# Anyone building a value investing agent? what are you using for data — retained record

> Retained reddit record from reddit:ValueInvesting.

Canonical: https://highsignal.app/data/records/859ece93c8af5009

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ValueInvesting)  published 2026-10-13 17:06 ingested 2026-10-14 02:44

[original source ↗](https://www.reddit.com/r/ValueInvesting/comments/1x52d8o/anyone_building_a_value_investing_agent_what_are/)

## retained text



```
been messing around with an LLM agent that reads 10-Ks and does the boring stuff: normalized earnings, ROIC, owner earnings, a rough reverse DCF. honestly getting the model to reason is the easy part. data is what’s killing me

tried pulling straight from EDGAR / XBRL companyfacts first. free and it’s the actual source, but every company has its own custom tags and segment data is a nightmare to clean up

then tried the usual APIs (FMP, yfinance etc). way easier but i can’t tell where a number actually came from. if it’s wrong the agent just runs with it and writes me a very confident thesis on garbage lol

also most of the stuff that actually matters for value work (leases, SBC, segment margins) is buried in the footnotes and i haven’t found anyone who has that structured

so for people who’ve built something like this:

* where’s your data coming from? paying for anything?
* do you feed it raw filings, structured numbers, or both?
* anyone dealt with point-in-time / restatements for backtesting? my backtests feel too good to be true and i’m pretty sure that’s why

curious what’s actually working before i waste another weekend writing parsers
```



document fetched 2026-10-14 00:44 · canonical https://reddit.com/r/ValueInvesting/comments/1x52d8o/anyone_building_a_value_investing_agent_what_are
