steel · synod
Seven minds. One verdict. For decisions that outlive the next sprint.
The synod is metalmind's deliberative council — seven personas spawned as parallel Claude Code subagents, each with its own context window, debating the question and synthesizing a structured verdict. For decisions that affect the next 6 months, not the next 60 minutes.
§ The seven seats
Each persona owns one question.
Switch the vocabulary toggle in the nav to see Kelsier's crew take the same seats.
01
Adversary
red team
"Where does this break?"
02
Strategist
long horizon
"What does this commit us to?"
03
Scientist
evidence
"What's the data say?"
04
Visionary
ambition
"What's the audacious play?"
05
Engineer
feasibility
"How does this actually ship?"
06
Philosopher
first principles
"What should we want?"
07
Humanist
second-order
"Who is hurt by this?"
01
Kelsier
the survivor
"What's the audacious play?"
02
Vin
the skeptic
"What are we missing?"
03
Sazed
the scholar
"What does history say?"
04
Elend
the idealist
"What should we want?"
05
Marsh
the engineer
"How does this actually ship?"
06
Dockson
the operator
"Who pays for it tomorrow?"
07
OreSeur
the loyal critic
"Who is hurt by this?"
§ The verdict
Structured. Auditable. No mush.
Every synod returns the same shape — a position, a confidence percentage, three risks, five next steps, and a minority report when the council can't agree.
› metalmind synod "should we adopt nats over kafka?"
position
Adopt NATS for inter-service messaging in the billing/notif boundary; keep Kafka for the analytics pipeline.
confidence
78%
risks
- JetStream durability semantics differ from Kafka log compaction.
- Operational maturity in your team is lower than for Kafka.
- Migration of in-flight messages requires a cutover window.
next 5
- Spike: NATS JetStream with the billing → notif edge in staging.
- Define retention + replay policy explicitly.
- Run the existing chaos suite against the new transport.
- Write the rollback playbook before the cutover.
- Re-convene synod after 30 days of staging traffic.
minority
Adversary holds: at current scale, the cost of any migration outweighs throughput wins. Revisit only if Kafka ops bill exceeds 2× NATS estimated TCO.
For the call you'll regret making alone.
Not 'should I rename this var.' Not 'what dependency should I add.' The architectural calls. The hiring calls. The ones where being wrong is expensive.