# Codebook: quarterly gross margin from audited filings, 10 August 2026

Every figure in `gross-margin-panel.csv` is computed from **SEC EDGAR XBRL company facts**: the
audited numbers as filed. Nothing here is a benchmark, a survey, a vendor disclosure or an
estimate. The scripts that produced it are in `scripts/`, including **the two that failed**.

## The frame

The **64 constituents of the BVP Nasdaq Emerging Cloud Index**, read from the data payload of
`cloudindex.bvp.com/companies` on 9 August 2026 and fixed there:
`/data/pricing-pages-2026-08/frame.md`.

Reused deliberately. It is the same population as the pricing-page corpus, already justified,
already public, and: the point: **not conditioned on the outcome**. It is an index of cloud
companies, not a list of firms with AI features or with falling margins. A universe assembled by
looking for margin compression would find it.

## Window and panel

**2022Q1–2026Q1, 17 quarters.** 53 of the 64 constituents have a complete run across it:
**901 firm-quarters**, balanced. The other 11 are accounted for below and never silently dropped.

| | n | |
|---|---:|---|
| **Balanced**: every quarter in the window | 53 | the panel every headline figure is computed on |
| **Partial**: listed or first filed with the SEC mid-window | 7 | ALKT, FIG, KVYO, RBRK, SAIL, SHOP, TTAN: in the CSV, `in_balanced_panel = False` |
| **Excluded**: no quarterly US-GAAP gross profit exists | 4 | see below |

**The four exclusions, with the actual reason:**

| | |
|---|---|
| **CLBT, MNDY, WIX** | Israeli-domiciled foreign private issuers. They file annually (20-F) under IFRS and report gross profit **annually or semi-annually**, never quarterly. There is no quarterly figure to include |
| **WDAY** | Workday files revenue under US-GAAP throughout the window but tags **no** cost-of-revenue element under any US-GAAP tag in it: the cost lines are company-extension tags. Gross profit is not derivable without leaving the audited taxonomy |

## Three decisions that change the numbers, all of them consequential

**1. The fourth quarter is derived, because it is never filed.** A company files three 10-Qs and
one 10-K. Q4 exists only as *annual minus the three quarters inside it*. Reading the API naively
returns 13 or 14 quarters of a 17-quarter window and silently drops every Q4: the seasonally
strongest one. Q4 here is the arithmetic difference, and it reconciles to the 10-K by construction.

**2. Quarters are assigned by period MIDPOINT, not period end.** Salesforce's quarter runs
1 November to 31 January. Assigned by its end date it lands in the following calendar Q1, a full
quarter late. Its midpoint is mid-December, so it is Q4. Sixteen of the 53 have off-calendar fiscal
years and mis-phasing them shifts the whole series against the calendar.

**3. SEC's own `frame` field is not used.** It is assigned sparsely and it is assigned *least* to
exactly the off-calendar filers that decision 2 exists for. Using it produced 7 quarters for
Workday and 0 for Wix, which reads as an absent company rather than an absent method.

## Fields

| Field | Meaning |
|---|---|
| `symbol`, `name`, `cik` | constituent, and its SEC identifier |
| `quarter` | calendar quarter containing the period midpoint |
| `period_start`, `period_end` | the filed period, exactly as reported |
| `revenue_usd` | first available of `RevenueFromContractWithCustomerExcludingAssessedTax`, `…IncludingAssessedTax`, `Revenues`, `SalesRevenueNet` |
| `cost_of_revenue_usd` | first available of `CostOfRevenue`, `CostOfGoodsAndServicesSold`, `CostOfServices`, `CostOfGoodsSold`; where a filer reports only components (AppFolio), the components are summed |
| `gross_profit_usd` | `GrossProfit` where filed, else revenue − cost of revenue |
| `gross_margin` | gross profit ÷ revenue |
| `in_balanced_panel` | whether the firm has all 17 quarters |

Where a filer restated, **the latest filed value wins**. Rows with a computed margin outside (0,1)
are dropped as a tagging error rather than carried.

## What this data cannot do, stated before anyone uses it

**Cost of revenue is one blended line.** No filer breaks out inference. The panel can show what
happened to the total; it cannot attribute any part of it to AI, and no arrangement of these
columns will make it do so. Anyone: including this site: who claims to isolate an AI effect from
these columns is inferring, not measuring.

**The design detects a mean compression of about 2 percentage points.** Simulated at the observed
between-firm spread: 50% power at −2pp, 84% at −3pp, and **17% at −1pp**. A null result here is
not evidence that nothing under a point happened. `scripts/power-and-robustness.py` is the
computation.

**Two diagnostics were built, tested and thrown away.** Both are retained in `scripts/` because a
failed instrument is evidence too:
- `diagnostic-v1-failed.py`: a levels regression of log cost on log revenue. The naive
  "elasticity above 1" reading **fires about 50% of the time on random data**, and the
  significance-tested version **fires 98.9% of the time on a pure accounting reclassification**
  with no change in cost structure at all.
- `diagnostic-v2-failed.py`: the first-difference rebuild. It survives the reclassification
  (3.9%) and then **detects a genuine 30% variable-cost shift only 3.8% of the time**, which is
  below its own false-positive rate. Robust and blind.

Neither is published as a recommendation, because a test that cannot separate an economics change
from a restatement: or that cannot see a real effect: is not a test.

## Reproducing it

```
python3 scripts/build-panel.py      # fetch + panel        (needs a SEC User-Agent with contact)
python3 scripts/analyse.py          # the headline series
python3 scripts/power-and-robustness.py
```

The SEC API is free, unauthenticated and rate-limited to 10 requests/second; it requires a
`User-Agent` naming a contact address. No scraping, no vendor data, no login.