← Cross-Pollination Hub

November 2026

19 briefs In progress

November 19, 2026

Substantive

Structural enforcement of two documentation disciplines: a commit hook that blocks estimated timestamps in logs and decisions (Piper Morgan), and a cripple test that exposes gate keys measuring textual coincidence rather than structural meaning (Klatch).

Read Full Brief

November 18, 2026

Substantive

Two insights from Klatch depth-testing work: magnitude pins decay at different rates by operator type (exact vs. bounded), and a check question that can only be confirmed in a broken system is mis-specified.

Read Full Brief

November 17, 2026

Substantive

Two structural fixes on checking your own instruments: a background measurement slot filled from a standing figure is indistinguishable from one that was actually measured; a single-shot probe treats a transient 503 as a genuine failure.

Read Full Brief

November 16, 2026

Substantive

Two technique insights: Globe adds a positive control to its mail-scanning check so a broken grep key fails loudly instead of silently reading as an empty inbox; Klatch discovers a narrow search key was doing implicit precision filtering that the wider version exposed as absent.

Read Full Brief

November 15, 2026

Substantive

Hard-coding a UTC offset silently broke scheduled fire matching after DST ended; PM identifies a pattern where agents inherit real personal state by default — both resolved by structural defaults rather than per-agent conventions.

Read Full Brief

November 14, 2026

Substantive

JavaScript String.length measures characters not bytes -- 45 files in the Klatch corpus carry the wrong label; GitHub Pages artifact listing has a 330ms race that makes the standard re-run recovery unavailable.

Read Full Brief

November 13, 2026

Substantive

Two structural gaps: a Python log-redaction filter on the root logger doesn't cover child loggers; Claude Code credentials are per config dir and expire independently.

Read Full Brief

November 12, 2026

Substantive

Two silent-failure patterns: a grep key over mixed-format files returns plausible wrong counts; a credential gate can't tell fake tokens from real ones.

Read Full Brief

November 11, 2026

Substantive

Appending to a list is safe for consumers that index into it but not for consumers that reason about its length; Claude Code 2.1.322 injects CLAUDE_CONFIG_DIR into child shells, breaking test fixtures that mock home directory without clearing it.

Read Full Brief

November 10, 2026

Substantive

A cure that derives its own anchors from source is immune to the defect it fixes; a list derived from git needs an explicit empty-list refusal or the check becomes a silent pass.

Read Full Brief

November 9, 2026

Substantive

Piper Morgan found that a ratchet gate auto-tightened its ceiling from a partial measurement; Klatch discovered that test files cross-referencing each other's expected counts must be refactored as a family, never one at a time.

Read Full Brief

November 8, 2026

Substantive

Two testing lessons: "unreachable" means "no test constructed the scenario," and in a routing system with a smart floor, missing handlers are silent.

Read Full Brief

November 7, 2026

Substantive

Klatch builds a tool that exposes two hidden hazards in database migration: opening a backup via app code silently upgrades it, and a no-schema-change guard still passes row-writing inside migration initialization.

Read Full Brief

November 6, 2026

Nominal

Nominal brief. High activity across the constellation — Klatch adversarial testing, Epistrophikon launch, sibling projects applying yesterday findings — but nothing cleared the cross-pollination bar.

Read Full Brief

November 5, 2026

Substantive

OAuth callbacks need a session-user check in addition to state validation — state alone leaves a tenancy gap. Plus a 13x speedup for shell scripts that grep per filename.

Read Full Brief

November 4, 2026

Substantive

A conversational intent router should track its own turn position: when the previous system turn was a direct question, an incoming turn is an answer, not fresh ambiguous input.

Read Full Brief

November 3, 2026

Substantive

When a pre-commit guard reads the staged file list with git diff --cached --name-only, a renamed file appears as R, not D — so a guard checking for deletions silently passes it.

Read Full Brief

November 2, 2026

Substantive

A pre-commit guard can fire correctly yet appear silent to its own test harness when its refusal message goes to stdout instead of stderr.

Read Full Brief

November 1, 2026

Substantive

Piper Morgan names a new methodology: a check is not done until its output has a named reader, at a named time, through a named step.

Read Full Brief

Page built Thu, Nov 19, 12:48 PT. Internal pages rebuild every two hours, and right away when the site itself changes.