I · Three questions a working researcher asks
From "where was that?" to novelty to replication.
Demo 1 · Literature lookup across domains1.1 ms · local
contrastive learning · audio-text alignment · before 2022 · with code
01CLAP · contrastive language-audio pretraining; code on GitHub. arXiv · 2206.04769 · Papers-with-Code · 2022-06
0.95
02Earlier · "audio-set + text head, NT-Xent loss" — 2021 workshop. Semantic Scholar · 2021 · § 3.2
0.89
03Your old reading list — exactly this paper, annotated. file://reading/audio-text/clap.pdf#highlights · local
0.82
Demo 2 · Methodology-anomaly scan on a draftscan · per-PDF
scan · draft/manuscript-v4.pdf · p-hacking · missing-error-bars · ai-generated-paper
01Fig. 3 — p = 0.049 across all 4 sub-groups; flagged for p-hacking. scan · anomaly.bio.p-hacking-signal · file://draft/manuscript-v4.pdf#p11
0.93
02Manuscript prose matches AI-generated stylometric signature in §3; ai-generated-paper. scan · anomaly.research.ai-generated-paper · §3
0.85
Demos illustrative. Your drafts and notes never leave the workstation.
II · What's included
arXiv, OpenAlex, IEEE — plus your draft.
Pair with Medical for clinical literature, Life for genomics, Patent for non-patent literature in IP work.
- arXiv (broad) · OpenAlex · IEEE Xplore · Semantic Scholarcross-disciplinary scholarly graph · ~250 M+ items
- full RFC set · OpenLibrary · Project Gutenbergstandards + monographs + classics
- p-hacking · missing-error-bars · replication-gapscan · anomaly.bio.p-hacking-signal · anomaly.physics.missing-error-bars · anomaly.bio.replication-gap
- AI-generated-paper · footnote-drift · outcome-switching · token-driftscan · anomaly.research.ai-generated-paper · anomaly.research.footnote-drift · anomaly.research.outcome-switching · anomaly.research.token-drift
- BibTeX / RIS export with snippetcite-ready, signed
Sources of recordarXivOpenAlexIEEE XploreSemantic ScholarRFCPapers with CodeOpenLibraryGutenberg
II.5 · Sample workflows
Real workflows. Real numbers.
Three workflows shaped for this bundle. Where the box reads Verified, the numbers come from a real run logged in this codebase. Where it reads Pending validation, the workflow shape is implemented and we are still gathering the headline numbers before we publish them.
Workflow 01
Verified
Hair-loss · novel mechanism candidates
Translational research
- InputHair-loss workspace: AA / AGA literature + adjacent immunology + dermatology.
- MethodD-3 Contradictions + D-4 Novelty across arXiv + PubMed + patent data.
- OutputA showcase doc listing the candidate hypotheses, the contradicting papers and the patent-family neighbours.
3 novel candidates surfaced
CGRP-receptor agonism (non-androgen axis), NFATc1 bulge-targeted modulator, AR-PROTAC topical scalp — Hair-loss showcase, 2026-04-27.
Workflow 02
Verified
Live discovery on a 62 k-passage dataset
Indago dogfood
- InputA live mixed dataset of papers + patents + tender docs (62 086 passages).
- MethodAll nine discovery primitives ran end-to-end against
path_filter=.indago/research.
- OutputPer-primitive timing + recall report; 130/130 unit tests green.
Live perf, real dataset
/isolated 2 071 ms · /unusual_pairs 679 ms · /contradictions 2 426 ms · /novelty verified end-to-end, 2026-04-28.
Workflow 03
Pending validation
Pre-submission methodology scan
PhDs · staff scientists
- InputA draft manuscript PDF.
- Method
anomaly.bio.p-hacking-signal + anomaly.physics.missing-error-bars + anomaly.bio.replication-gap + anomaly.research.ai-generated-paper.
- OutputAnnotated PDF with per-figure flags and a suggested-fix sheet.
Pre-submission catch
Scan family is shipped; pending a public reproducibility case study before quoting precision.
Verified workflows are reproducible against the index hash listed in each run. Pending workflows are a written commitment, not a customer quote — we publish numbers only when we have measured them.
Research bundle · from€499/ seat / month· lab / department licenses · scan family included