jhana

Final Eval

V3 graph corpus

309eligible sessions
...files
83,934nodes
2,414questions
Graph proof complete
Case register

Sessions

309
Opening register...
Select a session

Case files and benchmark questions

No session selectedChoose a matter from the register.
Trace inspector

Query Lab

7-way comparison
No session selectedRetrieval preview
Inspect the V3 graphSelect a file, replay a preserved query, or enter one above.
Measured canary

50 questions

350 answers
Opening comparison register...
Same question · same source bundle

Seven-lane answer and reference comparison

LaneRunsMeasured costMedianTool callsReferences
Loading measured runs...
Select a benchmark questionInspect seven answers, references, and full agent traces.
Completeness review

Canary questions

50
Opening evidence register...
Select a question

Suit and agent evidence comparison

Source documents
Frozen 50-question canary

Evidence completeness certificate register

Discovery, graph-semantic exposure, raw-primary inspection, and citations are reported separately.

Download statistics
Computing aggregate report...
Seven lanes

Aggregate statistics

LaneDiscoveryGraph semanticRaw primaryCitationCostTurnsTools
Loading certificates...
Select a question certificateInspect all seven exact traversal grades and the deterministic repair frontier.
Same corpus · same traces · five denominators

Completeness methodology comparison

Each percentage keeps the named build's own frontier, traversal credit, and primary statistic.

Download report
Loading method contracts...
Five builds · seven systems

Aggregate primary scores

Select a canary questionCompare all five completeness methodologies across the same seven retained runs.