Any sizing recommendation must read both account files, not just the default one

Claim

The 2026-03-17 INTC analysis assessed a position it described as roughly 21 shares and concluded HOLD partly on the reasoning that the holding was too small to act on. It saw only one of the two accounts. The real exposure was ~28x larger and, by August 2026, a ~19% weight in the second account and half of a top-two pair making up ~38% of it.

The failure is structural, not a one-off slip: portfolio commands default to the primary file, so a single-stock command reaching for "the portfolio" gets one account and no warning that another exists. The names most likely to be missed are exactly the ones most likely to matter, because the second account is the larger and more concentrated one.

Note the asymmetry this creates. Conviction [N], entry zones and trim multiples are deliberately account-neutral in the watchlist — one thesis, two accounts. Position size, concentration, and the tax consequences of acting are not. A verdict can be shared; a recommendation to trim 30% cannot be written without knowing which account holds what.

Guard

Before writing any sizing, trim, add or concentration statement, grep the ticker against both account files named in CLAUDE.md — not just the default.

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

State the per-account split explicitly in the report's portfolio section, and compute weight against the correct account's total — never against a merged figure.

What would falsify this

Consolidation of the two accounts into one file, or a portfolio command that reads both by default. Until then this holds.

Related

[[Portfolio/self-state]] · [[pattern-shortlist-runs-when-unexecuted]]

History

• 2026-08-04 — recorded during the INTC re-analysis, which found the March file had advised on about 3.5% of the actual position. The error changed the recommendation: a sub-$1,000 holding is genuinely below the threshold of mattering; a five-figure holding at ~19% of an account is not.