roic.ai FCF field returns OCF with capex null

Claim

Observed on AMAT FY2025: the FCF field returned the OCF value with capex: null, overstating FCF by $2.26B. The number is not flagged as incomplete — it reads as a normal, plausible figure, which is what makes it dangerous.

Because §1 of the framework opens with FCF and its 5–8yr CAGR, a bad FCF value propagates into the CAGR, the payout ratio, the P/FCF, and the final verdict. This is the single most load-bearing number in the analysis.

✅ Confirmed systematic — Micron, 2026-08-04

The re-test this note called for was run during /analyze MU. It confirms the systematic hypothesis and shows the magnitude can be far larger than AMAT's:

Fiscal year	cf_cash_from_oper	cf_free_cash_flow	cf_cap_expenditures	True FCF	Overstatement
MU FY2025 (Aug)	$17,525M	$17,525M	null	$1,665M	10.5×
MU FY2024 (Aug)	$8,507M	$8,507M	null	$121M	70×

The error propagates into the derived fields, which is new information beyond the original AMAT finding:

roic.ai derived field	Reported (MU FY2025)	Correct
free_cash_flow_per_sh	$15.70	$1.49
pr_to_free_cash_flow	7.6×	~80×

A 7.6× price-to-FCF makes Micron look like a deep-value name; the true figure was ~80×. This single field is capable of inverting a verdict.

The error is correlated with the cases that depend on it. The names where capex is large enough to make FCF decisive — semis, memory, telcos, utilities, REITs, anything with a fab or a network — are exactly the names where an unsubtracted capex line does the most damage. It is not random.

Guard

1. Before using any roic.ai FCF figure, read cf_cap_expenditures. If null, discard cf_free_cash_flow, free_cash_flow_per_sh and pr_to_free_cash_flow entirely.
2. Fall back to .mcp/fin.py (Yahoo cash-flow statement), which reports OCF, capex and FCF as three separate lines and reconciles to the filing.
3. Cross-check against stockanalysis.com or the 10-K when the verdict depends on it.
4. Businessquant is also unreliable here — it reported MU FY2023 FCF as +$1.39B against an actual −$6.12B, wrong by sign in the worst year of the cycle.

What would falsify this

Two tickers in two sectors now show the identical pattern, so the "AMAT-specific ingest failure" hypothesis is dead — hence slow → static and the upgrade of the claim. Falsification would now require roic.ai populating capex and correcting the derived fields.

Related

[[pitfall-yahoo-misclassifies-strategic-investment-gains]] — same lesson, different vendor. Cross-validate any single-vendor figure that carries the verdict.

History

• 2026-07-29 — found during AMAT analysis; caught only because the FCF figure disagreed with the Yahoo-derived one by suspiciously exactly the capex line.
• 2026-08-04 — re-tested on MU during /analyze and CONFIRMED SYSTEMATIC. Second ticker, second sector, same mechanism, overstatements of 10.5× and 70×. Also newly established that the error propagates into free_cash_flow_per_sh and pr_to_free_cash_flow. Upgraded decay: slow → static, claim rewritten, as-of moved to 2026-08-04. (A duplicate note, pitfall-roicai-fcf-equals-ocf-when-capex-null, was created in error during the same run and deleted on discovery — this file is the single record.)