A customer agreed to $10,000 and paid $6,000 in cash before the product existed.
Cash payment. Verbal agreement, no written contract. The customer can confirm it directly.
Each row states the claim, what the evidence actually is, and whether it can be shown. Where evidence is not public, that is stated rather than implied.
$10,000 AGREED · $6,000 PAID — $6,000 IN CASH, BEFORE A PRODUCT EXISTED
A Las Vegas travel wholesaler with more than a decade of operating history agreed to pay $10,000 for the system and paid $6,000 of it in cash before any product existed. The customer also supplied the development hardware and a phone, and covers the monthly AI tooling cost. The balance is due when the system is usable. There is no written contract — the agreement is verbal, and it has been honoured in cash. The $10,000 covers their use of the system; there is no separate licence fee on top of it. The founder sends a development update every week. The customer has also agreed to introduce other businesses in the travel industry once the system is ready. No introduction has produced revenue yet, and commercial terms for those introductions have not been set.
This is one customer, in one industry, buying the business system. It is presented as a real prepayment and a real relationship — not as product-market fit, and not as a purchase of the engineering control layer.
This is what actually happened in the conversation that ended with $6,000 in cash.
I explained what the system does and why it is built differently. He was interested.
Then I asked him how his business actually runs.
He told me, one piece at a time. I spent at least thirty minutes just aligning with him. I did not know his vocabulary; he explained it and I still only half understood.
So I started mapping it in front of him — acting as the process analyst, while thinking about how the thing would have to be governed and operated.
When he finished, he said: the others are messy too, and every one of us does it differently.
I asked how messy. He said: let me show you the sheet.
Then he said: this sheet is only part of it. There is far more old, unresolvable mess behind it.
Every month, money is under-counted.
And to reconcile anything, someone has to go dig through chat histories across every tool they use.
That sheet is the data measured below.
One of those channels leaves no record at all. A spoken phone call is not a weak record — it is an absent one. Everything downstream is then a reconstruction that nobody can verify.
One sheet, several people. Whoever has time does it. In the end nobody can read it.
That sentence is the cause. The twenty-two different header structures measured below are the effect. He said it before anyone counted them.
He was not shown a product. He was shown his own business, laid out, for the first time. He paid $6,000 in cash.
Four years of booking records from a live travel wholesale operation, provided by the customer. The figures below were computed from the files themselves, not supplied as a summary.
| 2023 | 12,492 bookings61,066 travelers |
|---|---|
| 2024 | 13,740 bookings75,786 travelers+24.1% YoY |
| 2025 | 13,151 bookings79,994 travelers+5.6% YoY |
| Total | 39,383 bookings216,846 travelers+31.0% over three years |
| Period covered | 2022-01 → 2025-12 |
|---|---|
| Booking records | 39,383 |
| Traveler count | 216,846 |
| Monthly worksheets | 37 |
| Partner company name strings | 1,420 |
| Top 3 partners | 48.8% |
| Top 10 | 67.5% |
| Top 20 | 79.2% |
| 823 partners appearing once, combined | 2.1% |
| Peak month vs quietest | May 4,912 · Jan 1,646 — 3.0× |
| Distinct header structures across 37 sheets | 22 |
|---|---|
| Records with the traveler name missing | 10,902 · 27.7% |
| Values in the phone column that are not phone numbers | 88.5% |
| Company names appearing exactly once | 823 of 1,420 |
| Near-duplicate company names in a 600-name sample | 43 pairs |
| Records with no confirmation number | 958 · 2.4% |
| Records in the 2025 workbook whose tour date is 2024 | 5,322 · 40% of that file |
| Records in the 2023 workbook dated 2022 | 933 |
| Distinct tour codes for 39,000 records | 19,607 — each used 2.0 times on average |
| Cell comments — people writing notes inside the grid | 3,750 |
|---|---|
| Cells with a background colour | 50,909 · 6.3% |
| Distinct fill colours used as status | 37 |
| Anywhere recording what each colour means | none |
| Formula cells — calculation logic scattered across the grid | 88,269 · 4.9% |
| Values found in the column headed “Phone number” | 确认 · sz确认 · SZ确认 · 没收到voucher |
| Distinct departure times in use | 21 |
A colour is a decision. A comment is a conversation. A formula is a rule. All three are real business logic, and none of them can be queried, exported, verified or handed to the next person. When the person who chose the colour leaves, the meaning leaves with them.
| Upper RT — one product, four column names | 100 / 110 / 120 / 130 Upper RT |
|---|---|
| Upper PT — one product, four column names | 120 / 130 / 140 / 150 Upper PT |
| Times the Upper RT column was renamed in three years | 8 |
| Names used for the phone field in 2023 alone | Phone number → Cell → Phone → Phone number |
| Distinct column names across 36 sheets | 47 |
| Of those, columns that appear in fewer than 30 sheets | 33 |
| Column count per sheet | 19 / 20 / 21 / 22 |
To answer "how many Upper RT tours did we sell", someone has to know that four differently-named columns are the same product.
Nothing in the file records that.
There is no product entity here. A product is a column heading, and it changes whenever the price does.
That is where twenty-two header structures come from — and why the count comes out different every time someone does it by hand.
This is one customer's data. It is not a claim about the industry, and no other company's data is being described here.
The founder analysed these three files and showed the customer what was actually inside them: twenty-two different header structures in one year's workbook, a phone column where 88.5% of the values are not phone numbers, and no way to answer how many partners the business actually has.
The customer paid $6,000 in cash before any product existed.
They had not seen a product. They had seen their own data.
Only counts and percentages are published. No names, phone numbers, invoice values or revenue figures from this dataset appear on this site or in any export.
The following is the account of an operator with more than a decade in this industry — someone who knows not only how his own company runs, but how his peers run theirs. It was given in conversation. It is informed testimony, not market research: no survey was conducted and no other company was asked directly.
NGF is positioned as “Weave your enterprise truth.” Describing their own operation, the operator described the exact absence of it: one sheet, several people, whoever is free does it — and in the end nobody can read it.
Two hours of conversation. No product. The customer said the system would help a lot of people — and paid $6,000 in cash.
Recorded as one person's account of their own market. It is not presented as verified industry data.
These figures come from a database audit of NGF dated 2026-08-08, not from a summary written for this website. Every one of them is recomputable from the schema.
This is system scale, not data volume. It shows how large the built system is. The customer's nearly five years of real operating data is a separate fact, stated separately.
This is a real record from the founder's own development, sanitized. It is the first published case where a completion claim was refused.
Nobody was watching. There was no reviewer, no auditor, no team. The founder was the only person who would ever know whether the claim was 22 or 20 — and the record says 20.
AI Self-Report ≠ Acceptance. Looks equivalent ≠ proven equivalent. Partial completion ≠ success.
Internal module names, file paths and source structure are withheld. The counts, the ruling and the reasoning are reproduced as recorded.
Four third-party figures. Each one is linked to its source, with the year it was published.
Gartner expects 90% of enterprise software engineers to be using AI code assistants by 2028 — up from under 14% in early 2024.
Gartner · 2026 ↗Gartner also expects 40% of enterprises may decommission AI agents they have already deployed, because governance failed.
Gartner · 2026 ↗CISQ puts the annual cost of poor software quality in the U.S. at $2.41 trillion, of which $1.52 trillion is accumulated technical debt.
CISQ · 2022 ↗The AI code tools market is roughly $9.4–10.1 billion in 2026 and growing about 28% a year. NCT is the governing layer inside that market — not an AI-compliance product.
Precedence Research · 2026 ↗These are third-party estimates, not our measurements. Published figures disagree: for the AI-governance category alone, estimates for the same year range from $249 million to $1.1 billion. Every line above links to the source it came from.
Each row states the claim, what the evidence actually is, and whether it can be shown. Where evidence is not public, that is stated rather than implied.
Cash payment. Verbal agreement, no written contract. The customer can confirm it directly.
A two-hour conversation with the operator. The account is published verbatim in summary form.
Photographs of the customer-provided hardware.
Three annual workbooks of booking records provided by the customer. All figures on this site were computed directly from those files.
Database audit dated 2026-08-08. FK_LEDGER.csv contains 1,213 rows. Every figure is recomputable from the live schema.
133 archived .app builds, B0001 through B0133, from 2026-08-28 13:04 to 2026-09-03 21:41. No missing build number. No duplicate. Build numbers and timestamps are strictly in the same order. Nothing was installed over, and nothing was deleted.
Development episodes, gate records, evidence packages from the founder's own work.
Separate development archives exist for NCT, NGF and NAO.
Nothing on this page is asserted without naming what would confirm it. Where the answer is "not public", the reason is stated. An unmeasured or unpublishable fact is never dressed up as a proven one.