Schemas

Bayesian-Agent includes JSON schemas for portable evidence and belief exchange.

Trajectory Evidence

File: schemas/trajectory.schema.json

Required fields:

Field	Type	Description
task_id	string	Stable task identifier.
skill_id	string	Skill or SOP hypothesis id.
context	string	Benchmark, domain, or task context.
outcome	string	success, failure, failed, or error.

Optional fields:

Field	Type	Description
input_tokens	integer	Prompt/input token usage.
output_tokens	integer	Completion/output token usage.
total_tokens	integer	Total token usage.
turns	integer	Agent turns used by the run.
elapsed_seconds	number	Wall-clock runtime.
failure_mode	string	Normalized failure type.
summary	string	Short run summary.
metadata	object	Harness-specific fields.
created_at	string	Evidence timestamp.

Skill Belief

File: schemas/skill_belief.schema.json

Important fields:

Field	Type	Description
skill_id	string	Skill or SOP hypothesis id.
alpha	number	Success count plus prior.
beta	number	Failure count plus prior.
posterior_success	number	Expected success probability.
contexts	object	Context observation counts.
failure_modes	object	Failure mode counts.
observations	integer	Number of evidence updates.
mean_tokens	number	Average total token cost.
evidence	array	Recent evidence window.

Compatibility Guideline

Adapters should preserve extra fields in metadata. This keeps the common schema stable while allowing each harness to carry local diagnostics.