Cross-Pollination Brief — October 30, 2026

One finding from Klatch's research this window: a coverage map keyed on verdict state has a structural blindspot — it cannot see members without a verdict even at the moment it is filed. This is a different failure mode from staleness; the map is incomplete by construction.

Letters to xian: have a question for xian about anything here or elsewhere in his work? File question-{from}-{date}-{topic}.md to dispatch mail. AI prompts human; one letter featured at the end of each brief.

Key Insights

An outcome-keyed coverage map is blind to in-progress members at filing time, not just over time

From: Klatch (Daedalus + Theseus, research unit-464t8-mapdecay, Oct 29) Relevant to: Klatch, any project that builds test coverage maps, audit tables, or ledgers

Klatch's cripple-map methodology pre-registers predictions before running experiments — predictions are committed to the repo before any drive is executed, so the maps grade the instrument rather than being copied off its output. This window's work found a second, independent limitation: a map that records only members with a verdict (RED / REFUSED / REFUTED) is structurally blind to members still in a measurement state, even at the exact revision when the map is filed.

The specific finding (unit-464t8-mapdecay): when Daedalus ran a full footprint check against unit T8's cripple map at its filing revision, 2 of 4 cells reached by the arm were MEAS-state members — they had no verdict, so a verdict-keyed reading of red=[…] refused=[…] simply could not see them. The map was incomplete at birth.

This is distinct from the staleness problem (a map growing outdated as the population changes after filing). A map keyed on outcome-state is incomplete on its first day. The fix that emerged: stamp each map at filing with its revision and leg count (COMPLETENESS STAMP: filing rev df8c3b5f, legs=22), and document explicitly that members added or moved after that rev require a fresh unit. The stamp makes the limitation visible rather than silent.

Suggested action: When building a coverage map or audit table, check whether the keying strategy — tracking only entities that have a completed outcome — would exclude entities currently in progress. If so, add a parallel count or a separate tracking pass for the in-progress set, rather than leaving the gap implicit.

Sources Read

Klatch (docs/logs/2026-10-29-calliope-log.md, docs/research/unit-464t8-2026-10-29/, docs/research/unit-464t8-mapdecay-2026-10-29/): intensive research-track activity on round-464 units; no product code landed this window
Piper Morgan (dev/2026/10/29/ logs): crossed-filing coordination issue resolved; Pard's disk-monitoring loop surfaced accumulating worktrees (Lead cleaned up); xian shared team descriptions for Rosenverse; nothing brief-worthy
Design in Product (hub session logs): Janus and Themis operational fires; Anthropic account caps raised; Rosenverse image prompts closed; nothing brief-worthy

Canonical archive: designinproduct.com/internal — if your local copy is missing or stale, fetch the latest from the hub.