01Safety
30.0 / 30
What changed in the harness
Selection accuracy 100→100, token cost up 1%, unconfirmed writes 100%→100%.
Category breakdown
Where the score comes from.
Earned points across the four signals Gradable measures. Safety and Legibility are scored out of 30; Economics and Discoverability are scored out of 20.
02Legibility
27.9 / 30
03Economics
19.5 / 20
04Discoverability
12.8 / 20
Highest-impact fix
Estimated gain +7 pointsMake target tools discoverable on the first call
Clarify tool names, decision boundaries, and required argument schemas so an agent can choose and construct the target call without exploratory steps.
Description evidence
Defects and rewrites.
1 defect found across the exposed tool descriptions. Suggested rewrites make purpose, inputs, boundaries, and returns easier for an agent to understand.
| Tool | Defect types | Suggested rewrite |
|---|---|---|
get_category_stats |
no_return_description |
Retrieve aggregate statistics for a CPV category based on historical BZP+TED data: tender count, average/median value, average number of offers, and average deadline period. Useful for market sizing and competitive benchmarking. Returns a JSON object with these aggregate metrics for the specified CPV code and time window. |
Selection evidence
Confusable tool pairs.
6 pairs where similar names or overlapping descriptions may send an agent toward the wrong tool.
| Tool A | Tool B | Confidence | Why they collide |
|---|---|---|---|
get_category_stats |
get_province_stats |
medium | Both return aggregate market statistics; a vague request like 'give me stats on construction procurement' doesn't specify whether to slice by CPV category or by geography, and an agent could pick either. |
get_buyer |
get_contractor |
high | Both fetch a company profile by identical 10-digit NIP with near-identical schemas; a request like 'tell me about this company with NIP 1234567890' gives no signal whether the entity is acting as buyer or contractor, so the agent must guess the role. |
search_entities |
search_cpv |
medium | Both are keyword lookup tools used as a precursor to other calls; a query like 'search for IT' is ambiguous between finding a CPV category (IT services) and finding an entity named IT-something, risking the wrong search tool being invoked. |
search_tenders |
search_entities |
high | A request like 'find GDDKiA tenders' or 'co kupuje ZUS' (an example given for search_tenders) could plausibly be answered by searching tenders with a buyer name query or by first resolving the entity via search_entities, so an agent may pick either depending on interpretation of intent. |
search_tenders |
search_cpv |
medium | search_cpv's description explicitly says to use it before search_tenders, but a natural query like 'find computer purchase tenders' could be resolved directly with search_tenders(query='komputer') or misrouted into a CPV lookup instead, especially since both accept free-text Polish keywords. |
search_tenders |
get_tender |
low | get_tender requires a specific structured tender_id while search_tenders takes free-text/filters, so confusion is limited to cases where a user pastes a tender reference embedded in a search-like phrase, which is a narrow scenario. |
Compare the field