Research Landscape: Extended Analysis and Recommendations (2026)

This document extends the project's literature review and methodology beyond the sources already cited in docs/literature-review.md and docs/annotated-bibliography.md. It surveys the wider scholarly, technical and industry landscape as of August 2026, evaluates it critically against what this project has built, and converts the findings into concrete, prioritised recommendations. It is a companion to docs/roadmap.md, not a replacement: the roadmap says what to do next in sequence; this document says why, with the evidence.

Scope note. Search access in this environment returns result summaries and occasional fetched pages, not universal full-text retrieval — several publisher domains (Spotify's developer blog, SAGE, ResearchGate, RSIS International) returned EGRESS_BLOCKED when fetched directly. Claims below are sourced from search-result summaries unless a direct quote or fetch is noted, and every specific claim below should be read with that caveat — the same discipline the project already applies to its snippet-sourced data tag (§5.1 of the research paper). Where a claim matters enough to appear in the paper, it should be verified against full text first.

1. What this project has already established (baseline)

Before extending outward, the baseline this document builds on:

63 artists, 23 events, 179 network edges, 169 normalised influence citations, seven analysis scripts, seven figures, a tested and revised conceptual framework (BMEM/BMEF), a paper at second draft with twelve numbered results.
The project's own strongest methodological move is the bill-size correction on concert hypergraph projection (1/(n−1) weighting per Newman's scientific-collaboration-network convention) and the three-pass reporting (observed/full/co_billing) that makes the correction's effect visible rather than asserted.
The project's own strongest epistemic move is naming what it cannot do (audience data, audio features) and retracting first-draft claims that outran the evidence — most visibly, renaming Dimension 4 from "preference network" to "co-appearance network" once the edge provenance was examined.

This is a genuinely unusual level of rigour for a project at this stage, and the sections below are written to sharpen it, not to relitigate it.

2. Scholarly literature: what's new since the last pass, and what was missed

2.1 A new primary source on the July 2024 uprising and hip-hop

Hossain, Tareq (2026). "Voice for justice: Hybrid cultural resistance of hip-hop songs and memes in July mass uprising, Bangladesh." (SAGE journal; DOI 10.1177/13678779261452687.) This was flagged in the literature review as "not yet reviewed in depth" — it should now be treated as core, not peripheral, reading. Per search summaries, the study examines 10 hip-hop songs and 77 memes across five meme groups shared on Facebook and YouTube during the uprising, arguing these functioned as tools of "hybrid cultural resistance" merging digital creativity with street activism. It names two anthem tracks — "Kotha Ko" (Speak Up) and "Awaz Utha" (Raise Your Voice) — with the latter's performer, Hannan Hossain Shimul, reportedly arrested on 25 July 2024 shortly after release.

Direct relevance to this project. The catalogue already has a hannan artist record and cites a "2024 breakthrough tied to the mass uprising" — the paper's §7.3 flags this as contradicting the project's own thin political_or_social marker count for hip-hop. Hossain (2026) is the exact academic source that would let the project correct that documentation artefact with a citable authority, rather than only asserting it introspectively. This is the single highest-value literature addition available: verify the full text, then re-research the hannan record (and any other 2024-adjacent hip-hop acts) against it, and cite it directly in §7.3 and in analysis/case-studies/hip-hop-lineage.md.

The Daily Star also ran a piece, "Sound of the July uprising," profiling protest rap songs — a secondary/journalistic source worth checking as corroboration and for additional song titles/artist names not yet catalogued.

2.2 Coke Studio Bangla now has a dedicated academic literature

The project's framework document (frameworks/bmem-validation.md, Dimension 5 and 6 discussion) already identifies Coke Studio Bangla as institutionally significant but had no academic citation for it. That gap is now closed:

"Beyond Appropriation and Commodification: The Making of Coke Studio Bangla's Complex Subjects" (ResearchGate record 388400913; author identifies as a "trained feminist anthropologist" per search summary — the exact author name needs full-text confirmation). The paper's stated argument, per its own framing, is that treating Coke Studio Bangla purely through an appropriation/commodification lens is inadequate, and that the platform should instead be read through the lens of "supply chain capitalism" — the scale and diversity of capital, labour, bodies and resources it mobilises.

This is directly usable in the project's Dimension 6 discussion, which currently treats Coke Studio Bangla only descriptively ("a beverage brand's cultural-prestige project"). The supply-chain-capitalism framing gives the project's "non-commercial funders" argument (BMEM revision #9) a sharper edge: Coke Studio Bangla is not non-commercial at all — it is commercial in a different register (brand equity, not direct song sales), and conflating that with the state-adjacent Joy Bangla Concert's genuinely non-commercial funding model understates a real difference the project's own data could measure (are Coke Studio pairings cross-strand at the same 1.00 rate as Joy Bangla bills? The catalogue does not yet have Coke Studio episode data to test this).

There is also a broader comparative literature on Coke Studio Pakistan (e.g. work reading it as a "soundtrack for reimagining Pakistan," and studies on its role during periods of instability) which supplies useful comparative theoretical framing for how brand-sponsored music platforms function as national-identity infrastructure across South Asia — worth a paragraph in the literature review's "Broader Theoretical Lenses" section, since the project currently treats this as a purely Bangladesh-local phenomenon.

2.3 Bangla-language music consumption and industry studies (2025)

Two 2025 studies close gaps the project's methodology document gestures at but never cites:

"Music Industry on Digital Platform: A Study on Five Major Music Labels of Bangladesh" (Rajshahi University Journal of Social Science and Business Studies, June 2025). Studies five labels — Chena Sur, Soundtek, Anupam Music, Laser Vision, G Series — across Facebook, YouTube, Instagram, TikTok and Twitter, reportedly finding YouTube is now these labels' sole meaningful income channel. This is a direct, citable source for the project's Dimension 6 (Industry Transformation) — currently the project's weakest-evidenced dimension by its own admission ("largely untested"). G Series already appears informally in industry background; this paper would let the project cite label-level digital-transition evidence directly rather than relying on Wikipedia label pages.
"Music Consumption on Digital Platforms: A Study on Bangladeshi University Students" (International Journal of Research and Innovation in Social Science). Reportedly finds Rajshahi University students have largely abandoned traditional formats for Facebook/YouTube/Spotify, prefer legal free sources over piracy. This is the closest thing found to primary audience-side survey data for Bangladesh specifically — not a substitute for the streaming API data the project is blocked on, but a citable secondary source that could support (or complicate) any inference the project draws about audience platform behaviour once it does get API access. It is also a methodological template: the project could commission or run an equivalent lightweight primary survey (a Google Form, distributed through university music societies) as a low-cost partial substitute for the blocked API data — this is the single most actionable idea in this section (see §7.2 below).

2.4 The pop strand remains the least-served in the literature — confirmed, not closed

The project's own paper (§3) states "no academic literature at all on the pop strand." This extended search did not find a dedicated academic treatment of Bangladeshi pop either — the closest adjacent material is comparative Coke Studio scholarship (which crosses into pop/fusion) and industry-side pieces (labels, digital consumption) that touch pop only incidentally. This gap is real and confirmed, not a retrieval artefact of the earlier pass. The project's own framing of this as a genuine research gap, rather than a project failure, holds up.

2.5 Computational precedent beyond Autul et al. (2024)

The project names Autul et al. (2024) as its sole computational precedent. Wider computational-musicology literature supplies methodological models the project should be aware of even though none targets Bangladesh directly:

Mauch, MacCallum, Levy & Leroi (2015), "The evolution of popular music: USA 1960–2010," Royal Society Open Science 2(5), arXiv:1502.05417. Uses MIR and text-mining on ~17,000 Billboard Hot 100 recordings to build an audio-based style classification and track harmonic/timbral diversity over time, explicitly framed as replacing anecdotal pop-history claims with measured ones. This is the closest large-scale precedent for exactly the "sound evolution" work this project currently cannot do (blocked on audio data) — and its method (MIR feature extraction + unsupervised style clustering + diversity/disparity metrics over time) is a concrete target architecture for the project's audio-feature phase once source audio or embeddings are obtainable.
CompMusic project (Xavier Serra et al., Universitat Pompeu Fabra, 2011–2017) and its Dunya corpora — a sustained programme building culturally-specific computational-musicology corpora for five non-Western traditions (Hindustani, Carnatic, Turkish-makam, Arab-Andalusian, Beijing Opera), explicitly rejecting the assumption that Western MIR features transfer cleanly to other traditions. This is the strongest available model for how a Bangladesh-specific computational musicology should be built if it ever moves beyond metadata-network analysis into audio: not "run Western MIR tools on Bangladeshi audio" but "identify what Baul/Bhatiali/Bhawaiya melodic and rhythmic structures actually are, and build description tools fitted to them." The project's dataset currently has no place to record this kind of tradition-specific musical information (see §6.2 recommendation on schema).
Decolonising-MIR literature. A 2026 ISMIR bibliometric study, "Beyond a Western Center of Music Information Retrieval," documents systematic underrepresentation of Global South scholarship and traditions in MIR research; a related 2024 paper, "Missing Melodies: AI Music Generation and its 'Nearly' Complete Omission of the Global South," reports that roughly 86% of dataset-hours and over 93% of researchers in AI music generation work focus on Global North music, and roughly half of surveyed papers use symbolic-generation methods that don't capture South Asian musical structure well. This literature reframes the project's own data gaps as a documented instance of a known, named, studied structural problem in the field — not simply "this project hasn't gotten to it yet." Citing this literature strengthens §10 of the paper (Limitations) by situating the project's audio-data blockage inside a recognised field-wide pattern, and gives the project grounds to argue (in any future funding or collaboration pitch) that filling it is a contribution of general interest to MIR, not just to Bangladesh studies.

3. Network methodology: what the wider literature says about this project's central technique

This is the area with the highest-leverage, lowest-cost improvements available, because the project's BMPN pipeline (analysis/scripts/build_bmpn.py, analyze_bmpn.py) is exactly the kind of construction the network-science literature has spent two decades refining.

3.1 The bill-size correction is right, and has a name

The project's 1/(n−1) per-event edge weighting is methodologically identical to Newman's (2001) collaboration-network weighting for multi-author papers — the same logic (more co-participants dilutes the strength of any one pairwise tie) applied to concert bills instead of papers. This is good company to be in, and the project should cite Newman (2001) directly next to its own description of the correction in docs/methodology.md and the paper's §5 — right now the correction is described but not attributed to prior art, which makes it look like an ad hoc fix rather than an application of an established technique. Low-cost fix: add the citation.

3.2 But the correction alone doesn't answer the harder question the literature has moved to

Bill-size weighting controls for how much an event contributes, but not for whether an edge is more than what raw co-occurrence would produce by chance. This is exactly the problem the bipartite-network-projection literature has spent the last 15 years on:

Tumminello, Miccichè, Lillo, Piilo & Mantegna (2011), "Statistically Validated Networks in Bipartite Complex Systems," PLOS ONE — introduces a hypergeometric-distribution test to assign each projected edge a p-value against a null model that accounts for each node's degree in the bipartite graph (i.e., how many events an artist appears at, and how large each event is), retaining only edges more frequent than chance would predict.
Neal, Domagalski & Sagan et al. and follow-on work on "the backbone of bipartite projections" (fixed degree sequence models, Stochastic Degree Sequence Model with Edge Constraints) — a family of null models purpose-built for exactly the co-authorship/co-sponsorship/co-attendance case the project's concert-bill network is an instance of.
A 2025/2026 arXiv paper, "How null-model constraints affect statistical validation in projected bipartite networks," shows the choice of null model (configuration model vs. hypergeometric approximation vs. partial configuration model) materially changes which edges survive validation — there is no single "correct" null model, only better- and worse-justified ones for a given system.

What this means concretely for the BMPN. The project's co_billing edge layer currently treats every shared bill as evidence at a weight set purely by bill size. It does not ask "is this pair co-billed more than two artists of their respective overall event-frequencies would co-occur by chance?" Lalon Band and Chirkutt's degree-18 hub status (flagged in the paper's own Result 7 as "programming policy, not audience overlap") is exactly the kind of high-frequency, low-information hub that statistical backbone extraction is designed to downweight or flag. Running a Tumminello-style hypergeometric test (or the simpler fixed-degree-sequence backbone method) over the current co_billing layer, even on today's 23-event dataset, would let the project report a second methodological result, comparable in kind to its already-published bill-size-correction finding: which co-billing edges survive statistical validation against a null model, and whether the "national festival circuit" community (currently the network's largest, per §7.4) is mostly validated structure or mostly artefact of a few artists appearing at many undifferentiated festivals. Given the project's demonstrated appetite for exactly this kind of self-critical methodological result, this is a natural next analytical script (analysis/scripts/validate_bmpn_backbone.py) rather than new data collection — it can run on the existing 23 events today.

3.3 Louvain's known defect, and what to do about it

Traag, Waltman & van Eck (2019), "From Louvain to Leiden: guaranteeing well-connected communities," Scientific Reports 9, arXiv:1810.08473 — demonstrates that Louvain (the algorithm analyze_bmpn.py currently uses, per the project's stated pipeline) can produce badly connected or even internally disconnected communities, empirically up to 25% badly connected and 16% disconnected in their benchmarks, especially under iterative refinement. The Leiden algorithm (same paper) is a drop-in, faster replacement with an explicit guarantee against this failure mode, implemented in the widely-used leidenalg Python package (which also works directly with networkx/igraph graphs, so this is a low-effort swap in analyze_bmpn.py).

Recommendation, low cost, moderate value: switch from Louvain to Leiden (or run both and report whether the six communities are stable across the swap) — cheap to do, and forecloses a specific, citable objection a reviewer familiar with the network-science literature would raise against the paper's central network result.

3.4 A sharper critique: is modularity the right question at all?

Peixoto (2023), Descriptive vs. Inferential Community Detection in Networks: Pitfalls, Myths and Half-Truths (Cambridge University Press; also arXiv:2112.00183) makes a stronger claim than the Leiden paper: modularity maximisation (Louvain or Leiden alike) is a descriptive method that systematically overfits — it finds "significant" community structure even in random graphs with no real structure, and Peixoto's benchmarks show inferential methods based on the stochastic block model (SBM) provide better statistical compression on 96%+ of tested empirical networks. This is a genuinely different and more serious challenge than the Leiden fix: the project's headline modularity figure of 0.428, reported as evidence the network is "genuinely structured," is exactly the kind of number this literature says should not be trusted at face value without a comparison against degree-preserving random graphs of the same size.

The project's data-quality-report.json and pipeline already compute an observed_uncorrected comparison pass — the infrastructure for a rigour step like this already exists. The concrete, cheap addition: generate ~100–1000 configuration-model random graphs preserving each node's degree from the observed 63-node, 153-edge network, run modularity maximisation on each, and report the observed 0.428 against that null distribution (a z-score or percentile). If 0.428 sits far above the random-graph distribution (plausible, given the paper's own qualitative account of why communities form — shared membership, cited influence — these are not random processes), this becomes a stronger, more defensible result, not a weaker one: "modularity 0.428, z = X against 500 degree-preserving null graphs" is a claim no reviewer familiar with Peixoto's critique can wave away, whereas the current unqualified 0.428 invites exactly that objection. This is the single most important methodological upgrade available to the project's central network finding, and it is entirely computable from data already in hand — no new collection required.

3.5 Comparative context: festival-lineup network studies elsewhere

Analyses of major Western festival lineups (e.g. BBC Data Unit's Glastonbury gender-representation network work, and independent lineup-uniqueness studies reporting festivals sharing artists with an average of ~7 other festivals in a dataset) confirm co-billing networks are a live, recognised genre of music network analysis outside academia too — but none found apply a rigorous null model either; most report raw overlap counts. This suggests the project, if it implements §3.2–3.4, would be doing more careful work than the informal industry-data-journalism baseline, which is a genuine differentiator worth stating explicitly in the paper's discussion of related work.

4. Audio and audience data: the landscape has changed since the roadmap was written, and not in the project's favour

The roadmap treats Spotify/YouTube API credentials as the blocker, to be resolved once the project owner registers for access. This needs revision: the landscape shifted materially on 27 November 2024, and the shift is a structural one, not a credentialing delay.

4.1 Spotify closed the exact endpoints this project needs — for new applications

On 27 November 2024, Spotify restricted its Web API, and — per multiple industry-press summaries (Music Ally, Digital Music News; direct fetch of Spotify's own developer blog post was blocked by this session's network policy and should be verified independently) — the specific casualties are Related Artists, Recommendations, Audio Features, and Audio Analysis, along with 30-second preview URLs and several playlist/editorial endpoints. Crucially: existing applications already in "extended quota mode" before that date keep access; new applications cannot get it. Spotify does not onboard new extended-mode apps below a 250,000-monthly-active-user threshold. This project, registering fresh, would very likely be a new application and would not be able to obtain audio features, related-artist adjacency, or recommendation data at all, regardless of how the credentials are registered. This is the single biggest correction this research pass makes to the project's own stated plan.

What this means for the roadmap's item #1 ("Register Spotify Web API and YouTube Data API v3 credentials — the one blocker holding back the preference network"). It is likely no longer true that registering credentials solves this. The Spotify half of that plan needs to be re-scoped:

Audio features are not coming back from Spotify for a new project. Search-industry commentary as of mid-2026 confirms there is still no official Spotify replacement 18 months on.
What Spotify does still give new apps: basic catalog search, track/ album/artist metadata (name, popularity score, release dates, genres where tagged), and playlist data reachable through non-deprecated endpoints. This is enough for some of what the project wants (a popularity proxy, genre tags, release timelines that corroborate the discography data already collected) but not enough for the preference-network construction the framework specifies (related-artist adjacency, playlist co-occurrence).

Recommended replacement path for audio features specifically: AcousticBrainz's final data dump. AcousticBrainz shut down in February 2022 but published its full ~7.5-million-track dataset (submissions ~29.4 million; dumps of low-level features — BPM, key, MFCCs, ~120 descriptors — and high-level features — mood, genre, instrument tags) under a CC0 licence before going dark, and the dumps remain downloadable from acousticbrainz.org/download. Coverage of Bangladeshi acts in that corpus is unknown and needs a direct check (search the dump for the project's 63 MusicBrainz-linkable artist names) — if coverage is thin or absent, the practical alternative the field has converged on is self-extracting features from source audio using open-source libraries (Essentia, from the same MTG/UPF group behind CompMusic and originally behind AcousticBrainz, or librosa) against YouTube-hosted official audio, which sidesteps Spotify entirely and is exactly the tool used to build the AcousticBrainz corpus in the first place. This is a real project (weeks of engineering, not a credential form), but it is the field's actual current path, and it produces features that are licence-clean and don't depend on any platform's continued goodwill.

4.2 YouTube Data API v3 is comparatively viable, with real constraints stated plainly

Unlike Spotify, YouTube's API was not deprecated in the same way. As of 2026, the free tier gives 10,000 quota units/day per Google Cloud project; a single search call costs 100 units (≈100 searches/day free), while reading a video's statistics costs 1 unit. For this project's actual need — pulling view/like/comment counts for a known, bounded list of official videos per catalogued artist, rather than open-ended searching — the quota is workable without a paid tier: video-statistics reads are cheap, and the expensive operation (search) can be minimised by resolving video IDs once and caching them. This is the more realistic near-term audience-side data source, and the roadmap should distinguish it from Spotify rather than bundling both under one "credentials" blocker. Recommendation: pursue YouTube Data API access first and independently of Spotify; it is both more likely to succeed and sufficient for a first-pass engagement-metric layer (views/likes/comments over time per official upload) that the current BMPN has no equivalent of at all.

4.3 A genuinely new, low-cost audience-data option the roadmap has not considered

Neither the roadmap nor the methodology document considers running a direct, small-scale primary survey — despite this being exactly the method the newly found "Music Consumption on Digital Platforms: A Study on Bangladeshi University Students" paper (§2.3 above) used, cheaply, to produce real audience-side findings for a comparable population. A short structured survey distributed through university music societies, Facebook fan groups for catalogued artists, or in partnership with a journalism contact would not scale to a full preference network, but it would produce the project's first primary audience-side data point of any kind — something no amount of API access substitutes for, because APIs give behavioural proxies, not stated preferences or identity narratives. This is cheap, ethically straightforward under the project's own stated ethics section (public, aggregate, consenting respondents), and directly answerable to the BMEM's "Listener Preference Networks" dimension in a way the co-appearance network structurally cannot be. This is the single highest-value new idea in this document for closing the project's most-repeated limitation (see roadmap item #2 discussion in §7 below).

4.4 Concert data: a real API exists and the roadmap hasn't named it

The roadmap's #2 and #3 priorities (hip-hop/urban events, university festival lineups) are framed purely as manual research tasks. setlist.fm's REST API is free for non-commercial use with a registered API key, indexes concerts by artist/venue/city/country, and uses MusicBrainz IDs for artist identity — which matters because MusicBrainz-artist-linking is a natural companion improvement (§6.3 below). Coverage of small/regional Bangladeshi acts on setlist.fm is almost certainly thin (it is crowd-contributed, and skews toward internationally-touring acts), so this will not solve the hip-hop/university gap outright, but it costs nothing to query against the current 63-artist catalogue and may surface diaspora-tour dates (a category the project already tracks, e.g. Artcell's Canada/Australia tours) that manual search missed. Bandsintown and Songkick have comparable but commercially-licensed data; lower priority given cost.

5. Industry and economic context: filling BMEM Dimension 6

The project's own validation document rates Dimension 6 (Industry Transformation) as "largely untested." This search closes part of that gap with citable, if still needing full-text verification, sources:

Ringback/caller tunes as an industry-defining revenue mechanism. Background research confirms that in the 2000s Bangladeshi mobile operators monetised ringback/caller tunes from label catalogues, initially without rights-holder consent, before an accommodation emerged; ringback tones were a very large (hundreds-of-millions-of-dollars-scale, in comparable South Asian/African markets) revenue category before streaming. The project's current industry narrative jumps from "cassette era" to "streaming" (frameworks/bmem.md Dimension 6) and has no place for this intermediate, telecom-mediated monetisation era, which by several regional accounts was the dominant Bangladeshi music-industry revenue model through roughly the 2000s–early 2010s, arguably more consequential than CD sales. This is a structural gap in the framework, not just a missing citation — recommend adding a "telecom/ringback era" phase between CD and streaming in BMEF's temporal-phases table (§3 of frameworks/bmef.md).
Labels (G Series, CD Choice, Soundtek, Anupam Music, Laser Vision, Chena Sur, Ektaar Music) and their digital pivot are now documented in the 2025 Rajshahi University study (§2.3) and in scattered label-history sources (Wikipedia label pages, Business Standard coverage of labels moving into drama/web-series production as music revenue thinned). Ektaar Music is specifically flagged in background sources as a royalty-payment pioneer — worth its own citation given the project's stated interest in monetisation as an untested BMEM claim.
Bangladesh-specific music-market sizing reports exist commercially (e.g. a "Bangladesh Music Market (2025–2031)" industry report was surfaced) but are paywalled market-research products, not academic sources — flag as a possible paid-access source to check if the project's owner has any institutional access, but do not rely on it uncited.

6. Concrete, prioritised recommendations for the project

Ranked by (evidence strength this document found) × (cost to implement) × (how directly it fixes something the project has already flagged as a weakness). "No new data" items are the highest-leverage because they act on data already in the repository.

6.1 No new data required — pure methodology, do these first

Null-model significance test for modularity (§3.4). Generate degree-preserving random graphs, report observed modularity as a z-score against them. Directly answers the Peixoto critique before a reviewer raises it. New script, ~half a day.
Swap Louvain for Leiden (or run both) in analyze_bmpn.py (§3.3). Cheap, forecloses a specific known defect.
Statistical backbone extraction on the co_billing layer (§3.2) — a Tumminello-style hypergeometric validation pass, reported as a fourth analytical pass alongside observed/full/co_billing. This would let the paper report, for the first time, which specific co-billing edges are statistically distinguishable from what artist/event degree alone would predict — directly relevant to the paper's own doubt about whether Lalon Band/Chirkutt's centrality is "programming policy" (their words) or real structure.
Cite Newman (2001) next to the bill-size correction in docs/methodology.md and the paper. One sentence.
Add the telecom/ringback era to BMEF's temporal phases (§5 above). Framework text edit, no data collection.

6.2 Literature and framework edits — verify then write

Verify and integrate Hossain (2026) on hip-hop and the July 2024 uprising (§2.1) — the highest-value literature addition found. Use it to correct the hip-hop political_or_social documentation artefact the paper's §7.3 already flags, with a citable source rather than an assertion.
Add a Coke Studio Bangla academic citation and the "supply-chain capitalism" framing (§2.2) to Dimension 5/6 discussion in frameworks/bmem-validation.md.
Cite the two 2025 Bangladeshi digital-consumption/industry studies (§2.3) in the methodology and Dimension 6 discussion.
Add a short "Decolonising MIR" paragraph to Limitations (§10 of the paper), citing the ISMIR bibliometric study and the "Missing Melodies" paper (§2.5), reframing the project's audio-data gap as a documented, field-wide structural problem rather than only a local blocker — this strengthens rather than weakens the project's credibility.

6.3 Schema and data-collection changes

Add music_brainz_id (and optionally wikidata_id) as optional artist schema fields. Enables querying setlist.fm and AcousticBrainz programmatically against the existing catalogue with no manual relinking later, and is a near-zero-cost schema addition (data/schemas/artist.schema.json).
Query setlist.fm's free API against the current 63-artist catalogue (§4.4) before doing more manual event research — may surface diaspora-tour dates cheaply.
Design and run a short primary audience survey (§4.3) through university music societies / artist fan groups — the project's first primary audience-side data point, addressing the single most-repeated limitation in the paper (no audience data of any kind) at a cost the blocked APIs cannot match. This should be scoped modestly: a validated instrument on genre identification and listening-community questions, not an attempt to reconstruct the full preference network from survey data alone.
Re-scope the roadmap's API-credentials item. Split "register Spotify Web API and YouTube Data API v3 credentials" into two separate, differently-prioritised items: YouTube (workable now, pursue first) and Spotify audio-features (very likely unobtainable for a new registrant; replace with the AcousticBrainz-dump-plus-Essentia-extraction path instead of waiting on credentials that a new application will not receive).

6.4 Longer-horizon, higher-cost

Explore MERT or comparable self-supervised audio-representation models (§2.5) as a lower-effort alternative to full Essentia feature engineering, once any source audio is obtainable — MERT operates at a fraction of prior large audio models' parameter count and is open-source, making it plausible for a project at this resourcing level, in a way a from-scratch deep-learning pipeline would not be.
If the project ever moves toward tradition-specific musical description (Baul, Bhatiali, Bhawaiya melodic/rhythmic structure — currently absent from the schema entirely), the CompMusic/Dunya model (§2.5) is the right template: build description tools fitted to the tradition rather than assuming Western MIR features transfer. This is explicitly a "next iteration" item, not near-term.

7. What this changes about the roadmap, stated plainly

The existing docs/roadmap.md "Immediate next actions" list treats API credentials as the single blocker unlocking almost everything audience-side. This document's central correction: that is no longer accurate for Spotify specifically, and conflating Spotify and YouTube under one "credentials" blocker obscures that YouTube is tractable now while Spotify's most valuable endpoints (audio features, related artists, recommendations) are closed to new applications regardless of credentials. The roadmap should be updated to reflect this — YouTube pursued directly, Spotify's audio-feature ambition redirected to AcousticBrainz + open-source extraction, and a cheap primary survey added as a genuinely new option the original plan did not consider. None of the network-methodology recommendations (§6.1) require touching this blocker at all and can proceed immediately on the existing dataset.

Sources consulted in this pass

Full bibliographic entries should be added to docs/annotated-bibliography.md after full-text verification; informal citations here, in order of appearance:

Hossain, T. (2026). Voice for justice: Hybrid cultural resistance of hip-hop songs and memes in July mass uprising, Bangladesh. SAGE. DOI: 10.1177/13678779261452687. (Full text not retrieved this pass — journal domain blocked; verify before citing in the paper.)
"Beyond Appropriation and Commodification: The Making of Coke Studio Bangla's Complex Subjects." ResearchGate record 388400913. (Author name and venue need full-text confirmation.)
"Music Industry on Digital Platform: A Study on Five Major Music Labels of Bangladesh." Rajshahi University Journal of Social Science and Business Studies (June 2025). ResearchGate record 392904977.
"Music Consumption on Digital Platforms: A Study on Bangladeshi University Students." International Journal of Research and Innovation in Social Science (RSIS International). (Author names not retrieved; domain blocked.)
Mauch, M., MacCallum, R. M., Levy, M., & Leroi, A. M. (2015). The evolution of popular music: USA 1960–2010. Royal Society Open Science, 2(5), 150081. arXiv:1502.05417.
CompMusic project (Serra, X. et al., 2011–2017), Universitat Pompeu Fabra; Dunya corpora.
"Beyond a Western Center of Music Information Retrieval: A Bibliometric Analysis of the First 25 Years of ISMIR Authorship." Transactions of the International Society for Music Information Retrieval (2026).
"Missing Melodies: AI Music Generation and its 'Nearly' Complete Omission of the Global South." arXiv:2412.04100 (2024).
Newman, M. E. J. (2001). The structure of scientific collaboration networks. PNAS.
Tumminello, M., Miccichè, S., Lillo, F., Piilo, J., & Mantegna, R. N. (2011). Statistically validated networks in bipartite complex systems. PLOS ONE, 6(3), e17994.
Traag, V. A., Waltman, L., & van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing well-connected communities. Scientific Reports, 9, 5233. arXiv:1810.08473.
Peixoto, T. P. (2023). Descriptive vs. inferential community detection in networks: pitfalls, myths and half-truths. Cambridge University Press. arXiv:2112.00183.
Spotify for Developers blog, "Introducing some changes to our Web API" (27 Nov 2024). (Direct fetch blocked by session network policy; corroborated via Music Ally and Digital Music News summaries — verify against the primary post before citing.)
AcousticBrainz project data dumps (acousticbrainz.org/download); MetaBrainz blog, "AcousticBrainz: Making a hard decision to end the project" (Feb 2022).
YouTube Data API v3 documentation and third-party 2026 quota guides (elfsight.com, getphyllo.com, socialcrawl.dev) — verify current quota figures against Google's own documentation before relying on them operationally.
setlist.fm API documentation (api-evangelist.com mirror).
Background industry sources on Bangladeshi ringback-tone monetisation and record labels (G Series, CD Choice, Ektaar Music) — Wikipedia label pages and Business Standard coverage; treat as secondary/background, not citable-grade.

Last updated: 2026-08-27.