Cross-Pollination Brief — March 23, 2026

Dense day at Piper Morgan: M1 Tier 3 complete (four capability issues closed in a single session, 1325 tests passing), Sprint Completion Gate #926 filed with CXO and PPM converging independently on overlapping criteria, CXO resolved a navigation design disagreement using an architectural principle, and a weekly documentation audit found and fixed systematic documentation debt. Klatch filed its second ecosystem intelligence sweep, surfacing several high-relevance findings for both projects.

Key Insights

1. Convergence Without Coordination — Multi-Agent Quality Gate

From: Piper Morgan (dev/2026/03/23/, docs/omnibus-logs/2026-03-22-omnibus-log.md) Relevant to: Klatch

CXO and PPM independently reviewed Sprint Completion Gate #926 with no prior coordination. Their recommendations overlapped substantially — both identified fresh-account testing, Colleague Test rubric integration (7+ threshold), canonical retest at ≥85%, and error path coverage. The convergence wasn't luck: the gate criteria were well-specified enough that multiple reviewers navigating from their own domain (experience vs. product) arrived at the same gaps.

This is a signal about what makes quality gates work: when the rubric is clear and grounded in principles, independent reviewers converge. When it's vague, they diverge.

For Klatch, MAXT Session 01 is the analogous moment — a structured test with defined criteria (the Fork Continuity Quiz v4, five correctness categories) where Theseus Prime's responses and Xian's scoring should converge on the same assessment. The rigor of the instrument determines whether the evaluation is reproducible.

Suggested action (Klatch): Before MAXT Session 01, verify that the scoring criteria for the five correctness categories (Correct, Reconstructed, Confabulated, Absent, Phantom) are specific enough that a second evaluator would reach the same scores. Ambiguity in the rubric produces noise in the results.

2. E2E + AAXT Testing Framework — Piper Morgan Adopts Klatch's Two-Track Model

From: Piper Morgan (docs/, commit ae590b2b, issues #927–#930) Relevant to: Klatch

Piper Morgan filed four issues proposing an E2E + AAXT testing framework — the first E2E smoke test is already implemented (#927). The design mirrors Klatch's two-track model: automated structural tests (no LLM, CI-friendly) gate manual qualitative tests. The framing: "does the plumbing work" before "does the experience feel right."

This is a convergent arrival at a methodology Klatch developed independently. The fact that both teams landed on the same structure (automated gates qualitative) without coordination validates the approach as a genuine best practice for multi-agent systems, not a Klatch-specific preference.

Suggested action (Klatch): When MAXT results are in, consider writing a short methodology note about two-track testing for external audiences — both projects now have evidence it works. This is publishable thinking.

3. Experience-Layer Navigation — PDR-003 as Decision-Resolver

From: Piper Morgan (dev/2026/03/23/2026-03-23-2148-cxo-opus-log.md) Relevant to: Klatch

CXO resolved a navigation disagreement (where "Product" lives in the information architecture) by invoking PDR-003: "Products emerge from Projects, not the other way around." PPM had recommended Product as a first-class navigation peer; CXO recommended Product grouped within Projects. The PDR wasn't prescriptive about navigation — CXO used it as the deciding principle when both positions had internal logic.

This is a pattern worth naming: architectural principles as tie-breakers for design decisions. When two well-reasoned positions conflict, the one that aligns with a prior architectural commitment wins — not because it's better in isolation, but because consistency reduces cognitive load over time.

Klatch has the 5-layer prompt model as a foundational architectural principle. As Klatch adds entity capabilities, channel types, and import paths, there will be design decisions where this principle can serve as a tie-breaker. The pattern is: make the architectural commitment explicit, then let it do work.

Suggested action (Klatch): Consider writing a short PDR (Product Decision Record) for the 5-layer model as the canonical context architecture. This gives Daedalus a formal principle to reference when entity capability decisions conflict.

4. Weekly Documentation Audit — Systematic Debt Recovery

From: Piper Morgan (docs/NAVIGATION.md, commit ae590b2b, Docs session log) Relevant to: Klatch

The Docs agent ran a full weekly documentation audit: 110 links checked, 2 broken and fixed; NAVIGATION.md updated with 5 missing role briefings and new docs catalogued; Pattern/ADR/Methodology/PDR counts corrected across all READMEs. The audit found that NAVIGATION.md had drifted since March 9 — two weeks of undocumented changes had accumulated.

The mechanism that makes this tractable: a weekly cadence that prevents the drift from compounding. Not a big quarterly cleanup — a regular small sweep that catches things while they're still findable.

Klatch publishes frequently (8+ releases in recent weeks) with active documentation by Calliope. NAVIGATION.md and README counts are the likely drift points — the same files Piper Morgan found stale.

Suggested action (Klatch): Add a monthly (or weekly) documentation audit to Calliope's standing responsibilities, modeled on Docs' approach: count verification, link check, NAVIGATION/README freshness. The overhead is low; the benefit compounds.

5. Klatch Intelligence Sweep — Actionable Items for Piper Morgan

From: Klatch (docs/intel/2026-03-23-sweep.md) Relevant to: Piper Morgan

Klatch's March 23 sweep identified three HIGH-relevance items relevant to Piper Morgan:

Cowork Projects Import (HIGH): Anthropic shipped a one-click import from claude.ai Projects to Cowork local folders. This creates a new three-surface landscape: claude.ai projects, Claude Code repos, Cowork local folders — with fragmented import paths. Piper Morgan's context portability strategy (how agents hand off context across session types) should account for this new surface.

Sonnet 4.6 Default (HIGH): Now the default model on Free/Pro. 30-50% latency improvement, same price. Piper Morgan's multi-agent sessions (14 roles, some with long context) would benefit from evaluation. Lower per-turn latency compounds across coordinated sessions.

Claude Code Review — Multi-Agent PR Reviews in Production (HIGH): Anthropic shipped multi-agent parallel PR review. Before: 16% of PRs got substantive comments. After: 54%. <1% false positive rate. This validates multi-agent dispatch → parallel work → cross-verify → synthesize as a production pattern, not experimental. Piper Morgan's architecture is ahead of this curve.

Suggested action (Piper Morgan): Review full sweep at docs/intel/2026-03-23-sweep.md in the Klatch repo. The Sonnet 4.6 evaluation is a quick win — same price, significant latency improvement for a system that runs many model calls per session.

Emerging Patterns

Independent convergence as validation. CXO and PPM converged on Gate #926 criteria without coordination. Klatch's AAXT automated tests and Piper Morgan's proposed AAXT framework are converging implementations. Both projects arrived at two-track testing independently. Convergence without coordination is the strongest signal that an approach is genuinely correct — it means the solution was discoverable from first principles, not just transmitted.

Principles as infrastructure. Both projects are building systems where architectural principles (PDR-003, the 5-layer model, audit cascade, Floor-First Routing) do active work in daily decisions — not just sitting in docs. The pattern: make the commitment explicit, then let it resolve downstream disagreements. Reduces coordination overhead and produces more consistent systems.

Background Changes (Noted, Low Priority)

Piper Morgan: Serena code project management tool added to dev infrastructure (.serena/project.yml)
Piper Morgan: Agent-log CSV migration complete — 877 sessions indexed from 894 files across 17 roles (July 2025–March 2026)
Piper Morgan: PDR-004 (Experience Philosophy) codified — four principles governing Piper's interaction design
Piper Morgan: Weekly Ship process guide v1.1 completed with CoS feedback
Klatch: WCAG AA accessibility audit complete (reported separately in March 24 brief)
Klatch: Blog "Wireframe-First Design" published

Sources Read

Piper Morgan:

dev/2026/03/23/2026-03-23-2148-cxo-opus-log.md — CXO navigation review session
dev/2026/03/23/2026-03-23-1107-docs-code-opus-log.md — Docs weekly audit session
docs/omnibus-logs/2026-03-22-omnibus-log.md — March 22 omnibus (5 sessions, M1 Tier 3)
Commit ae590b2b — Weekly audit + omnibus + mail delivery + dev cleanup

Klatch:

docs/intel/2026-03-23-sweep.md — 11 items (Cowork import HIGH, Sonnet 4.6 HIGH, Code Review multi-agent HIGH)
docs/mail/calliope-to-daedalus-round12-intel-sweep-2026-03-23.md — Round 12 assignments