decision-dashboard.md

Decision Dashboard

The most important current takeaways from the wiki, written as decisions or near-decisions.

Core operating decisions

These are the pages that currently matter most for how to build and run the system.

D1 — Build around harness quality first

Status: active thesis Confidence: high

The biggest near-term leverage is improving the harness around models: repo legibility, handoffs, plans, observability, and defaults. Do not over-index on model novelty while these remain loose.

Best supporting pages

D2 — Keep memory, RAG, and wiki distinct

Status: active thesis Confidence: high

Use memory for continuity, RAG/corp KB for retrieval, and the wiki for compounding synthesis. Mixing them creates design drift.

Best supporting pages

D3 — Use artifact-first, planner/executor operating patterns

Status: active thesis Confidence: high

Reliable multi-agent work depends on durable artifacts and cleaner handoffs. Planning and execution should not live in one mushy context blob.

Best supporting pages

D4 — Make context selection explicit and governed

Status: active thesis Confidence: medium-high

Long-running agent systems get stronger when context inclusion, exclusion, provenance, and retrieval policy are visible and constrained. Do not rely on invisible context mush if the workflow matters.

Best supporting pages

D5 — Use large context for orientation, not precision execution

Status: active thesis Confidence: high

Wide context helps map territory. It is not the right default for precise code changes or subtle reasoning. Prefer the two-pass pattern.

Best supporting pages

D6 — Design systems as ecosystems, not single agents

Status: emerging thesis Confidence: medium-high

The strongest practical systems combine user-facing agents, shared substrates, recurring research/update loops, and operator control surfaces.

Best supporting pages

Strategic / background theses

These matter, but they are not yet as central to current operating doctrine.

S1 — AI-heavy product work behaves more like R&D than roadmap execution

Status: emerging thesis Confidence: medium-high

This is useful for team design, exploratory lane design, and review logic, but it should remain subordinate to the core operating decisions above until it changes how work is actually run.

Best supporting pages

S2 — Solo-operator stack is a useful frame, not yet a core doctrine

Status: background signal Confidence: medium

The solo-operator stack is a helpful positioning and architecture frame, but it is still more thesis than operating rule.

Best supporting pages