← RSI Lab
The Solutions-Architect loop: the artifact that evolves is an architecture spec — a services table with prices, an availability note, an IaC excerpt. The frozen judge scores four things at once: service economy (fewest independently-operated services — but a merge that breaks separation of concerns scores DOWN), cost efficiency (every dollar traces to a stated need), availability fit (redundancy sized to the SLO — over-provisioning is a defect too), and boundary integrity (compliance/security boundaries survive every consolidation).
The Solutions-Architect loop: the artifact that evolves is an architecture spec — a services table with prices, an availability note, an IaC excerpt. The frozen judge scores four things at once: service economy (fewest independently-operated services — but a merge that breaks separation of concerns scores DOWN), cost efficiency (every dollar traces to a stated need), availability fit (redundancy sized to the SLO — over-provisioning is a defect too), and boundary integrity (compliance/security boundaries survive every consolidation).
Architecture is a moving target — prices, traffic, team size, and regulations all drift. That's why every volatile factor lives in the workload profile below (never in the frozen judge): when the world moves, update the profile and re-run with your last accepted architecture as the new baseline. Prices are dated inputs — any $ figure not in ALLOWED FACTS is treated as fabrication and auto-rejected.
Watch: two real consolidations — zero keys needed
Workload profile — what conditions the optimization: app type, goals, regulations. Compiles into INTENT + ALLOWED FACTS below.
App type · industry
Regulations
Availability target · critical path
Budget ($/mo) · traffic · team size
Providers in play
Pricing anchors + hard requirements (the judge treats any $ not here as fabrication — paste your Infracost output)
then pick an architecture below (or paste yours) and run
Run the consolidation loop — same engine + house rules as the parent Lab
Architecture EVOLVES — services table + availability note + IaC excerpt
INTENT (from the profile — drift auto-rejects)
ALLOWED FACTS (profile clauses + pricing anchors — the only evidence)
Judge FROZEN — how the architecture is scored
Via negativa (improve by subtraction)
FROZENDensity Necessity Clarity after cuts Nothing lost
First-principles rebuild
FROZENRequirements covered No cargo cult Directness Load-bearing structure
Infrastructure optimization architect
FROZENService economy Cost efficiency Availability fit Boundary integrity
Well-Architected pillar review
FROZENCost fitness Reliability / SLO fit Operability per head Tradeoff candor
Loop — proof-gate pairs well: every merge must argue dimension-by-dimension why it wins; evolution surfaces divergent proposals for contested trade-offs
Run
model
loops
When the target moves — prices change, traffic 10×es, a regulation lands: update the profile, load your last accepted architecture from the library (☆ on a finished run) as the new baseline, and re-run. The judge stays frozen; the frame is yours to move. Each accepted proposal ends with a sensitivity note naming the factors it depends on — the conditions under which you should come back.