Public research system · working architecture · 2026
A modular research system for making Belarusian music discoverable, traceable, and understandable across borders—without inventing certainty, flattening culture, or overwriting archival originals.
Research pipeline
Each stage distinguishes evidence from inference.
Owned files, permitted collections, metadata, public evidence, and recovery leads.
Exact file checksums, source links, and stated rights. Audio recognition is not implemented.
Belarusian-first descriptions with Russian and English access layers.
Search, relationships, transparent derivatives, and human review.
Exportable evidence for future studies and teaching. Learning outcomes have not yet been evaluated.
The track does not “move” or disappear into another app. The three modules read or add to one linked research record. The original audio is never silently uploaded, replaced, or rewritten.
Tool suite
The interface works like a research utility cabinet: each tool solves a distinct problem, while shared identifiers and evidence records keep the results connected.
Verifiable file identity & provenance
Create a portable record for origin, rights, context, exact file identity, verification events, and documented derivatives.
Discovery & context engine
Describe music to search an external source-linked catalog, or explore the bounded reviewer corpus and local passports. Keep external discovery, language evidence, and your own annotations distinct.
Authentic, documented derivatives
Create and audition a conservative access derivative without allowing it to masquerade as the historical master.
Data view · current truth
Music Atlas has two separate search spaces. External discovery retrieves public Wikidata music metadata when you request it; catalog coverage and missing fields are disclosed. The local workspace retains 18 seed records and your validated passports, annotations, and analytics. External results do not silently enter your archive. There is no central collection of visitors’ files.
Every passport created on this device appears as a list with basic counts, source status, language, date span, and export controls.
Open local corpus ↗An opt-in, moderated dataset for rights-cleared metadata and selected audio deposits. Contribution consent will be separate from audio permission.
Not collecting public submissions yetEnglish-, Belarusian-, and Russian-language research queries, structured filters, evidence-linked dossiers, a six-page seed-source register, local corrections, and live result-set distributions.
Open live analytics ↗JSON preserves complete records; CSV opens as a table for analysis. Both remain user-controlled and can seed the future shared pilot.
Export from Archive Passport ↗Identity model
The system must never collapse a checksum match, a recording match, a shared composition, and a historical source into one claim.
Can establish: an exact SHA-256 match.
Cannot establish: authorship or historical truth.
Can propose: MP3/WAV or bitrate variants of one recording.
Requires: perceptual fingerprint + threshold.
Can distinguish: studio, live, demo, remix, or broadcast.
Requires: metadata and evidence review.
Can connect: covers and arrangements.
Cannot assume: same performer or recording.
Can connect: label, date, edition, and track sequence.
Requires: release evidence.
Can document: collection, donor, URL, deposit, or recovery chain.
Does not certify: every claim inside the source.
Music Atlas · working pilot
Describe an artist, genre, decade, or format in English, Belarusian, or Russian, inspect the interpreted filters, and request external catalog results. A Belarusian artist does not automatically imply Belarusian-language vocals. Unsupported mood descriptions are disclosed, not treated as verified audio features. This is metadata retrieval, not a search of every music site or an audio-listening AI.
Other research questions
Dance works and remixes in Belarusian Belarusian folk artists Electronic music from the 2000sRestoration Lab · working pilot
Restoration is a new version, not a corrected past. The master remains untouched; every derivative receives its own fingerprint, method log, and review record.
Preserved as received, with provenance and rights status.
Model, software, settings, edits, intent, and uncertainty logged.
Compared with the master and clearly labeled as processed.
Current implementation: the browser Lab verifies an exact passport-linked master, performs conservative high/low-pass filtering, gain and normalization, enables A/B listening, exports a WAV and JSON intervention log, and registers a new derivative fingerprint. AI reconstruction, denoising claims, and perceptual-quality findings remain outside this pilot.
Bounded pilot
Research boundaries
These are operational constraints, not decorative ethics language.
Metadata, audio, lyrics, and images enter only through a documented legal or consent basis.
AI suggestions keep their source, confidence, model/date, and human correction history.
Restored or enhanced media remains a derivative with its own fingerprint and intervention log.
Location and sensitive context are collected only when the research question and permission justify them.
Technology map
ShazamKit and AcoustID already demonstrate recording recognition. The system’s research contribution is the provenance-aware cultural layer that connects an audio match to works, releases, sources, language, contested context, and explainable evidence.
Live module