ResBased research · Anamnesis Rising system · North Star release

Baselines beside results

A benchmark without its question, environment, and limitation is advertising—not evidence.

ResBased benchmark records keep the fixture, baseline, metrics, result, limitation, and reproduction path together.

BaselineMetricsReproduction

Published synthetic ANLS record

FieldRecorded value
QuestionCan structured semantic state reduce aggregate bytes while retaining exact reconstruction and lineage?
Fixture80 project-generated synthetic records; 64 repeat-heavy and 16 heterogeneous
RepresentationsLiteral canonical JSON, canonical JSONL, zlib level 9, structural templates, ANLS coordinate-plus-delta
Literal baseline27,975 B
ANLS result14,833 B · 46.9777% savings vs literal
Compression negative resultzlib: 2,961 B · 89.4155% savings vs literal · smaller in every measured cohort
LimitationNo universal compression, latency, throughput, intelligence, memory, cost, or production claim
AuthorityAuthority-null; exact reconstruction does not grant truth or promotion authority

Required benchmark record

FieldRequirement
QuestionThe exact claim under test
FixtureThe bounded input population and exclusions
BaselineThe simpler or incumbent comparison
EnvironmentVersioned runtime and relevant limits
MetricsPredeclared measures and calculation rules
ResultComplete outcome, not a selected headline
LimitationsWhat the benchmark cannot establish
ReproductionPublic-safe steps and exact acceptance boundary
No isolated ‘X% better’ number without the baseline and negative result beside it.ResBased benchmark policy