What companies told investors to expect, set against what they then reported, read from the quarterly results releases they filed with the SEC. Each of the 2,334 rows is one measure for one quarter: the range or figure guided in one release, the result in the next, and where the result landed. Every row links the filing it came from.
Download
What is in it
- The 500 companies the Vanguard S&P 500 ETF held, 498 of them read: 5,929 results releases, filed from 2023 to 2026.
- 140 companies have at least one matched line. The rest guide in words, on a call or not at all, or the reader could not match a guide to a result, so companies that publish a number and report the same measure next quarter are over-represented.
- 20 measures in all. Those with at least a hundred lines:
These counts take every line as it is. How often companies beat their own guidance counts each company's quarters once and sets ranges apart from single figures.
Load it
curl -O https://canlicapital.com/datasets/company-guidance/guidance-lines-2026-11-02.csv
shasum -a 256 guidance-lines-2026-11-02.csv
python3 -c "
import csv
rows = list(csv.DictReader(open('guidance-lines-2026-11-02.csv')))
print(len(rows), rows[0]['company'], rows[0]['measure'], rows[0]['position'])"
How it was made, and what uses it
- Read with
guidance_record in canli-mcp, free and open source, which matches each guide to the next result by measure, basis and unit.
- beat_probability, the first Canli model, is fitted and tested on it: the chance a company's next result lands above, within or below its own guide.
Limits
- A hand read of a random sample of 30 lines against both quotes found 29 right. Other errors will remain.
- Only what the releases print: guidance given on a call or in a slide deck is not here.
- A record of the past. It is not a forecast and not investment advice.