Compile observed pass rates, retries, and cost per pass into the cheapest task-tier routes that clear your reliability gate.
Eval cells appear here.
Each task and tier pair retains run count, pass rate, retries, spend, and cost per successful run.
A cheap tier cannot receive a task class when its observed pass rate misses the explicit reliability gate.
The compiler chooses the lowest cost per pass only after reliability eligibility is established.
The authored CSV sample gives the standard ambiguous-task tier one pass from two runs. Its observed pass rate is fifty percent, below the default seventy five percent eligibility gate. At three pixels per percentage point the observed bar spans one hundred fifty while the gate spans two hundred twenty five. The elite ambiguous cell has three passes from three runs and clears that gate. These tiny sample proportions are not confidence bounds or measured future reliability. The offline saved copy lacks Papa Parse, disables Compile policy, and cannot demonstrate the calculation. This explanation follows its written sample and routing rule. In the written sample, routine standard spends three hundred cents for three passes, costing one per successful run. Ambiguous elite spends nine hundred cents for three passes, costing three per pass. With five projected tasks in each class, the eligible policy costs five plus fifteen, totaling twenty. All-elite routing costs thirty. At ten pixels per projected currency unit the bars span two hundred and three hundred. This is a cost-per-success projection, not a real dispatch or billing receipt. The source chooses the cheapest eligible historical cell only after checking the pass-rate gate. Lowering the gate from seventy five to fifty would admit ambiguous standard. That cell spent two hundred cents across one pass and one failed run, so its cost per successful run is two, not one. Routine five tasks still cost five, while ambiguous five tasks cost ten, totaling fifteen. At ten pixels per projected policy currency unit the bars shrink from two hundred to one hundred fifty. The cheaper policy trades away the sample eligibility threshold; it does not establish acceptable risk. Missing parser functionality means these are source-derived projections rather than demonstrated exports or live delegation.