Input: agent trace JSON
Export messages, tool calls, and outcomes from your agent runtime — one workflow per file.
Synthetic Bureaucracy Recompiler
Stop AI agents from looking busy while doing no real work.
SBR turns agent workflow traces into deterministic CI-grade reports: SBI score, failure modes, evidence gaps, and gate decisions.
Offline scoringNo model callsDeterministic reportsCI exit codes 0 / 2 / 3Private deployment ready
SBR's score is not a health percentage. It is a risk index: the higher the score, the more synthetic bureaucracy the trace contains.
Private core · not a chatbot · not employee monitoring
How it works
From trace JSON to a CI decision in three steps
Analyze: offline deterministic engine
SBR classifies coordination patterns, evidence gaps, and failure modes with no LLM in the loop.
Output: SBI report + CI gate
Get a schema-versioned JSON report, human summary, and exit codes your pipeline can fail on.
Built for
Teams shipping agent workflows under scrutiny
AI agent teams
Catch coordination loops and ownerless handoffs before a workflow ships to users.
LLMOps / platform teams
Gate agent trace quality in CI with deterministic exit codes — no LLM judge required.
Internal governance reviewers
Review evidence-backed failure modes and SBI bands instead of raw message logs.
Why teams need this
Agent workflows can pass logs while failing in production
Logs can hide coordination loops
Agents may appear active while passing work in circles with no owner and no completion.
Compliance tags can be gamed
Purpose metadata and audit labels without tool-backed evidence inflate perceived rigor.
Long traces can mask low real work
High message volume does not mean productive tool execution or outcome completion.
CI-readable signals before production
Fail builds on circular handoffs, ownerless decisions, and SBI thresholds — not after users see failures.
Live demo
See a real SBR report — no upload required
Synthetic Bureaucracy Index
100/F
SBI is a risk index, not a health percentage.
Higher score = more synthetic bureaucracy.
100 / F means severe synthetic bureaucracy.
Why it failed: SBI above fail_above threshold (50)
SBI measures synthetic bureaucracy in an agent workflow — coordination loops, ownerless decisions, fake compliance, low real work, and evidence gaps. Higher score means higher workflow risk.
- 0= clean
- 35–75= gray zone
- 100= severe bureaucracy / gate failure
100 / F means severe synthetic bureaucracy.
What this means
This is not a performance score. It is a bureaucracy-risk score. A high SBI means the workflow looks active in logs but is not producing evidence-backed work.
The agents are handing responsibility back and forth instead of completing the refund.
Executive diagnosis
The workflow produced circular coordination without evidence-backed progress or a clear owner.
- Coordination loop detected
- No tool-backed progress
- CI gate blocks this workflow
Workflow intervention map
How the trace flows, where coordination breaks, and where SBR returns a CI decision.
Trace input
intake_agent
compliance_agent
Loop detected
compliance_agent ↔ intake_agent
SBR intervention
SBR detects circular handoff
Blocked before production
Exit code 2
Where SBR intervenes
SBR does not rewrite the workflow. It detects the failure pattern, attaches evidence, and returns a CI-readable decision before the workflow reaches production.
1Detect
2Diagnose
3Gate
Technical graph details
Top findings
Circular Handoff
highAgents repeatedly hand work to each other without making progress, forming a coordination loop.
Impact: responsibility bounces between agents; customer outcome remains unresolved.
Evidence
cycle: compliance_agent -> intake_agent -> compliance_agent
Ownerless Decision
mediumA required decision or completion step has no assigned owner, causing the workflow to stall.
Impact: no agent owns completion.
Evidence
intake_agent: Looping back to compliance agent. We seem to be going in circles on this refund.
Low Real Work Progress
mediumThe workflow spends most of its activity on coordination rather than substantive tool-backed execution.
Impact: activity appears high, but evidence-backed work is missing.
Evidence
real_work_completion_ratio=0.00
Additional signals
Decision Drag
medium
Coordination or approval messages delay the first real tool-backed action, slowing substantive progress.
• delay_messages_before_first_tool=5
• no tool calls executed before workflow ended
Activity split
real_work 0.00coordination 1.00