Why Research (and Why Now)
Research papers aren't just academic exercises. They're the highest-leverage way to spread methodology. If LEWS works, a paper means other organizations can replicate it. If Open Permit's legal AI approach generalizes, a paper makes it available to the field.
I've spent the last two years building systems that solve real problems across animal welfare, civic tech, CS education, and healthcare. Each of those systems has a methodology — a way of framing the problem, designing the solution, and measuring whether it worked. That methodology currently lives in code, documentation, and my head. A research paper turns it into a citable, replicable, peer-reviewed artifact.
The timing is right for a specific reason: the data exists. LEWS produced real scoring data. Open Permit has 200+ processed permits. The Adventurers Guild has user progression metrics. The Supabase India block produced a natural experiment in infrastructure resilience. I'm not proposing hypothetical research — I'm proposing to write up what already happened.
Paper 1 (Priority): LEWS — Formalizing Lock-in Early Warning for Emerging Animal Technologies
First Priority
LEWS (Lock-in Early Warning System) was built at the EA Animal Welfare Hackathon 2025 to assess lock-in risk for emerging animal farming technologies before they become irreversible. The scoring methodology is novel, the dataset is real, and the team is intact. This is the paper that writes itself.
Research Question
Can lock-in risk be quantified for emerging animal farming technologies before irreversibility? What variables are most predictive of high lock-in trajectories, and how should uncertainty in early-stage assessments be represented?
Dataset
LEWS 7-variable scoring data from the EA Animal Welfare Hackathon 2025. Scored assessments across multiple emerging technologies with uncertainty bounds, historical trajectory comparisons, and variable weighting documentation.
Methodology
Variable normalization across the 7-factor scoring framework, uncertainty integration using credence-based weighting, historical trajectory comparison against known lock-in cases. The methodology section is mostly documented in the LEWS system itself.
Target Venue
Animal welfare journals, EA research forums (EA Forum long-form), arXiv cs.AI for the technical methodology. Workshop track at a relevant conference as a first step before full journal submission.
Co-authors
LEWS team: Tim, Thomas, Rapha + project lead. All intact from the hackathon. Roles: Abid (technical methodology, AI components), Tim + Thomas + Rapha (domain expertise, welfare framework grounding, review).
The LEWS hackathon produced real scoring data. The methodology paper writes itself — we just need to formalize what we already built into a structure that passes peer review.
Paper 2: Open Permit — AI-Assisted Legal Objection Generation at Scale
High Priority
Open Permit has processed 200+ real planning permits across 8+ countries and built 40 jurisdiction-specific legal frameworks. That's a dataset unique in the world — no one else has systematically applied LLMs to cross-jurisdictional civic objection generation at this scale.
Research Question
How effective is LLM-generated legal content for civic objection letters across jurisdictions? What factors most affect letter quality, legal accuracy, and jurisdiction coverage completeness?
Dataset
200+ permits across 8+ countries, 40 legal frameworks with jurisdiction-specific citation formats, objection letter quality metrics from human review, permit type distribution across factory farming, development, and environmental categories.
Methodology
Human evaluation of letter quality by legal practitioners in target jurisdictions, legal accuracy assessment via citation validity checking, jurisdiction coverage analysis — how well does the system generalize to jurisdictions not in the training corpus?
Target Venue
ACL workshop on legal NLP, or ICAIL (International Conference on Artificial Intelligence and Law). Both have strong communities for exactly this problem space.
Unique Angle
First systematic study of cross-jurisdictional AI civic advocacy. No comparable paper exists — the closest work is in legal document summarization, not objection generation. The civic advocacy framing + multi-country scope is genuinely novel.
Paper 3: Speciesist Language Detection — Extending Inclusive Language Tools
Medium Priority (Open Paws work)
Open Paws maintains a no-animal-violence rule corpus for language moderation. The underlying NLP problem — detecting speciesist framing in text — is analogous to existing racial and gender bias detection work but has received almost no systematic treatment in academic NLP.
Research Question
Can NLP tools reliably detect speciesist framing in text, analogous to racial and gender bias detection? What are the precision/recall characteristics, and where do false positives cluster?
Dataset
Open Paws no-animal-violence rule corpus (annotated examples), supplemented by academic grounding from Hagendorff (2022), Takeshita (2023), and Leach (2024) speciesism NLP papers. Existing academic work provides the theoretical frame; Open Paws provides the practical dataset.
Methodology
Rule-based detection pipeline (current Open Paws implementation), false positive analysis across different text types (news, social media, academic), comparison with existing bias detection toolkits (Perspective API, Detoxify) on the same corpus.
Target Venue
FAccT (Fairness, Accountability, and Transparency), or ACL BioNLP workshop. FAccT is the highest-signal venue for this type of bias detection work; the animal welfare framing would be novel in that community.
Paper 4: Infrastructure Abstraction for Resilient Multi-Service Deployments
Medium Priority
The Supabase India block in 2025 took down 29 services simultaneously and forced a full infrastructure migration. That incident produced a natural experiment in cloud provider dependency, migration complexity, and architectural patterns for regional resilience.
Research Question
What architectural patterns minimize downtime risk when a cloud provider becomes unavailable in a specific region? How does migration complexity scale with service count and data interdependency?
Dataset
Supabase India block incident data (3-day outage, 29 services affected), before/after latency measurements, migration timeline per service, cost analysis pre- and post-Yotta migration, architectural diagrams before and after.
Methodology
Case study methodology documenting the incident, migration execution, and architectural changes. Before/after latency comparison, cost analysis, migration complexity scoring by service type. Generalization section: what patterns would have prevented the incident.
Target Venue
IEEE Cloud Computing, or USENIX SREcon. SREcon is particularly relevant — it's practitioner-focused and values real incident postmortems as research contributions.
Paper 5: Gamification as Mechanism for CS Skill Development in Emerging Markets
Future Priority
Adventurers Guild uses RPG-style progression (F-rank to S-rank) to structure CS skill development. As the platform passes 500+ users, longitudinal data on engagement and skill acquisition is accumulating — enough to ask whether the gamification design actually produces different outcomes than traditional approaches.
Research Question
Does RPG-style progression (F→S rank) improve sustained engagement and skill acquisition compared to traditional internship or course models? What rank transitions correlate with measurable skill jumps?
Dataset
Adventurers Guild user progression data: quest completion rates, rank advancement timelines, dropout rates by rank, skill assessment results before and after rank progression events. Cohort data from first 500 users.
Methodology
Cohort analysis comparing engagement metrics across rank levels, skill assessment before/after rank progression milestones, dropout analysis, comparison with published engagement metrics from non-gamified CS learning platforms.
Target Venue
ICER (International Computing Education Research), or CHI. ICER is the top venue for computing education research; CHI is broader but highly respected for gamification and engagement work.
How to Start Writing
The biggest obstacle to writing the first paper isn't skill — it's starting. This is a concrete 6-step path for Paper 1 (LEWS) that avoids the paralysis of "I need to do more research before I can write."
Pick Paper 1 (LEWS) — team is intact, data exists
The LEWS team (Tim, Thomas, Rapha) worked together at the hackathon and all expressed interest in formalizing the methodology. The scoring data exists. The methodology is documented in the system. Activation energy is lowest here.
Write the Methods section first
The Methods section is already implicit in the LEWS blog post. Start there — describe the 7-variable scoring framework, the uncertainty integration approach, and the historical trajectory comparison in formal academic language. The rest of the paper structure falls out from a solid Methods section.
Get a senior researcher as co-author or advisor
EA network, GTU faculty, or Open Paws academic connections. A senior co-author provides: (a) access to venues that require institutional affiliation, (b) peer review credibility, (c) accountability for actually finishing. Approach via EA Forum or direct email to relevant researchers.
Submit to arXiv first
arXiv establishes a timestamp and gets the work in front of the community before formal peer review. EA Forum can amplify the arXiv post. Community feedback from EA Animal Welfare researchers is accessible and constructive — use it to improve the paper before journal submission.
Iterate based on reviews
EA Animal Welfare community is unusually accessible and engaged. Post to EA Forum, share in relevant Slack channels, reach out directly to researchers cited in the paper. Incorporate substantive feedback into the revised version. This iteration loop is faster in EA than in most academic communities.
Target a workshop paper first
Workshop papers have lower bars than full conference papers, faster turnaround (3-4 months vs 12+), and provide genuine community feedback. One accepted workshop paper builds the publication record needed for full conference submissions. Don't aim for Nature on the first paper — aim for published.
Target Venues
| Venue | Type | Relevance | Difficulty |
|---|---|---|---|
| arXiv cs.AI | Preprint | All papers | Easy (no review) |
| EA Forum | Community | LEWS, Open Permit | Easy |
| Animal Welfare journals | Peer-reviewed | LEWS | Medium |
| ACL workshops | Workshop paper | Open Permit, Speciesism | Medium |
| FAccT | Full conference | Speciesism, Gamification | Hard |
| ICER | Full conference | Adventurers Guild | Hard |
| USENIX SREcon | Full conference | Infrastructure | Hard |
Key Lesson
The hardest part of academic writing isn't the writing — it's realizing you already have the data, the methodology, and the results. You just haven't written it up yet. That's where we are with LEWS. The scoring framework is novel. The dataset is real. The team is intact. The only missing piece is sitting down and writing it in academic format.
If you're a researcher working in animal welfare tech, civic AI, or inclusive language NLP — reach out. These papers are better written collaboratively. I have the systems, the data, and the implementation. I'm looking for co-authors who bring domain expertise, academic affiliation, or experience navigating the specific venues. Email or EA Forum DM.