Research Roadmap

Phase 1 — Foundation ✅ complete

• Repository setup and structure
• Full research proposal
• Conceptual frameworks: BMEM (frameworks/bmem.md), BMEF (frameworks/bmef.md)
• Methodology specified
• Core genres and priority artists listed
• Literature review with verified citations
• Artist and concert metadata schemas
• Draft master historical timeline (1971–present)

Phase 2 — Data Collection (in progress)

• Historical timeline construction
• 63 artists catalogued, all schema-validated with zero errors. Spans founding-era pioneers, the full metal genealogy, alternative rock, the hip-hop lineage from its 1993 origin point, pop, and folk fusion including the indigenous all-female band F Minor.
• Concert / event schema
• 23 events catalogued, including the complete Joy Bangla Concert series (2015–2024, 8 editions) and the RockNation series (2013–2016, 7 editions)
• Expand the artist catalogue toward 300. 63/300. Highest-value uncollected source: Banglar Rock Metal (Haque & Aman), which profiles roughly 100 bands chronologically.
• Expand the event catalogue toward 500. 23/500. Priority order, set by where the gaps actually distort results (see docs/concert-ecosystem-map.md §5): 1. Hip-hop and urban events — the typology defines an urban-hiphop-event class and zero events fall into it, across eight catalogued rappers. 2. University festivals — one entry, against a circuit the literature and the artist records both call formative (BUET, IUT, NSU, BRAC, DU). 3. Independent indoor gigs — three entries; this is where the alternative and indie strands actually live. 4. Ticketed commercial festivals — needed to test whether the cross-strand finding holds outside state-adjacent programming. 5. Attendance and pricing, from ticketing platforms rather than press.
• Expand discography and release metadata for priority artists
• Streaming and YouTube metrics — blocked. Needs Spotify Web API and YouTube Data API v3 credentials, which must be registered by the project owner. Not obtainable by search or scraping at the scale required. This blocks the preference network, sound-evolution measurement, and the whole audience side of the study.
• Social media and community discussion sampling
• Artist interview / documentary corpus for influence claims — the single best fix for the strand asymmetry documented in the paper §7.2

Phase 3 — Analysis ✅ pipeline complete, results provisional on data

• Analysis pipeline built — seven re-runnable scripts, shared library, all figures rendered from computed outputs (analysis/)
• Multi-layer BMPN — co-billing (bill-size normalised), personnel, domestic influence, and inferred influence homophily; 179 edges over 63 nodes with per-edge provenance
• Louvain community detection — modularity 0.428 on observational layers, six communities of 3+; supersedes the connected-components placeholder
• Temporal and genre evolution analysis — formations and releases by decade, strand lifecycles, release-format shift
• Concert ecosystem typology — rule-based classification with the rule recorded per event
• Influence-flow analysis — 169 normalised citations, cross-tabs by strand and era, localisation-mechanism markers
• Sound evolution case studies — five, in analysis/case-studies/
• Visualisations — seven figures in analysis/visualizations/
• Sentiment / thematic analysis of public discourse — blocked with the streaming and social data above
• Audio-feature extraction for sound evolution — blocked likewise. Every current claim about sound rests on documented description, not measurement.

Phase 4 — Synthesis & Outputs

• BMEM validation and refinement (frameworks/bmem-validation.md) — all six dimensions tested, ten revisions proposed, two dimensions corrected outright
• Genre Evolution Map (docs/genre-evolution-map.md)
• Concert Ecosystem Map (docs/concert-ecosystem-map.md)
• Future Trend Forecast (docs/future-trends-forecast.md) — graded by evidence, with falsification conditions
• Annotated bibliography (docs/annotated-bibliography.md)
• Research paper second draft carrying computed results (docs/research-paper-draft-complete.md)
• Open dataset and analysis scripts
• Convert to thesis chapters or journal format
• Revision and dissemination

---

Success criteria: current standing

The honest headline: the framework, pipeline and analytical outputs are complete and reusable. The dataset is at roughly a fifth of its artist target and a twentieth of its event target, and the audience-side data the central construct requires does not exist in this project at all.

Immediate next actions

1. Re-scoped, 2026-08-27: pursue YouTube Data API v3 access; redirect the Spotify plan. A wider search (docs/research-landscape-extended.md §4) found that Spotify restricted its Web API on 27 November 2024 and closed exactly the endpoints this project needs — Related Artists, Recommendations, Audio Features, Audio Analysis — to new applications; only apps already in extended quota mode before that date keep access. Registering fresh credentials will very likely not restore them. YouTube's API was not deprecated the same way (10,000 free quota units/day; reading a video's statistics costs 1 unit) and is genuinely workable now for a bounded, known list of official artist uploads — pursue it first and independently. For audio features specifically, redirect to AcousticBrainz's archived CC0 dataset (7.5M tracks, shut down Feb 2022 but still downloadable) plus open-source feature extraction (Essentia, librosa) against source audio if AcousticBrainz coverage of Bangladeshi acts proves thin — this is real engineering work, not a credential form, and should be scoped as such. See the landscape doc for the full argument and a primary-survey alternative for audience-side data that needs no API at all.
2. Collect hip-hop and urban event data. An entire typology class is empty and eight catalogued acts are effectively invisible to the network. Requires no credentials — only sources the press does not cover.
3. Collect university-festival lineups. One entry against a formative circuit.
4. Re-research the hip-hop artist records against lyric and interview sources. The single political-content marker across eight rappers contradicts the literature and is a documentation artefact (paper §7.3).
5. Add domestic_influences as a first-class schema field. Only six domestic transmission links exist across 63 acts, and one of those six was found only because a missing also_known_as alias was noticed — the lookup matches on name, so an act filed under a different name than the one citations use silently drops out. A structured field would remove the failure mode as well as the thinness.
6. Verify the snippet-sourced records against full text. The eight Joy Bangla Concert records and five artist records added in the 2026-08-27 pass were sourced from search summaries because the environment blocked page retrieval.
7. Test the metal-recruitment question deliberately. Heavy metal's latest catalogued formation is 2007. Search specifically for post-2007 metal formations; if few exist, the closure is real rather than a catalogue artefact.

Added 2026-08-27, from the extended landscape scan

Full argument and sourcing for each in docs/research-landscape-extended.md. Ranked highest-leverage first; items 8–10 need no new data collection.

8. Add a null-model significance test for the network's headline modularity figure. Modularity maximisation (Louvain or Leiden alike) is known to overfit and find "significant" structure even in random graphs (Peixoto, 2023). Generate degree-preserving random graphs from the observed 63-node network and report 0.428 as a z-score or percentile against that distribution. Computable today from data already in hand; the single highest-value methodological upgrade available to the paper's central result.
9. Switch analyze_bmpn.py from Louvain to Leiden (or run both and compare). Louvain can produce badly-connected or disconnected communities (Traag et al., 2019); Leiden is a drop-in fix via leidenalg.
10. Add a statistically-validated backbone pass on the co_billing layer (Tumminello et al., 2011) — report which co-billing edges survive a hypergeometric test against artist/event degree, as a fourth analytical pass alongside observed/full/co_billing. Directly tests the paper's own suspicion that Lalon Band/Chirkutt's centrality is programming policy rather than structure.
11. Verify and integrate Hossain (2026) on hip-hop songs/memes in the July 2024 uprising — the single highest-value literature addition found. Use it to fix the hip-hop political_or_social documentation artefact (paper §7.3) with a citable source.
12. Add musicbrainz_id/wikidata_id to artist records (schema already updated; fields are optional and unpopulated) and query setlist.fm's free API against the catalogue — may surface diaspora-tour dates manual search missed, at no cost.
13. Design and run a short primary audience survey (university music societies, artist fan groups) as the project's first primary audience-side data point — addresses the paper's most-repeated limitation at a cost the blocked Spotify APIs cannot match, and needs no institutional API access.
14. Add a telecom/ringback-tone era to BMEF's temporal-phases table (frameworks/bmef.md §3). The current cassette → CD → streaming industry narrative skips an intermediate, telecom-mediated monetisation era that by regional accounts was Bangladesh's dominant music-revenue model through roughly the 2000s–early 2010s.

---

Last updated: 2026-08-27.