The Nineteen Rules
Five override states, seven principles, seven tactical rules. All earned, none assumed.
Thirty-plus experiments. Measured results. Some surprises, some confirmations, a couple of expensive mistakes. These are the rules that survived.
Three layers: named overrides that take over when conditions demand it, core principles that run all the time, and tactical rules for specific situations.
Named Protocols — Override States
These suspend normal operations. Four have executable skills. One is a policy flag.
Alpha — Production emergency. Everything else stops. SSH in, read logs, fix it. Document after.
Gamma — Urgent fix. Something’s broken in code. Shortest path to working. Leave TODO(gamma) markers on every hack.
Delta — Use everything, waste nothing. Session time running out. Scan all projects for parallelizable work, swarm it. Quality still matters — no garbage.
Omega — Knowledge preservation. Context getting long or session ending. Write findings to files, update docs, leave breadcrumbs. Nothing discovered should be lost.
Epsilon — Emergency cost control. Kill all paid API calls. Free tiers and local models only. (Policy flag, no skill.)
Core Protocols (1–7)
1. Test, don’t guess. Prove, don’t assume. Deliver, don’t hoard.
2. The mission comes first. Every test serves a project. Every finding reaches the projects that need it.
3. Context beats compute. An 870-token glossary made Haiku match Sonnet. Feed the model what it needs before reaching for a bigger one.
4. Cheap for plumbing, expensive for people. Users deserve quality. Pipelines don’t care.
5. Fall back, don’t fall over. API → Groq → Local MLX. The delta is 0.4 points. Uptime is worth more.
6. Two fix attempts, then instrument. Stop guessing. Add logging. Data beats theory, especially when frameworks are doing magic you can’t see.
7. Ship the breadcrumb, not the encyclopedia. Signpost in every affected project — experiment number, three-line summary, path back to full docs.
Subprotocols (S1–S7)
S1. Let no free API credit go unused.
S2. If a rule can do it, a model shouldn’t. $0, instant, 100% reliable.
S3. Local when it’s free, cloud when it matters. Apple Silicon eats OCR, transcription, and batch for breakfast. Pay for judgment.
S4. Rotate, don’t pray. Key touched git, logs, or history? It’s compromised. Rotate immediately.
S5. Don’t fine-tune until your prompts are perfect. I spent $26 and got a worse model.
S6. The specialist beats the generalist. Purpose-built APIs outperform scaling general models.
S7. EU when equal, exotic when fun.
Protocols are amended when experiments prove them wrong.