The economics half is done.
The half that matters isn’t.

Institutional host / PI

You’d own: Affiliation, IRB cover, and eligibility for the funders that matter.

Almost every serious funder requires either a PhD-holding PI or an institutional host. That — not money — is the binding constraint right now, and it is the reason a fundable design is sitting unfielded. A host also usually brings the one thing that would compress the timeline most: an existing sampling frame that can be re-contacted instead of built from scratch.

The ask: A department or centre willing to host, plus a senior co-investigator.

Qualitative / sociological lead

You’d own: The interpretive core of P1, and P2 outright.

The economics half of this is done. The half that matters — what people give up first, who inside the household absorbs it, whether the grievance is framed as class, sector, party or geography — needs somebody whose actual training is in it. Social movements, moral economy, gendered labour, or urban informality all fit.

The ask: Co-investigator on P1; lead on P2 if the e-rickshaw strand is your thing.

Agricultural economist / agrarian sociologist

You’d own: P3, the Boro strand — and it has a hard November deadline.

This is the channel most likely to turn into a 2027 food-price and foreign-exchange problem, and almost nobody is watching it. Planting decisions get made in November at Tk 135 diesel against a frozen procurement price. If nobody is in the field by then, the decision that sets 2027 goes unmeasured and can only be reconstructed afterwards from what people remember.

The ask: Lead or co-lead P3; a route to DAE / BADC data would be worth as much.

Bangla NLP / computational social scientist

You’d own: P4 — where there is already a corpus, a pipeline and a known problem.

1,996 articles are collected and a v0 frame classifier runs, but it is a keyword dictionary and it is not good enough: against a single-coder pass it scores F1 0.36 on government-failure framing and 0.20 on apology/empathy. It needs a BanglaBERT-class model, a validated sentiment layer, and two human coders to get kappa above 0.7. The corpus is built; the hard part is open.

The ask: Model work, or simply two coders willing to work through the codebook.

Survey partner in Bangladesh

You’d own: Fielding W0 within the window.

A CATI-capable survey firm or research centre that can run a 15–20 minute bilingual instrument across 1,500–2,500 households, stratified by the main earner’s sector, with quotas that do not quietly exclude women and the poorest. Speed matters more than scale here: a smaller W0 fielded in October beats a larger one fielded in December, because the larger one is no longer a baseline.

The ask: A quote and an honest answer on the earliest realistic field date.

Built and public

• A sourced impact analysis across inflation, fares, agriculture, industry, power and macro, with every derived figure marked and its method shown
• A full audit log — every search, every figure, its source, and every derivation behind the analysis
• Six costed study designs with sampling, instruments, analysis plans and ethics
• P4’s corpus: 1,996 articles across Prothom Alo, Samakal and The Daily Star, 2022 and 2026 windows, collected politely with robots.txt and crawl-delay respected
• A working pipeline — volume timeline, frame shares with Wilson intervals, exploratory NMF topics, a stratified coding sample and a validation script
• A mapped landscape of institutions, adjacent researchers and funder eligibility

Known to be missing or wrong

• No institutional host, no IRB, no PhD-holding PI — the binding constraint
• P4’s frame classifier is a keyword dictionary and fails on most frames (F1 0.36 government-failure, 0.20 apology/empathy). Only protest holds up, at 0.95
• No human coding yet, so no kappa — the single Claude pass is a stand-in, not validation
• No validated Bangla sentiment model
• No Facebook or YouTube data; Meta Content Library needs an application, YouTube needs an API key
• Protest “events” are a news-mention proxy, not ACLED event counts
• Corpus skews to Prothom Alo and to Dhaka-centred, centrist outlets. It is not “Bangla public discourse”
• Zero household data. That is the whole point.

Authorship

Whoever does the work is on the paper. A lead on P2 or P3 leads that paper. The existing analysis is groundwork, not a claim on anyone else’s output.

Data and code

Code and derived metadata are public. Article text stays unpublished for copyright, and anything identifying stays unpublished for ethics. De-identified survey microdata is intended for public release after the first papers.

Ethics

Consent in Bangla, recorded verbally for phone waves. The e-rickshaw and charging work touches illegal activity, so names are never stored alongside it, diaries are encrypted, and charging-point locations are never published. Partisanship is never stored alongside identifiers. Respondents get paid for their time.