AgentInputLab Studio

01 / METHOD

Select the intervention

Every version comes from the frozen registry. Studio-ready is narrower than registry-ready.

Algorithm Semantic family Version profile Runtime profile

PROFILE

Select a method to inspect its implementation boundary.

02 / RUNTIME

Pin the execution context

Only deterministic replay is enabled in this local Studio build.

Model backend Backend capability

Seed Samples Top k Temperature Soft word budget Clarify threshold

Parameters remain in the run manifest even when a selected algorithm does not consume them.

03 / DATA

Choose an evaluation cell

Registered datasets are visible; only verified prepared assets may run.

Dataset / benchmark cell Preparation status

Task: —
Scale: —
Metrics: —
Families: —

SAMPLE

A sample appears for locally prepared fixtures.

04 / CONTROLS

Protect the comparison

Disabling a control marks the run exploratory in the compatibility report.

Paired baselineEvaluate the same frozen examples before and after intervention. Fixed seedDerive every sample seed from one recorded run seed. Shared retrieverUse exactly one corpus, tokenizer, BM25 k1, and b configuration. Freeze dependenciesRecord backend, corpus, and retriever identities in the result. Fail on fingerprint driftStop if a dependency identity changes across the paired run.

Configure an experiment

Select the intervention

Pin the execution context

Choose an evaluation cell

Protect the comparison

Run queue

Experiment configuration workflow

Select the intervention

Pin the execution context

Choose an evaluation cell

Protect the comparison

Run queue