# The deal set

10,000 sales transactions. Synthetic, courtesy of Andy, from the AI in GTM School
Class 0 data set. Nothing in it describes a real company or a real person.

Closed history runs 2023 to 2025. Open pipeline carries expected close dates into 2026.

It is a working export rather than a tidied one. It has not been cleaned, and it is
not guaranteed to be internally consistent.

## Reaching it

No account, no key, no sign-in. Three ways in, and any of them is fine.

**Paste the sample.** `sample-200.csv` is 200 rows from the same data, small enough
to paste into any assistant. It carries every deal stage, territory, main loss reason
and special tag, so all the shapes are present. It does not reproduce the full set's
proportions, and it does not contain every loss subtheme.

https://storage.googleapis.com/vibecoding-201-data/sample-200.csv

**Give your agent the whole file.** 10,000 rows, 2.7 MB. Fine for a coding agent
that can read a file; too large to paste into a chat window.

https://storage.googleapis.com/vibecoding-201-data/deals-10k.csv

```bash
curl -O https://storage.googleapis.com/vibecoding-201-data/deals-10k.csv
```

**Or open the source.** The spreadsheet this came from is
[here](https://docs.google.com/spreadsheets/d/1XMqMIhcA5BGe8bD5RNIfVFbE1vuLFL5cKyf9WukrhFA/edit).
Good for looking at. It rate-limits when a lot of people query it at once, so work
from the files above during the exercise.

**Getting a 403 from a cloud workspace?** Some sandboxed environments — Claude
Cowork among them — block outside downloads. That 403 is the sandbox's, not the
bucket's: these links work in any browser. Download the file on your own machine
and attach it to the chat or project, then continue. The sample (55 KB) pastes
anywhere; the full set (2.7 MB) attaches fine.

## The 36 columns

The files are public and read-only. Your tool imports a copy into its own database
and owns that copy; nothing you build writes back to this bucket.

**The file's headers are Title_Case: `Transaction_ID`, `Deal_Stage`,
`Last_Activity_Date`.** This page writes the same 36 columns in lowercase, and the
file's column order differs from the order below. Match by name,
case-insensitively, never by position. SQL tools do this on their own; code that
matches names case-sensitively finds nothing and reports **zero rows from a tool
that looks finished**. If your first number is 0, check this before checking
anything else.

| Column | Type | Filled | Values |
|---|---|---|---|
| `transaction_id` | text, primary key | 100% | 10,000 distinct |
| `transaction_date` | date | 100% | the date the deal was created |
| `expected_close_date` | date | 91.5% | |
| `last_activity_date` | date | 98.0% | |
| `seller_company` | text | 100% | 100 distinct |
| `buyer_company` | text | 100% | 100 distinct |
| `product_category` | text | 100% | 12: Analytics Tools, Cloud Services, Consulting, Hardware Equipment, Integration Services, IT Infrastructure, Marketing Services, Professional Services, Security Solutions, Software Licenses, Support Services, Training |
| `quantity` | integer | 100% | 5–100 |
| `unit_price` | numeric | 100% | 51.09–4999.28 |
| `total_amount` | numeric | 100% | 255.45–498957.00 |
| `discount_percent` | integer | 100% | 0, 5, 10, 15, 20 |
| `discount_amount` | numeric | 100% | 0–96005.40 |
| `final_amount` | numeric | 100% | 242.68–489606.00 |
| `sales_rep` | text | 100% | 50: `Rep-1` … `Rep-50` |
| `sales_rep_tier` | text | 100% | Junior, Mid, Senior |
| `territory` | text | 100% | Central, East, North, South, West |
| `customer_segment` | text | 100% | SMB, Mid-Market, Enterprise |
| `industry` | text | 100% | Education, Finance, Government, Healthcare, Manufacturing, Retail, Technology |
| `company_size` | text | 100% | `1-50`, `51-200`, `201-1000`, `1001-5000`, `5000` |
| `decision_maker_persona` | text | 100% | 12: CFO, CIO, CISO, Director IT, Director Operations, Founder / CEO, Head of Ops, Head of Procurement, Owner-Operator, VP Engineering, VP Operations, VP Sales |
| `contact_count` | integer | 100% | 0–7 |
| `deal_stage` | text | 100% | Prospecting, Qualification, Proposal, Negotiation, Closed Won, Closed Lost |
| `win_probability` | integer | 100% | 0, 10, 25, 50, 75, 100 |
| `deal_size_category` | text | 100% | Medium, Large, Enterprise, Strategic |
| `quarter` | text | 100% | Q1–Q4 |
| `year` | integer | 100% | 2023, 2024, 2025 |
| `month` | text | 100% | January – December |

### Columns that only closed deals carry

The nine columns below describe how a deal ended, so an open deal has nothing to put
in them. Their fill rates are what they are for that reason, not because data is
missing.

| Column | Type | Filled | Values |
|---|---|---|---|
| `sales_cycle_days` | integer | 50.2% | 14–240 |
| `competitor` | text | 50.2% | 39 distinct, including `No competitor`, `No-decision` and `Internal build` |
| `has_buyer_interview` | text | 50.2% | Yes, No |
| `loss_reason_main` | text | 24.8% | Company Perception, Price / Value Position, Product Performance, Sales Experience |
| `loss_reason_subtheme` | text | 24.8% | 27 distinct |
| `buyer_reported_reason_main` | text | 24.8% | same four values as `loss_reason_main` |
| `buyer_reported_reason_subtheme` | text | 24.8% | 27 distinct |
| `win_reason` | text | 25.5% | 10: Best feature fit, Best champion enablement, Strongest discovery, Best implementation plan, Best ROI case, Trusted brand / reputation, Best references, Strongest integration story, Best executive sponsorship, Best price |
| `special_tag` | text | 7.9% | Champion churn, Consolidation, Incumbent switching cost, Internal build, No-decision / status quo won, Regulatory / legal block, Stalled / deprioritized |

`loss_reason_*` records how the deal was coded internally. `buyer_reported_reason_*`
records what the buyer said. They are two separate fields and they are populated
independently.

## Using this in your spec

This list is the whole contract. If a plan comes back referring to a table, a column,
a status, or a permission that is not written above, the plan invented it — and that
invention is now yours to own.

That is the count to keep.
