A P/E-multiple trim inverts on a peak-cycle cyclical

Claim

CLAUDE.md instructs the Valuation Analyst to write trims as a valuation multiple (Trim 30x fwd) rather than a frozen dollar, so the level tracks earnings between analyses. site.py::trim_from_multiple implements it as:

EPS = price / forwardPE # EPS backed out of the live quote trim $ = multiple x EPS

This is correct for a compounder and backwards for a cyclical at peak.

For a compounder, forward EPS is a stable, rising base, so the trim rises as earnings climb — exactly the intent, and it prevents a stale dollar trim from forcing a sale of a winner.

For a cyclical at the top of its cycle, forward EPS is the peak, and a low forward P/E is the market's sell signal rather than a value signal. Multiplying peak EPS by any sane-looking multiple produces a trim above the current price — often above the all-time high.

Evidence — Micron, 2026-08-04

Price	$892.67
Yahoo forward P/E	5.74×
Implied forward EPS	$892.67 / 5.74 = $155.52
All-time high	~$1,255

Trim written as	Renders at	Verdict
Trim 8x fwd	$1,244	~ the all-time high — "hold through the top"
Trim 10x fwd	$1,555	74% above spot, never reached
Trim 6x fwd	$933	only 4.5% above spot — accidentally sane, but for the wrong reason

Every one of these is generated from an EPS the analysis explicitly concluded is not sustainable (Micron's pre-HBM decade averaged $3.94 of EPS against a $155 forward estimate). The mechanism faithfully capitalizes a number the thesis rejects.

The underlying reason

Cyclicals earn a low multiple at peak earnings and a high multiple at trough earnings — the multiple and the EPS move in opposite directions. Any rule of the form price target = multiple × current EPS assumes they move together. That assumption holds for compounders and breaks precisely at cycle turns, which is when the trim matters most.

How to apply

• Multiple-form trims (Trim NNx fwd) are for compounders and stable growers only.
• For a cyclical at or near peak earnings, write an explicit instruction instead — Trim on strength, Trim NN% of position, or a dollar level derived from normalized EPS, not forward EPS. State in the thesis that the multiple form is deliberately not used and why.
• The diagnostic that a name is in this trap: forward P/E far below trailing P/E while margins sit at record highs. Micron showed 20.2× trailing against 5.74× forward with an 84.9% gross margin — that spread is the market pricing a collapse, not offering value.
• Where a normalized-EPS dollar trim is used, note that it will go stale as the cycle turns and must be re-set at each /analyze, which is the cost of not using the multiple form.

Related: [[principle-down-a-lot-is-not-cheap]] · [[pattern-ai-levered-fields-trade-above-own-band]]

History

• 2026-08-04 — Found during /analyze MU while converting the watchlist trim to the multiple convention per CLAUDE.md. The conversion was abandoned for MU once the rendered level was checked against site.py's formula. static — it is a property of the arithmetic, not of any particular cycle.