# Six jobs you can spec against the deal set

Starting points, not a menu. If you brought your own job, build that one.

Each is narrow enough to specify in fifteen minutes and each starts with a verb.
The Job and the User are given. **You write the Done** — that is the part being
practised, and it is the part that is hard.

Each job lists the number of rows one precise Done returns. If your number is
different, your Done and this one are not asking the same question. That difference
is the exercise, not a mistake.

---

### 1. Identify

**Job.** Identify open deals that have gone quiet, and list them by the rep who owns
them.

**User.** A RevOps analyst who sends the list to sales managers on Monday morning.

**Done.** *You write it.* Treat "gone quiet" as no recorded activity since
5 May 2026.

> A precise Done returns **634** rows across all 50 reps.
> Against the 200-row sample: **15** rows, plus **4** with no activity date.

---

### 2. Reconcile

**Job.** Reconcile how a lost deal was coded internally against what the buyer said,
and list the deals where the two disagree.

**User.** A sales enablement lead preparing a quarterly loss review.

**Done.** *You write it.*

> A precise Done returns **866** rows, out of 2,479 closed-lost deals.
> Against the 200-row sample: **20** rows, out of **55** closed-lost.

---

### 3. Route

**Job.** Route open deals that have fewer than three known contacts to the rep's
manager.

**User.** A sales manager who wants them worked before the quarter closes.

**Done.** *You write it.* The data has no manager column — decide what routing
means and say so.

> A precise Done returns **1,463** rows.
> Against the 200-row sample: **31** rows.

---

### 4. Prepare

**Job.** Prepare a weekly digest of quiet open pipeline for a single territory.

**User.** The regional director for that territory, who reads it in a Monday email.

**Done.** *You write it.* Same definition of quiet as job 1.

> For **East**, a precise Done returns **148** rows and **$19,392,867** in
> `final_amount`. The other four territories run between 105 and 132 rows.
> Against the 200-row sample, East returns **2** rows and **$126,666**.

---

### 5. Summarize

**Job.** Summarize one rep's closed-lost deals into a coaching note their manager
can read in two minutes.

**User.** A first-line sales manager preparing for a one-to-one.

**Done.** *You write it.* This is the one where a checkable Done is hardest, because
the output is prose. Write a Done anyway.

> Each rep has between 35 and 65 closed-lost deals; the median is 50.
> `Rep-20` has 63.
> Working from the 200-row sample, use **Rep-33** (**4** closed-lost rows there);
> Rep-20 has none in the sample.

---

### 6. Approve

**Job.** Approve for correction every deal whose expected close date falls before the
deal was created.

**User.** A RevOps administrator who will run the correction after someone signs off.

**Done.** *You write it.*

> A precise Done returns **2,248** rows.
> Against the 200-row sample: **42** rows.

---

### 7. Yours — the app you already built

**Job.** Build the system behind the app you already have: the storage, the
sign-in, the refusals, the tests, and the owner it is missing.

**User.** Whoever already depends on the app, or is about to.

**Done.** *You write it* — and because nobody publishes a target number for your
app, the check is different: **run your Done twice and get the same number both
times.** A Done that cannot produce the same answer twice is not yet a check.

The path differs in two places, and only two:

- **Start with the repository-assessment prompt** (`prompt-pack.md`, number 1)
  instead of picking from the six above. What it returns is your draft PRODUCT.md.
- **Your contract is your own schema.** Ask the agent to print every table, column,
  and permission your app actually has. That printed list replaces `schema.md` in
  the counting exercise: anything your plan names that your schema does not
  contain, your plan invented.

Everything else — the plan prompt, the count, the security test, the nine checks —
runs against your app unchanged.

---

## Two things worth deciding out loud

**What the tool does when it finds nothing.** Every job above returns rows today.
Say what happens on the week it returns zero.

**What happens if it runs twice.** Jobs 1, 3 and 4 produce a list somebody acts on.
Say whether running it again on Tuesday creates a second list, updates the first, or
does nothing.
