OVERVIEW.md

Overview — what this lab is doing, where it stands, and how to join

A five-minute orientation for anyone deciding whether to contribute. For this week's deadlines, use NOW.md. For navigation, use MAP.md and the directory in README.md; this page is for judging whether the program is worth your time. Outward manuscripts live under submissions/.

In two minutes

This lab asks one question about coordination: when a worker, a mediating system, and a counterpart work together, is the arrangement irreducible — does it genuinely bind all three — or does it factor into independent two-party pieces? It answers by modeling the arrangement as a small Boolean dynamical system and computing exact integrated information (Φ, from IIT 4.0, via PyPhi). A form is triadic (irreducible; the lab's name for it is algorithmacy) or dyadic (it factors; literacy). On systems small enough, that verdict is exact, not a proxy.

The work runs as four programs on that one question, each a different lens on the same arrangement. The computational program fixes a protocol — a literature pass, five hypotheses set before any computation, an instrument-validated run, a paper — and has carried 74 questions through it, with a measure-validation arc that established Φ as the instrument first and six synthesis papers tying threads together. The qualitative program reads real coordination settings against that catalog of priors, on the questions of process and meaning fieldwork answers. The recurrence program pairs Φ with cross-recurrence quantification, a behavioral measure of how the parties' states track each other over time and which one leads. The survey program measures algorithmacy as a lived competence in real workers through self-report, the measurement side of the validation gap. Structure, fieldwork, behavior, measurement, on one arrangement, with Φ the structural instrument the qualitative and recurrence arms are built around.

The honest standing

The distinguishing feature of this lab is that it tells you where it is weak, and you should know that before contributing.

• The instrument is real and the results are exact and reproducible. Every number re-derives from a committed script under CI. That part is solid.
• Most individual "laws" are rediscoveries. When the program ran the literature on its own results, it found that bidirectional coupling is IIT's own requirement, that pivotality and substitutability are cooperative game theory's (the Shapley value), that the platform-position theory is platform economics' competitive bottleneck and the outside-option principle, and that the commit-versus-convey distinction is the communication literature's own. The synthesis paper states this plainly.
• A little is genuinely new. A graded membership relation (complex membership rising with causal pivotality, where IIT membership is per-form binary) and an extremes-only quorum law. Both are bounded; see the synthesis and its committee review for how much weight they bear. The recurrence program adds a newer one: pairing exact Φ with cross-recurrence quantification is unclaimed in the literature. It began on synthetic trajectories and has since run on real open-source coordination data (v8–v10), where the veto-player and disintermediation priors appear in a real organization and governance style changes the gate. That is the first real-data evidence in the lab, on a methodological side rather than the core thesis.
• The real contribution is the bridge, not any single law. One exact, computable criterion — Φ irreducibility on a coordination model — turns out to be the common image of five separate disciplines' tests, and it names which parties are bound (the major complex), which a one-line heuristic cannot.
• The validation gap is open. Everything is in-silico: evidence about Boolean models of coordination, not measurements of organizations. No real worker, platform, or message has been measured. Closing this is the program's largest open need.

If you want a program that overclaims, this is not it. If you want one that subjects its own results to deep research and an adversarial committee and reports what survives, read the synthesis and its committee review and decide.

What is already here

The instrument is the shared base; the four programs build around it.

• The instrument and its validation — foundations/: a complete manuscript showing no cheap proxy recovers exact Φ, so the exact computation is needed. The classifier, controls, and the major-complex reading live in org_frontier/classifier/. The benchmark pin is PyPhi's feature/iit-4.0 line at b78d0e342 (pyphi.Network, Subsystem, new_big_phi). Ternary and mixed-radix forms use third_party/pyphi_iit4_mv, also packaged on rogerSuperBuilderAlpha/pyphi@feat/multivalued-iit4-radix-k. Upstream PyPhi 2.0 is wmayner/pyphi main at 134a8f31 (tag v2.0.0rc4): pyphi.iit4_2023() reproduces the pinned controls, and pyphi.iit4_2026() is the Mayner, Marshall, and Tononi (2026) default.

The computational program — exact Φ on Boolean models, the structural reading and the prior catalog:

• The logbook — 74 questions in org_frontier/questions/, each a full paper, plus a per-probe log in org_frontier/probes/PROBES.md, and the thread and study corpus the catalog of priors is drawn from.
• Synthesis & reviews — the six pipeline papers (core-membership law, coordination-logic atlas, discriminant boundaries, platform position, the field protocol) and the cross-program synthesis.

The qualitative program — real settings read against the priors:

• The qualitative arm — org_frontier/qualitative/: a methods guide, an open topic agenda, and worked study proposals, where a real coordination setting is read against a named prior and the prior is held open. A study may stand on its own or feed the field protocol toward a verdict.
• The field arc — org_frontier/field/: a protocol for modeling one real coordination arrangement and the verdict it yields, demonstrated on ten mock organizations, with the validation gap kept explicit.

The recurrence program — Φ paired with a behavioral measure of coordination:

• The recurrence arm — org_frontier/recurrence/: cross-recurrence quantification paired with exact Φ. Φ reads a coordination's structure; cross-recurrence reads a run of it — how strongly the parties' states track each other, and which one leads. A corpus sweep, twenty experiments, and the bridge at four and five parties show where structure and behavior agree and where they part. It has since read real coordination data: three open-source projects' commit, pull-request, and review histories (v8–v10), where the veto-player and disintermediation priors appear and a heavy review process spreads the merge gate and binds a deeper core. The pairing is unclaimed in the literature.

The survey program — measuring the competence in people:

• The survey arm — org_frontier/survey/: the human-subjects arm. It carries algorithmacy from theory into measurement, a self-report scale for the construct's three facets — counterpart inference, signal compression, rule-change tracking — fielded on people inside a real coordination arrangement. Its first study is a pre-registered three-wave panel, the instrument and hypotheses committed before any data. This is the measurement side of the validation gap; no data is collected yet.

Where you could contribute

Concrete open threads, roughly easiest to hardest. Each is a real gap, not busywork. The empirical and bridge arms below each come as a ready-to-run handoff packet — a front-door README, the pre-registration discipline, and a runnable scaffold that takes a real input to a verdict. The five are indexed in org_frontier/HANDOFF_PACKETS.md.

• Pick an open question. The pre-written agendas are answered — all fifty in RESEARCH_AGENDA_50_V2.md and items 1–12 of RESEARCH_AGENDA_V4.md — and each answer names the residual it leaves. V4 item 11 is answered (GRADED_HOLDS: the joint-observation cliff tracks duty correlation, not channel granularity). V4 item 12 is answered (M3_FLIPS_VERDICT: induced exact Φ already decides V4 items 1–3; overlay background-subset Φ moves the V4 item 1 subset-estimand key). Scaffold one with python -m org_frontier.protocol.new_question and run it through the protocol.
• Open the Ergodicity × Algorithmacy track. Forty questions on when ensemble coordination statistics transfer to an individual trajectory (org_frontier/ergodicity/). The first cell — studies/ergodic_components_vs_phi/ — asks whether triadic forms differ from dyadic forms in attractor / basin structure; hypotheses are fixed, not yet run.
• Stress a standing result. The extremes-only quorum law (study D) uses clean threshold counts; does it survive weighted or noisy quorums? The membership law (study A) uses a single-node influence proxy; does a higher-order pivotality measure sharpen the Shapley correspondence past the null-player corner? (A first pass exists in org_frontier/threads/shapley_membership — the exact Shapley value reaches AUC ≈ 0.87 vs ≈ 0.63 for single-node influence; extending it to larger forms and a coalition-formation account is open.)
• Settle a construct question the committee flagged. Find one coordination arrangement that is algorithmacy but not directive algorithmic management (or show none exists). That case is what would establish the construct's marginal value.
• Run the first real field case. The field protocol (org_frontier/field/PROTOCOL.md) is ready and its weakest step — eliciting determination rules from evidence with reported inter-rater reliability — is specified but undemonstrated. A single honest real-world case, run and reported, is the highest-value contribution available, because it begins to close the validation gap.
• Bring qualitative methods. The qualitative arm (org_frontier/qualitative/) opens the program to interview-based, ethnographic, and document work — reading a real coordination setting against a named prior, as a stand-alone description or upstream of a field verdict. The methods and an open agenda of settings are laid out for contributors.
• Bring the recurrence instrument a new real series. The recurrence arm (org_frontier/recurrence/) has read three open-source projects' event histories (v8–v10), confirming the veto-player and disintermediation priors and measuring how a heavy review process changes the gate. The open edges are a different kind of coordination: a bot-merged project where the merge actor is a machine, a physiological series such as the neonatal vitals, or a dispatch or handoff log with a named gatekeeper — ideally where the determination actor is recorded so both instruments run.
• Field the survey with a real cohort. The survey arm (org_frontier/survey/) has a pre-registered three-wave panel ready, with the instrument, scoring, and hypotheses committed. It needs a cohort that coordinates through a system it does not control, and the access and ethics clearance to run it. Collecting the first wave is what turns the algorithmacy scale from an instrument into a measurement.
• Push past the size ceiling. Exact Φ is feasible to ~10–12 elements. The proxy-bridge attempt (foundations) does not yet recover the verdict from cheap signals; making it work would unlock larger arrangements.
• Review, don't just submit. The program runs a reproducibility-first public review; maintainers and reviewers are needed (see MAINTAINERS.md).

How to start

1. Read GETTING_STARTED.md — environment, the instrument-validation gate, the model in one paragraph, and the contributor path end to end.
2. Read one worked question (e.g. org_frontier/questions/q43_thompson_interdependence/) and the research protocol.
3. Scaffold your own with python -m org_frontier.protocol.new_question, commit your hypotheses before computing, register your numbers for CI, and open a pull request into contrib per PUBLISHING.md.

The pitch, in one paragraph

A coordination arrangement is irreducible when a mediating system commits a determination that binds all parties, and reducible when it merely conveys while the parties stay separable. This lab makes that line exact and computable, names which parties a real arrangement would bind, and submits every claim to deep research and adversarial review before keeping it. Most of what it has found, other fields already knew in their own terms; the lab's contribution is to compute the same line once, across all of them, on explicit models — and the largest thing left to do is to point the instrument at a real organization and report honestly what breaks.