Research
Read the governing sources, map uncertainty, and separate evidence from assumption.
About RN Collins + Aloha AI
Work on questions too messy for one discipline.
RN Collins is a researcher, educator, builder, and law student working across neuroscience, policy, technology, and organizational decision-making. She researches an entity and its work, finds the gaps or pressure points, makes them concrete and visible, and creates what the situation requires.
The practice
One method, four forms of work.
Decision design
Name the people, authority, tradeoffs, stop conditions, and review points before selecting software.
Implementation
Build a bounded, reversible path with human review and a record someone else can inspect.
Education
Turn the reasoning into tools and learning materials people can use without surrendering their judgment.
How the work is actually done
Four rules that show up in every build.
These are not values statements. Each one is visible in the artifacts below and on this site, and each one has a cost — it makes the output smaller, later, and less flattering than the version that skips it.
Read the source, not the summary
A framework, an agency record, a court order and a vendor blog post about that court order are four different objects. The practice reads the first three and treats the fourth as an unverified pointer. The Source Desk on this site names all seventeen records it relies on, what each one is used for, and where each one stops.
Publish the limit next to the claim
Every build carries the thing it cannot establish on the same surface as the thing it can — not in a footer, not in terms of service. When that pairing cannot be written honestly, the claim gets narrowed until it can.
Make “do not proceed” a real output
A decision aid that can only produce encouragement is a sales tool. Five specific answers in the Opportunity Studio override its own score entirely; the vendor and pilot instruments carry gates that stop the process rather than deducting points.
Report the gap rather than hide it
A monitor that finds nothing looks identical to a monitor that is broken. Where a source could not be retrieved, a link is dead, or a sweep did not run, the build says so in words — because a quiet page is the most misleading artifact in this kind of work.
Public evidence
Selected work you can verify.
These links go to publishers, a federal index, or a government record—not self-authored claims.
Disability in Radiology
Peer-reviewed research indexed by the U.S. National Library of Medicine.
Psychedelics as medicine
Science-policy analysis published by the MIT Science Policy Review.
Caregivers in research funding
Open-access policy proposal published in Frontiers in Education.
NYC Youth Agenda research
Public civic-research record from the City of New York.
Self-published builds
What the practice produces, open to inspection.
12 public builds, each live and each carrying its own stated limit. The figures were read off the pages themselves on August 29, 2026; the limits are quoted from what each build says about itself. Two of them — a hand-read authority monitor and an unreviewed automated tracker on the same subject — are here specifically because the difference between them is the whole point.
These are self-published. Opening one confirms that it exists and says what is described here. Unlike the four records above, no outside publisher, index or government body has vouched for them, and this section does not ask you to treat them as though one had.
Four kinds of object recur across them. Monitors watch a field that keeps moving and report the date of their last successful collection separately from their last attempt. Registers fix a snapshot, say when it was taken, and refuse to quietly update it. Decision tools make one judgement inspectable and export it as a file the user keeps. Atlases map a body of knowledge and publish the size of the hole in it — which is how a link check across 132 catalogued entries became a headline finding of 58 broken or missing recordings rather than a maintenance chore.
The counts are load-bearing, and each one is a number some other version of the same page would have rounded away: eight authorities of which seven were read and one is flagged unread; 175 organisations surviving from 177 recorded; 66 builds enumerated by a script after a hand-kept list had drifted to 32. Where a figure is smaller or uglier than the round number nearby, that is the figure.
Build index
66 builds · regenerated 29 August 2026
The complete estate, generated rather than curated: a script asks GitHub what repositories exist and Vercel what is deployed, cross-references the two, checks that each production URL answers, and writes the data file the page renders from. Each entry keeps its own status in its own words — retired prototype, paused with the back end off, research preview — instead of being flattened into a uniform present tense.
Where it stopsIt replaced a hand-kept list that had drifted to 32 entries while the estate held more than twice that. Presence in the index means a repository and a deployment exist and the URL answered; it is not a claim that a build is finished, maintained, or in use.
Legal AI Guidance Monitor
8 authorities · 7 opened and read · 1 flagged unread
Primary authorities on lawyers' use of AI. Each record separates what the document holds from how far it can be stretched, and distinguishes advisory ethics opinions from a standing order that binds one courtroom and a published sanctions order.
Where it stopsThe eighth authority could not be retrieved and is listed as unread rather than quietly dropped. Ethics opinions interpret rules of professional conduct; they do not enact them.
AI Governance Tracker
5 saved searches · weekly · 0 reviewed before publication
The deliberate contrast to the monitor above: five saved searches run once a week across the open web, published newest first without human review, and labelled as such at the top of the page.
Where it stopsResults mix commentary, law-firm client alerts and vendor material with official documents. It is a place to start research, not a record of what an agency has done.
Psych Ops Source Watch
6 source profiles · 5 failure modes behind a quiet sweep
A regulatory monitor that publishes its own blind spots: what each watched body actually controls, what a change there would mean, and the five different things an empty sweep can mean — only one of which is reassuring.
Where it stopsDiscovered items stay unreviewed candidates until a person confirms the document, its date, its jurisdiction and its operational consequence. Silence is reported as unknown, never as all-clear.
Platform Moderation Evidence Lab
5 records · 5 reading rules · 1 dead source kept and marked
Five public records about platform moderation of drug-policy content, each paired on the same card with the claim it is routinely asked to support and cannot.
Where it stopsOversight Board decisions reach the Board by appeal or referral, never by sampling, so they cannot carry a claim about how often anything happens. A published policy is a ceiling, not a measurement.
Psychedelic Operations Directory
177 recorded · 175 after de-duplication · 23 categories
A field survey recorded once and published unchanged. It publishes only what can be stated without a per-claim source: name, category, a scope descriptor that makes no claim, and a destination link where one answered.
Where it stopsInclusion is not vetting, endorsement, licensure verification, or a finding about quality, safety, solvency or compliance. Pricing, credentials and rankings are withheld because no source was recorded for them.
AI Build Cost Workbench
7 steps · dated provider rate catalog · scenario comparison
Turns an early AI idea into a reviewable cost scenario across labor, model usage, infrastructure, vendors, operations and risk — with every rate exposed as an editable assumption rather than a project-type price sticker.
Where it stopsThe rate catalog is a hand-typed, versioned snapshot of published list prices. It is not a live feed, not a quote, and it does not know your contract.
Drug Market Transition Evidence Lab
6 judgement dimensions · 1 primary source required per record
Builds a source-anchored record of one policy, rule, court decision or market event, then makes six judgements inspectable — each requiring an evidence note, with "unknown" a permitted result.
Where it stopsIt does not claim live monitoring, automatic refresh, or independent verification of any outcome. A policy announcement is not a market outcome.
Entheogen Atlas
10 substances · 57 centuries of confirmed use
Maps the archaeology, neuroscience, case law and phenomenology of ten substances across the span from the oldest chemically confirmed psychedelic use to the present.
Where it stopsDosing, interaction clearance, individualised legal-risk scoring and live-law tracking were deliberately removed from the edition, and the page says why.
Fadiman Knowledge Atlas
132 entries · 1968–2026 · 58 dead or missing URLs
A catalogue of 58 years of one researcher's recorded public output, with a link check on every entry — which is how the finding emerged that a large share of the record is no longer retrievable.
Where it stopsIt is a proposal for knowledge infrastructure, not a commissioned archive. Cataloguing a paper does not validate it; the atlas says so on the microdosing study it lists.
Zero → Frontier
11 sections · 237 curated resources · 0 sign-ups
A source-forward AI literacy guide with a route builder that shapes a reading path around purpose, stakes and available time, with no account and no email.
Where it stopsEducational reference only. Every resource carries a named source and a visible freshness limit rather than an implied currency.
Myelin Learning Lab
4 courses · 6 field briefs · 3 builders
Four applied courses built around the decisions responsible AI work actually requires — verify the claim, classify the data, define the human control, leave an evidence record — plus printable field briefs and three builders.
Where it stopsNon-credit education only. No account, identity verification, accredited certificate, CE credit, or licensing-board reporting.
The operating stance
Evidence before theater. Context before software. A bounded build before transformation claims.
Hawaiʻi is the place from which the practice operates and a reminder that technology enters living systems with histories, obligations, and uneven consequences. It is not a claim to speak for all of Hawaiʻi or decorative brand material.
This page identifies selected public work; it is not a complete curriculum vitae. Aloha AI does not present RN as a licensed attorney, auditor, certifier, or substitute for regulated professional advice.