Changelog

What changed, and when

Written by the people who made the change. Fixes are listed with the same weight as features, because a fix is usually what somebody was waiting for.

Version 1.4

July 16, 2026

Founder questions, in priority order

Every memo now ends with a ranked set of questions to ask the founder, each with a one-line reason it matters. The sharpest question sits at the top.
Questions now pull the specific number or name from the deck where there is one, so a question reads as "Brookline is 22% of ARR, what is the renewal date" rather than a generic prompt.
Fixed a case where a memo with no clear ask still produced a round-size question. It now asks for the ask instead of inventing one.

Version 1.3

June 3, 2026

Decks and data rooms up to roughly 60,000 characters, about fifty pages, are now read in a single pass, so a memo reasons across the deck and the exhibits together.
Long or self-contradicting documents now escalate to a frontier model automatically. The routing table on How the AI works is generated from the code and shows the escalation rule.
Evidence quotes that cannot be located in your own document are dropped rather than displayed. If any were dropped, the memo says how many.

Version 1.2

April 22, 2026

Every memo now carries a metrics table: revenue, growth, retention, margin, CAC and the round, taken verbatim from the deck where they exist.
A number that is missing is now surfaced as a founder question rather than filled in with a guess.
Risk severities were recalibrated so a clean company comes back with fewer high flags, because a memo that flags everything is a memo nobody reads twice.

Want something on this list? Tell us. Most of what ships here started as an investor saying a memo missed something.