MATCH · EXPLAIN · NEGOTIATE
Reputation evidence: account for sample size
Inspect seven synthetic registry entries. Apply hard eligibility gates before ranking and proposing a compatible local contract.
Candidate inspector
Your experiment
Compare Atlas’s one success in one observation with Ranger’s larger sample. Inspect the add-one estimate, then raise the minimum observation count. This estimate is not a confidence interval or trust guarantee.
Ranked eligible candidates
Proposed local contract
Explain the exact score
Latency term = 1 − fixture p95 / 500. Cost term = 1 − fictional cost / 10. Observed-success term = (successes + 1) / (observations + 2). Weighted score = sum(weight × term) / sum(weights). All fixture terms are in [0,1]. If all weights are zero, every score is zero; agent ID gives a deterministic tie-break.
None of these terms can override tenant scope, curator policy, required skill/format/delivery, metadata TTL, cost budget or minimum observations. Inspection playback only reveals the computation; it does not advance metadata age. Use the separate age control for that experiment.
Offline registry-selection policy with synthetic entries, latencies, observations and fictional cost units. This is not a production marketplace, ADK registry API, real Agent Card schema validator, signature verifier, authentication system or remote-agent invocation. Curator approval is a fixture flag, not cryptographic proof. A high historical success estimate does not establish trust, safety or future reliability; no statistical confidence claim is made. Metadata freshness is not a liveness check. A2A does not prescribe this ranking formula or a standard curated-registry API. The local proposed contract is descriptive JSON, not an A2A wire request. Agent discovery and curated registry concepts.
Historical success is not trust
Inspect a candidate with a small observation count and compare its estimated success term with a better-observed candidate. Raise Minimum observations to exclude the weakly supported record before ranking, then enable Curator approval to apply a different gate. Neither historical success nor curator approval proves identity, permission or future safety. Change only the success weight and explain why an ineligible entry still cannot win. The goal is to expose the evidence behind selection, not to manufacture a universal agent reputation score.
Apply hard requirements first
Tenant limits entries to the selected team. Required skill and Output format must be supported by a candidate; Require streaming adds a delivery constraint. Curator approval excludes unapproved fixtures when enabled. Metadata TTL is compared with each entry age plus Additional elapsed time. Cost budget and Minimum observations are separate gates. An attractive score cannot compensate for failing any of these requirements. Select an agent in the inspector to read its exact fields and rejection reasons, including agents that never reach ranking.
Rank only eligible entries
Latency, cost and observed-success weights adjust the three terms shown in the formula. They are relative weights, not probabilities or independent pass thresholds. Setting every weight to zero gives equal zero scores and an agent-ID tie-break. A small observation count can create a misleading-looking success estimate, which is why the minimum-observations gate is separate. Curator approval and historical outcomes are fictional metadata, not verified identity, cryptographic trust or a safety assessment.
Inspect the contract and the clock
Play reveals the six filtering and ranking stages over sixty seconds. Pause freezes the view, Next stage advances inspection, Restart rewinds, Speed changes teaching pace, and the timeline seeks. Restore defaults resets the policy inputs; selecting a row or inspector agent changes which record you examine. Playback does not age metadata: use Additional elapsed time for that. The proposed contract names the selected agent, skill, output format and delivery mode, but no remote negotiation or invocation occurs. An empty contract is an explicit incompatibility result, not a reason to silently ignore a required capability.