Choose for the whole chain, not the first step.
Compound observed stage reliability, include retry time, then select the fastest profile that still clears your deployment floor.
Measured stage matrix
pass rate × latency
per profile and stage
per profile and stage
Synthetic illustration. These names mirror the source’s tiering concept, but the values are not ByteDance or Seed2.0 measurements. Replace them with repeated observations from your own workflow.
| Profile | Workflow stage | Pass rate % | Latency s |
|---|
Ready to analyze the synthetic fixture.
Constraint-aware result
fastest expected success
among qualifying profiles
among qualifying profiles
No route selected yet.Analyze the matrix to see how small stage failures compound through the complete workflow.
RECOMMENDED QUALIFYING PROFILE
—
QUALIFYING—
FASTEST RAW—
RELIABILITY FLOOR—
| Profile | Chain success | Base latency | Expected completion | Successes/hour | Weakest stage | Gate |
|---|
Interpretation boundary: Stage independence is an explicit simplifying assumption. The sample does not measure or rank Seed2.0, Claude, Gemini, or any provider. Correlated failures, queueing, cost, quality severity, and recovery strategies require separate measurement.