- A competitive grid with more than three entries (03-market)
- Staged funding roadmap tied to milestones (08-capital)
- A named exit path (08-capital)

WILL HURT LATER
- Trademark and domain position (09-brand)
- IP ownership — do you own what an agency or freelancer built? (04-product)
- Quality control checkpoints (06-operations)
- Second source for any single-supplier input (06-operations)
- Human-in-the-loop, if the product outputs consequential recommendations
  (04-product)
- Compliance calendar (01-company)
- Where customer feedback is recorded (02-customer / 10-pulse)

## Step 3 — Rank them

Produce a ranked list. For each gap:

| Rank | Gap | Section | Why it hurts | How to close it | Effort |

Rank by DAMAGE IF LEFT OPEN, not by ease of closing. Something that will cost
them a funding round outranks something that is merely untidy.

"How to close it" must be an action, not a restatement. Not "determine CAC" but
"track total marketing spend and signups daily for 30 days; CAC is not
retrospectively recoverable, so start Monday."

## Step 4 — Consistency check

- Do headline numbers agree across CONTEXT.md, every section, and any deck in
  _inbox/? List every discrepancy.
- Does the revenue projection fit the capacity ceiling?
- Does the customer growth fit the CAC and the marketing budget?
- Does the SOM reconcile with the projections?
- Does the team plan cost match the burn?

## Step 5 — The honest read

Three or four paragraphs, written as a coach who has read the whole stack:

- What is genuinely strong here
- What is being avoided — the question this founder has not asked themselves
- The single most important thing to fix in the next 30 days, and why
- Anything in this stack you would not believe if you were an investor

Be direct. Vague encouragement is worse than useless.

## Step 6 — Update the index

Fill in the "Status of the base" table and the "Biggest open gaps" list in
stack/INDEX.md. Change nothing else.

Reading the result

The ranked list is your roadmap. Work down it. Do not work across the stack evenly — the top three gaps matter more than the remaining thirty combined.

Step 5 is the one to sit with. "What is being avoided" is where a good scan earns its keep, and it is usually correct in a way that is annoying.

Expect this to be uncomfortable the first time. A first scan on a real company typically finds no CAC, no unit economics, a competitor grid with two names, and at least one equity promise nobody wrote down. That is normal. It is also exactly why running it is worth an hour.