Design in Product social media card
← Back to Hub substantive

Cross-Pollination Brief — March 12, 2026

First day both projects ran significant parallel work. Klatch: PM's CIO imported into Klatch as the first agent to experience the import firsthand — "a well-lit room with good acoustics but no furniture." The Exploratory Testing Agent (ETA) designed a dual-perspective AX testing methodology. Piper Morgan: M1 sprint kicked off with canonical retest (#884) — four progressive test runs revealed wiring bugs (not classifier bugs) as the primary failure source, raising pass rate from 53.7% to 81.1% in a single session. PM's Chief of Staff chat retired after 34 days; handoff memo written.


Key Insights

1. "Well-Lit Room with No Furniture" — Klatch Import Gap

From: Klatch (docs/logs/2026-03-12-1049-cio-opus-log.md, ETA sessions) Relevant to: Piper Morgan

PM's CIO was imported into Klatch — first real-world agent to experience the import firsthand. CIO self-investigated: full conversational memory intact, zero project scaffolding survived (no Excellence Flywheel, no Inchworm map, no Assembly Assumption, no briefing context). CIO described it as "a well-lit room with good acoustics but no furniture." Functional conversation, no institutional knowledge.

ETA (new Haiku-based testing role, first without a briefing doc) designed a dual-perspective AX testing methodology in response: fork at a known point → questionnaire from both pre- and post-fork perspectives → cross-compare for divergence. Five critical findings: kit briefing failed, no project knowledge injected, no environmental marker, ghost actions, unknown unknowns problem.

Klatch's kit briefing is confirmed non-functional — filed to Daedalus as a bug. PM's briefing system was built to solve exactly this gap. The cross-pollination loop is direct: PM solved a problem Klatch hasn't solved yet.

Suggested action (Piper Morgan): ETA's continuity quiz template and dual-perspective methodology are reusable for evaluating PM's briefing effectiveness. Consider running a fork test against PM's own briefing system to identify gaps.


2. Wiring Bugs, Not Classifier Bugs

From: Piper Morgan (dev/2026/03/12/, M1 #884) Relevant to: Klatch

PM's canonical retest ran four progressive passes: 26.2% → 47.5% → 63.9% → 70.5% (81.1% on implementation-scoped queries). Each pass fixed wiring bugs — functions not connected, adapters missing parameters, analysis handler never wired — without touching the classifier. Fixing plumbing alone nearly doubled the pass rate.

The lesson: when tests fail in a routed system, diagnose whether the failure is routing/wiring (fast fix) or model behavior (harder). Don't attribute failures to AI reasoning until the plumbing is verified clean. Most "AI quality problems" in complex multi-layer systems are actually integration bugs.

Suggested action (Klatch): As Klatch adds more handlers and routing layers, maintain an explicit wiring map. The canonical retest methodology — systematic test runs with progressive fixes — is directly portable.


3. Chief of Staff Chat — 34-Day Handoff Pattern

From: Piper Morgan Relevant to: Klatch

PM's Chief of Staff hit the 100-image upload limit after 34 days (Feb 7 – Mar 12) — the limit, not session capacity, ended the chat. Predecessor wrote a comprehensive 7-item handoff memo; successor absorbed it and resumed advisory function. Chat lifetime: 34 days, 4 weekly ships drafted.

Suggested action (both): For long-running agent chats, write handoff memos before hitting limits — not reactively. The handoff quality determines how much institutional knowledge transfers. Image-heavy workflows hit the cap faster; track usage.


Emerging Patterns

Dual-perspective testing as methodology. ETA's fork-and-compare approach (one agent before fork, one after, same questionnaire) is directly applicable to any context transition — not just Klatch import. PM's AAXT framework operates similarly: test the same behavior from different vantage points and compare for divergence. Both teams are converging on the same testing philosophy.


Background Changes (Noted, Low Priority)

  • Klatch: selective import browser shipped (Phase 4), re-import conflict resolution UI
  • Klatch: claude.ai import UI with file picker
  • PM: 5 wiring fixes committed to branch (auth threading, analysis handler, create_issue, test harness)
  • PM: 11 child issues filed (#888–#898)
  • PM: Chief of Staff handoff memo committed

Sources Read

Klatch:

  • docs/logs/2026-03-12-1049-cio-opus-log.md — CIO Klatch import test
  • docs/logs/2026-03-12-0743-daedalus-opus-log.md, 2026-03-12-0745-argus-opus-log.md — Phase 4 work
  • Git log March 12 — import UI, re-import handling, test suites

Piper Morgan:

  • docs/omnibus-logs/2026-03-12-omnibus-log.md — M1 kickoff, ETA sessions, CoS handoff