A method, run end to end.
The Algorithmacy Lab is a public research program built around a method: a six-stage, AI-assisted protocol that takes one question from prior-work review to a finished quantitative paper, with five falsifiable hypotheses fixed before any computation and refutations reported as first-class results. The protocol is grounded in established norms — Platt's strong inference, Chamberlin's multiple working hypotheses, pre-registration — and is made runnable, end to end. The thesis below is the domain it is currently applied to; the protocol is the contribution.
The method.
One question, six stages, run as a pipeline. Each stage is specified so the whole thing can be executed start to finish rather than described in the abstract. The discipline is front-loaded: the hypotheses are committed before the instrument is run, so a result can disconfirm them.
-
01 · Review
Survey the prior work the question sits in.
-
02 · Deep research
Read the literature closely enough to state a real, contestable claim.
-
03 · Hypotheses
Fix five falsifiable hypotheses before computing — so the test can fail.
-
04 · Methods
Specify how each hypothesis will be tested against the instrument.
-
05 · Run
Execute the tests against the exact-Φ ground-truth instrument.
-
06 · Paper
Write a full paper — nulls and refutations reported as first-class results, not omitted.
The full write-up: How to research with AI → · The protocol spec: RESEARCH_PROTOCOL.md →
The thesis it's applied to.
The program's domain is a single, made-computable claim: a coordination form is either dyadic — it factors into independent pieces, and so demands literacy — or triadic — it is structurally irreducible, and so demands algorithmacy. To make that testable, a worker–system–counterpart arrangement is modeled as a small Boolean dynamical system, and one reads whether its most-integrated state factors apart (dyadic) or stays bound (triadic).
The verdict is computed with exact integrated information — Φ from IIT 4.0, via the PyPhi library. It is an exact reading of the model, not a proxy or an approximation. The foundations/ arc holds the measure-validation experiments that established this instrument before it was turned on the organizational questions.
The foundations arc: foundations/README.md →
The track record.
- 134+ logged probesEach a small experiment with its own question, hypothesis, method, and result, recorded in the logbook.
- Six research programsIterative arcs of questions, each building on what the last one settled or unsettled.
- A 50-question agendaA standing list of open questions the program is working through.
- ~1/3 nulls and refinementsAbout a third of the results are nulls or refinements, logged as such. That honesty is part of the design, not a footnote to it.
The experiment catalog: pyphi_org_theory_catalog.md → · The logbook: PROBES.md →
A worked example.
The first question taken fully through the automated pipeline asked whether the dyadic/triadic verdict reproduces Thompson's (1967) ordering of interdependence — pooled, then sequential, then reciprocal, in rising order of coordination demand.
It does not. Four of the five pre-registered hypotheses were confirmed and one was refuted; under a uniform convention the verdict scores pooled as dyadic and ranks both sequential and reciprocal as triadic, collapsing the top of Thompson's ordering. The headline is a negative result about a classic typology — which is exactly what the protocol is built to surface and report. Read it as a worked example of the method, not as a settled claim about coordination.
The pilot paper: Thompson's interdependence types against the dyadic/triadic verdict →
Scope & limitations.
Read these results for what they are. They are in-silico results on small Boolean models, computed exactly. That exactness is internal: when two models agree, that is internal validity, not evidence about any real organization. A validation gap separates these computations from empirical claims about firms, teams, or platforms.
- Proof-of-method, not findingsThis is a research agenda and a runnable protocol — not peer-reviewed findings about real organizations.
- Exact, not approximateThe Φ verdicts are exact for the models specified; the open question is whether the models capture the things they stand for.
- The thesis is not provenNothing here should be read as the dyadic/triadic thesis being empirically established. It is being tested, in a controlled in-silico setting, in the open.
The repository.
Everything is public. The lab proper lives in org_frontier/; the instrument-validation work lives in foundations/. Start at the repository, or go straight to a document.
- algorithmacy-lab →The repository — open code, probes, and papers.
- How to research with AI →The method write-up.
- RESEARCH_PROTOCOL.md →The protocol specification.
- Experiment catalog →The full catalog of experiments.
- PROBES.md →The logbook — question / hypothesis / method / result.
- Pilot paper →The Thompson interdependence worked example.
- foundations/README.md →The instrument-validation arc.
Convened around the same coordination form: the conference tracks →