← All examples Levirge Brain · examples

Dense results

A measured before/after on the same index: what a search returns without Brain's summary layer, and the single consolidated answer it returns with it.

1

The query

"cyclooxygenase inflammation cancer" — one question against one index, asked twice. The only variable is whether the summary layer is in the way of the results. Measured by Levirge's evaluation harness on BEIR SciFact, a public 5,183-document biomedical benchmark corpus, 15 Aug 2026.

Each step below shows the readable version first, with the raw payload behind a tab — the same bytes an agent receives, from a live pull on the same corpus.

2

Without the summary layer — eight fragments

Ranked document hits. Each is a real source; none is an answer. Two are only loosely on the subject, so the model pays to read them and then pays again to discard them — every time anyone asks.

1  Cyclooxygenase-Dependent Tumor Growth through Evasion of Immunity
2  Tumor-associated macrophages and high COX-2 → poor prognosis
3  COX-2 inhibitors enable repeated vaccinia virus administration
4  PI3Ks qualify as drug targets for cancer, inflammation …  (loosely related)
5  Osteoarthritic cartilage degradation indicators …         (loosely related)
6  15-Hydroxyprostaglandin dehydrogenase down-regulated in colorectal cancer
7  Omega-3-derived lipid mediators
8  COPD, neutrophils, cigarette smoke and COX

Thousands of tokens, two of them off the point, and the synthesis is still the model's job.

3

With the summary layer — one consolidated answer

The top result is a single entity summary, consolidated from 11 source documents, each claim carrying the source it came from. The agent cites this and stops.

Cyclooxygenase-2 (Cox-2) — an enzyme implicated in inflammatory
and pathological processes: viral clearance, lipid-mediator
generation, and cancer development.

- Viral clearance: Cox-2 inhibitors circumvent viral-clearance
  limits via neutralising antibodies                     [cited]
- Lipid mediators: bioactive signals (15-epi-LX, 18R-HEPE)
  via NSAIDs + omega-3                                   [cited]
- Renal vasoconstriction                                 [cited]
- Cancer: survivin correlation                           [cited]
- COPD airways                                           [cited]

The summary also carries a small entity-relation graph — cox-2 → survivin, PGE-2, arachidonic-acid, aspirin — so the next question about a neighbouring subject starts from the relations, not another eight-document read.

4

What was measured

Consolidation only helps if the dense answer actually surfaces — and only if it doesn't cost ranking quality. Both were measured on the same corpus and query set.

evaluation run 5,183 documents
summaries in top-10   35%          — they surface, not buried
rank-1 for            25% of queries
nDCG@10               0.71         — BM25 baseline 0.665

 consolidation does not cost retrieval quality
The same corpus consolidates interleukin-1b from 7 documents and azacytidine from 3. Consolidation is per entity, so the ratio follows how much has been written about a subject — and so does quality. Thin input shows: an entity with only a handful of sources can open by defining itself in terms of itself, because there is not enough cross-document material to say something denser.
5

Where this came from

Both result lists come from the same index and the same query. The effect is a property of consolidation, not of this corpus: how dense your answers get depends on how many sources say something about the same subject.

provenance 15 Aug 2026
corpus      BEIR SciFact — public biomedical benchmark, 5,183 documents
variable    the summary layer — same index, same query, both runs
raw tabs    a live pull for "interleukin-1b inflammasome" on the
            same corpus, 15 Aug 2026 - a second query, because it
            is the one we hold verbatim payloads for
payloads    titles excerpted; summary text verbatim apart from one
            unrelated line - a consolidation defect we logged
            rather than quietly leave in the example
harness     Levirge evaluation harness, 15 Aug 2026