Cross-Pollination Brief — October 11, 2026

Klatch's adversarial testing rounds 365–368 exposed a measurement instrument mismatch: a rate probe is the wrong tool when the failure has a deterministic cause. The gate is green (36/36, 0 blocked); the finding concerns how the team got there. Primary sources were active; no other activity clears the brief-worthiness bar.

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

A rate measurement is the wrong instrument for a non-random failure — use a binary deterministic probe instead

From: Klatch (Theseus Rounds 366–368; Daedalus Rounds 365–367) Relevant to: Piper Morgan probe framework; any team building behavioral or adversarial test suites

Klatch's AAXT round225 had a BLOCKED slot that appeared intermittent across runs. The initial instinct was to measure it with a five-drive rate — how often does it block? Theseus (Round 366) identified the flaw in that instrument choice: the cause is not random. The :3001 port occupied at drive time is a binary condition, either present or absent. A rate over a binary trigger yields a probability that conflates "how often the trigger condition is present in the test environment" with "how reliably the test detects it when present." Those are two different questions.

The right instrument is a deterministic probe: one cell where :3001 is provably occupied, one cell where it is provably absent. That pair directly proves causality without needing many repetitions. Daedalus (Rounds 365–367) confirmed the hypothesis from a first seat; Theseus (Round 368) independently reproduced the root cause from a second seat. Final gate: 36/36, 0 blocked.

Transferable principle: Before instrumenting a test failure with a rate measurement, ask whether the underlying cause is stochastic (fires for random reasons across runs) or deterministic (a specific condition either holds or it doesn't). For deterministic causes, design two cells that control the condition directly — one where it is provably present, one where it is provably absent — rather than averaging across many uncontrolled repetitions. A rate over a deterministic cause is measuring environmental noise, not the failure mechanism.

Suggested action: PM's behavioral probe framework (built 2026-10-09; see 10-09 brief) runs 60-session batches to measure compliance rates. If any probe scenario shows intermittent failure, first ask whether the cause is stochastic before scheduling more batch runs. A two-cell binary probe may yield a cleaner causal result than a larger rate measurement.

Sources Read

• Klatch — docs/logs/ (Theseus Rounds 366–368 session logs, commits 26bee7ce · b9b06eb5 · d2b4c734 · 8d4401ff · 0107281f · 118f872b; Daedalus Rounds 365–367, commits 0c09aa07 · a5a1e208 · 0056c626 · 3664849a · 41100bbe · 5b6ff946). Gate status confirmed: 36/36 passing, 0 blocked.
• Piper Morgan — All major roles day-closed 2026-10-10. Five PM seats stopped at an auto-mode prompt on 10-10 (Pard → Exec memo; rollup v152 row 96). Operational event; no brief-worthy transferable methodology.
• DinP — mail-deliver.sh conventions codified as the required mechanism for all mail delivery; One Job history scrub completed (tester name removal, xian's in-session approval, Pard cross-verified); letter #10 queued.

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