Metrics and Narratives in Portfolio Research Notes

Why portfolios invite mixed language

Holdings summaries, fact sheets, and internal memos often weave numbers into a story arc. Readers experience fluency; editors lose the seam between a defined metric, a restated claim, and an interpretive gloss. InvestCAD treats that seam as the primary object of portfolio research support: not what to hold, but how the note names what it knows.

Three channels that must stay labeled

• Measured fields — values with a definition, unit, and capture or reporting date.
• Quoted claims — language taken from issuers, managers, or third parties, attributed and dated.
• Analyst narrative — interpretation that must be marked as such and never presented as a measured field.

When a paragraph collapses all three, a later reader cannot tell whether a figure was observed, paraphrased, or inferred. That ambiguity is how soft advice language enters “research” notes without anyone intending it.

Definitions before comparisons

Comparing two holdings narratives is meaningless if “duration,” “exposure,” or “concentration” are used differently across sources. A short glossary block at the top of a note — even if incomplete — beats a polished comparison that silently switches definitions mid-page.

Working with assisted extraction

AI assists are useful for pulling tables, normalizing synonyms, and flagging repeated phrases across documents. They are poor at inventing missing metrics or “completing” a holdings picture. Prompts should ask for definition candidates and missing fields, not for attractiveness or implied positioning.

A portfolio research note that lists empty cells is often more honest — and more useful — than a note that fills every cell with fluent prose.

Documentation gaps as findings

Gaps are findings. If a public document omits a definition, the note records the omission with the capture date. Teams sometimes fear that empty cells look unfinished. In opaque domains, a document without gaps usually smoothed them away.

1. Inventory metric names as they appear in sources.
2. Attach definitions or mark “undefined in source.”
3. Separate narrative adjectives into a labeled interpretation block.
4. Reject any draft sentence that implies a buy, sell, or sizing instruction.

Closing checklist

Before circulation, ask: Can a second reader see which lines are measured, quoted, or interpretive? Are capture dates present? Is allocation language absent? If any answer is no, the note is not ready — regardless of how polished it reads.