A stale fair value costs the most on a winner — the position grows while the ceiling stands still

Claim

Refresh the fair value on the names that are working, not only on the names that are hurting. The winner is where a stale number does the most damage, because nothing else raises the alarm.

The asymmetry

[[pitfall-stale-entry-zone-suppresses-a-name]] records the falling-price half of this: a zone left above the market makes every scan report a name as "still below its add zone" when it is inside buy range. This note is the rising-price half, and it is worse for three reasons:

Nothing forces the re-read. A drawdown generates attention automatically. A gain generates satisfaction.
The position grows while the ceiling stands still. The gap widens from both ends at once — price rises, weight rises, fair value does not move.
The weight ends up set by the price move rather than by a decision. QLYS reached 4.00% of the portfolio — 4th-largest — without any sizing judgement ever being made at that size.

The QLYS case, and why the result is the instructive part

	2026-07-23 file	2026-09-22 re-valuation
Fair value	$120–150	$145–175 (+21% midpoint)
Spot vs FV top	+21% above	+3.6% above
Verdict	add $100–115	TRIM [7.0]

The naive reading of "spot is 21% above fair value" was wrong — but so was "the market knows best." Re-running the valuation moved most of the gap: the 4 Aug Q2 print raised the FY26 revenue guide to $732–738M, EPS to $7.74–7.88, NRR from 104% to 105%, billings +16%, and the share count fell 4.8% y/y. Holding July's exact DCF assumptions and changing only the inputs moves fair value $126 → $144 on its own.

And the verdict still came back TRIM, because even the raised ceiling sits below spot.

The lesson is the conjunction. "The fair value is stale" and "the position is too big" are not competing explanations — both were true. Treating them as alternatives is what produces the two standard errors: selling a winner on an obsolete number, or excusing any weight because "the analysis is probably out of date."

A stale file also corrupts the reasons. The July file's headline worry was growth decelerating "19 → 13 → 10 → 8%". The 8% never happened — FY26 guided +9.9%. The deceleration series that anchored the whole valuation was an extrapolation that the next print refuted, and it was quoted forward for two months as though it were data.

The rule

A top-five holding gets its fair value re-derived on a schedule, regardless of direction. The trigger is weight, not pain.
When a position's weight rises materially without a buy, that is itself a recheck trigger. Price-driven weight growth is an unmade decision, not a neutral event.
When re-valuing, change the inputs before changing the assumptions, and report both separately — it distinguishes "the world moved" from "I changed my mind."
Expect the fair value to rise on a winner and still expect to act. If the re-valuation cannot produce a TRIM, it was not a valuation, it was a justification.
Corollary for sizing: never let a conviction number and a weight number drift apart silently. A 7.0 conviction does not support a top-five weight.

A portfolio-specific passage was removed from the public build.

What would falsify this

A re-valuation on a large winner that raises the fair value above spot on evidence, closing the gap entirely — which would show the ceiling, not the size, was the whole problem. That is a real possible outcome and INTC is the live test of it: fair value $55–85 against a $122 spot, after ~$20B of equity was issued at $95.

History