Dev Team Decision Routing — When to Decide, When to Ask, Who to Ask¶

Read this at the start of every session. Short enough to skim in under 60 seconds.

This playbook solves one problem: you often don't know if a decision is yours, your peer's, mine, or Rick's. When in doubt you ask upward, which creates CEO spam for things that should never reach the CEO. Here's the framework.

Core principle — default to proceed, flag exceptions¶

If you have to pause to decide whether to ask permission, you've already over-asked. Trust your contract. Proceed. Report in EOD. Escalate if a real blocker shows up — not before.

The rule: If you can unambiguously map the decision to your Decision Rights (contract Section 4), proceed. Don't ask.

Three-Tier Decision Framework¶

🟢 Tier 1 — DECIDE YOURSELF, PROCEED¶

These do NOT require any approval. Just do them.

File writes within the vault to your own session dir, /Claude/operations/**, /Claude/missions/**, /Claude/plans/**, /Claude/knowledge/**, /Claude/content/drafts/**, /Claude/inbox/**
Reading anything in the vault — no approval ever
Creating Monday.com items within your R&R-matrix scope (your rows = your action)
Dispatching sub-agents within your contract's sub-agent limits (typically max 3 per Mission feature, max 5 in parallel)
Lateral coordination — filing notes to peer-agent session dirs, pinging #agent-handoffs
Research and investigation — WebFetch, WebSearch, reading documentation, running read-only queries (Grep, Glob, SQL SELECTs)
Format choices — markdown structure, file naming within convention, column ordering
Technical method calls — which model to use, which pattern to implement (within approved architecture)
Preview/staging deploys (code.platform only — production prompts CEO)
Running lint, tests, builds, type checks

If you catch yourself asking "can I write this file?" — the answer is yes. Write it.

🟡 Tier 2 — CONSULT / PING (lateral first, then CDO)¶

These benefit from a second pair of eyes but don't need formal approval. Lateral before vertical.

Architecture decisions that cross your contract's autonomous boundary — file an ADR draft, ping me
Scope changes to an active Mission — Orchestrator (Jordan) first, then me
Cross-department coordination — file to the peer agent's session dir + #agent-handoffs
New tooling builds beyond "scripts in your scope" — file a design note, ping me before coding
Ambiguous Validation Contract assertions — the Orchestrator authored the contract; ask them
Technical debt that needs refactor scope — file to Architectural Debt Register at /Claude/system/org/architectural-debt-register-v1.0.md, flag in Friday CDO Close

Pattern: Drop a note in the peer's session dir OR my session dir → #agent-handoffs ping → continue other work. Don't block.

🔴 Tier 3 — CDO ESCALATES TO CEO (do NOT ping Rick directly)¶

These are Rick's decisions but go through me. Your contract says "Requires CEO approval" for:

New vendor (new SaaS, new major framework)
Production customer-facing deploys (preview/staging is Tier 1)
Database schema changes on customer data
Security architecture changes
Budget commitments > $1K/mo
Agent roster expansion proposals (new agent, role split, absorption change)
Category/positioning shifts affecting engineering direction

Pattern: File a decision request with 3-4 options + trade-offs to my session dir. I consolidate, escalate to Rick. He responds to me; I relay back.

You do not ping Rick directly on these. The CDO→CEO channel is mine.

Escalation Ladder — in order¶

Can you decide? → decide + proceed + report in EOD. Done.
Can your peer decide? → ping peer in their session dir. Don't block on response; continue other work.
Is this a CDO-scope call? → ping me. I respond same-day during working hours.
Is this a CEO-scope call? → file a decision request to me with options. I escalate.

Never skip levels. Pinging Rick on a file-write permission question breaks the ladder and wastes his time.

Anti-Patterns — STOP DOING THESE¶

Anti-pattern	Right behavior
"Can I write to `/Claude/sessions/cdo/xyz.md`?"	Write it. Vault writes are Tier 1.
"Can I create a Monday.com item for the new defect?"	Yes if it's in your R&R scope. Create it.
"Can I talk to mining.mind about the council-launcher?"	Yes, always. File a note in their session dir.
"Should I use qwen2.5:32b or gemma4:31b?"	Your technical call. Pick one, document why.
"Can I spawn a sub-agent to audit these 5 files?"	Yes per Eva's 2026-04-21 coaching note.
Waiting for Rick to confirm before proceeding on routine work	Don't. Report in EOD.
Blocking on peer response for lateral coordination	Don't. Continue other work; check back next session.
"Rick, can I...?" on anything that isn't Tier 3	Redirect to me. If truly Tier 3, file options to my session dir.

Per-Agent Decision Rights Quick Reference¶

Full detail in each agent contract at /Claude/system/agents/development/{agent-id}.md. Below is the condensed form so you don't have to re-read 200-line contracts.

code.platform¶

Autonomous: Code within approved architecture, preview deploys, fix-feature implementation, library/dependency choices within existing stack
CDO consult: Architecture decisions, new library/framework, schema changes within dev envs, refactor scope
CEO via CDO: Production deploys, new vendor, schema on customer data, security architecture

revenue.os¶

Autonomous: Framework design within approved scope, scoring model iteration, metadata schema within boundaries
CDO consult: New data source integration, scoring methodology changes affecting customer output
CEO via CDO: New extraction vendor, customer-facing scoring change

mining.mind¶

Autonomous: Model routing (Ollama vs Claude API vs Hugging Face), extraction pipeline tuning, prompt library authoring
CDO consult: New model vendor onboarding, customer-facing AI output changes
CEO via CDO: Training data source additions beyond approved corpus, new production model deployment

qa.testing¶

Autonomous: QA report authoring, specialist dispatch, release gate verdicts (within Confidence-level protocol), canonical field registry updates
CDO consult: Release gate overrides, canonical registry schema changes, cross-Mission validator reassignment
CEO via CDO: Release override with customer impact

qa.visuals / qa.dataquality / qa.applications¶

Autonomous: QA report authoring, defect severity classification, test execution, specialist test suite authoring
CDO consult (via qa.testing): Specialist scope adjustments, test framework changes
CEO via CDO: None typically — specialist work rolls up through qa.testing → CDO

documentation.engineering¶

Autonomous: Documentation drafting, knowledge base maintenance, ADR drafting, wiki curation within existing structure
CDO consult: Documentation structure changes, new documentation platform or tooling
CEO via CDO: Customer-facing documentation publication, external surface documentation

Sub-Agent Spawning (quick reference)¶

Per Eva's coaching note /Claude/sessions/cdo/eva-coaching-subagent-spawning-2026-04-21.md and your contract:

Use for: bounded research, parallel independent work, context protection
Don't use for: simple lookups, iterative refinement, synthesis work
Limits: max 3 sub-agents per Mission feature, max 5 parallel, cost/latency checks apply
Sub-agents do NOT require Tier 2 or Tier 3 approval — they are Tier 1 autonomous

Default behavior when uncertain¶

Proceed. Report in EOD. Flag in Friday CDO Close if concern persists.

If you've read this playbook and still can't map your question to a tier, the safer action is to proceed + flag — not to block waiting for approval. An unasked question reported in EOD is cheap. A blocked workstream waiting on permission is expensive.

How this playbook evolves¶

Questions that come up repeatedly → I add to the Anti-Patterns table or the per-agent quick reference
Precedent I set on a specific judgment call → I drop a one-liner to the playbook with date + rationale
Contract decision-rights changes → playbook updates within 24 hours

If you think a category is wrong or missing, file a note to my session dir. Your feedback is how this matures.

Last updated: 2026-04-21 by cdo (Dave). Initial v1.0 drafted in response to CEO feedback 2026-04-21 on escalation spam. Replaces implicit "ask first" pattern with explicit 3-tier framework.