Research agenda

Principle #

Do not try to test “Space Immanence” all at once. Break it into components.

Hypothesis families #

H1 — Phenomenological clustering

Sustained contemplative inquiry into self and world falls into two basic orientations:

• toward represented content / world / this;
• toward representational activity / awareness / that.

Test: Collect reports from several traditions. Have researchers classify them without knowing the tradition, control for translation and modern nondual vocabulary, and compare expert and novice reports.

Risk: The categories may come from the interpreter rather than the reports.

Related research: Hood’s Mysticism Scale measures Stace’s (1960) introvertive and extrovertive types of mystical experience, a partly overlapping distinction (unity without content, and unity found in the perceived world). Factor analyses in several cultures give mixed results on whether the two separate, and Stace’s treatment of Advaita and of a Zen satori example does not independently support assigning those traditions to different types. The literature’s relevance to H1 depends on a mapping from those types to the two orientations that has not been validated (note).

Status (October 2026): First tests on public meditation data (Gamma and Metzinger's MPE-92M questionnaire, Zen against TM practitioners) are in the lab record: one run invalid, one stopped by its agreement rule, one held at its outside read, and an exploratory analysis giving small difference estimates that depend on how disputed questions are coded. None is a filing-compliant test. Version 2.4 states the pairing in Part IV: this is the orientation toward represented content, that toward awareness. That matches the tests’ pre-registration; until v2.4 the list above stated the opposite (note). The author fixed the pairing after the exploratory analysis had been seen, and under it that analysis leans slightly toward the prediction. The choice rests on the paper’s other readings and the traditions’ usage and not on the data. Stace’s classic typology of mystical experience (Mysticism and Philosophy, 1960) gives the split no independent support: it treats Advaita’s pure consciousness as introvertive, and an outside read of the text found its Zen satori example discussed on the same side (note). No claim moves on these results, and a confirmatory test must be filed fresh.

H2 — Human-specific versus universal topology

The two orientations reported in experience may be shared by all consciousness-relevant folds, or may arise from human embodied attention.

Test: Compare human meditation reports, animal self-model indicators, collective cognition, and artificial agents.

Risk: Non-human topology may be inaccessible through human categories.

H3 — Predictive-processing implementation

Human folds are implemented through self-models that make predictions at several levels.

Test: Compare self-model dynamics under meditation, depersonalization, psychedelics, flow, panic, and ordinary selfing. Measure changes in precision weighting, interoceptive inference, boundary formation, and metacognitive access.

H4 — Meditation / psychedelic route distinction

Some altered states may loosen constraints on self-modelling by reducing activity. Others may do so by saturating or overloading existing models.

Test: Compare nondual meditation, 5-MeO-DMT, psilocybin, deep absorption, and ordinary metacognition using EEG/MEG/fMRI plus high-resolution phenomenology.

H5 — AI aperture criteria

An artificial system is a candidate phenomenal aperture only if it meets the working paper's four fold criteria — integrated self-modelling, temporal continuity, counterfactual sensitivity, and a self-maintaining boundary — plus, as graded extensions, metacognitive access and non-trivial stakes.

Test: Evaluate base LLMs, memory-enabled assistants, agentic systems, embodied robots, recurrent architectures, and active-inference systems.

H6 — Physics pressure

If spacetime is not fundamental, perspective may be a better explanatory frame than ordinary emergence in some contexts.

Test: Compare spacetime emergence, spacetime functionalism, relational quantum mechanics, causal set theory, loop quantum gravity, entropic gravity, and holographic approaches.

Risk: These comparisons may establish no explanatory advantage for a perspective-based account. This concerns H6; P1’s claim about experienced locality and duration is a separate matter.

Strong weakening conditions #

Each part needs conditions under which it would lose support.

The empirical arm, live #

Three programmes have moved beyond outline concepts; the third, the meta-problem programme, ran one pre-registered test in September 2026 and is recorded below. The Swarm Instrument tests whether adversarial groups of language models reach accurate answers or merely spread convincing accounts. Its preregistered program (Studies 1–3) includes the Study 1 pilot, where a homogeneous debate swarm matched a single agent’s accuracy at matched compute while expressing more confidence; the Study 2 independence × contact comparison; and Study 3 on contagion. The results ledger records every committed prediction, including the seven that came back null, with open per-item data. Coherence Without Contact proposes studies of human readers; its registration is pending. Those programmes use their own numbering. The wider agenda below uses letters to keep the studies distinct.

These studies can bear on the framework’s methodological claims: that a coherent account need not be well supported, and agreement need not be true. Their results provide no support for its metaphysical claims. Building an instrument around the thesis does not make the instrument evidence for the thesis.

The meta-problem programme (September 2026): a pre-registered test of one prediction from the diagnosis, and its null #

What was tried. In September 2026 the diagnosis was put through a five-stage programme on a research branch: state the hypothesis that the problem judgements (the explanatory gap, the zombie, the further ingredient, Mary, acquaintance, intrinsicness, the hard/easy divide) are what a container self-model produces through four generic mechanisms; build matched structural models of three self-model organisations under one rule module; derive each judgement without using an experiential term; subject every document to adversarial model review with the findings kept; and run one pre-registered empirical test. Nothing in the programme was permitted to attach to a claim node without a named transfer rule, and nothing did.

What it found, in order. First, the four mechanisms do not derive all seven judgements. Four of them are one fact (an entry the location query cannot resolve), intrinsicness collapses onto the further-ingredient judgement, and the three judgements specific to the self (the zombie's conceivability form, acquaintance, the hard/easy partition) rest on transparency, which the models stipulate rather than derive. Second, a narrowing of the fold to an indexical anchor (now, this one, actual) predicted the same family of judgements across time, actuality and the self; the literature check found that three of the seven do not travel. Third, no discriminator between the fold and illusionism survived the comparisons the programme examined, at the level of explaining the judgements: both say the judgement-carrying element misrepresents; illusionism grants the anchor as a real functional state, and this proposal's additional commitment is to it as the fold's inward orientation, which is the hard problem proper. Fourth, the one executed test, a pre-registered framing study on three language models (a bank of 56 matched item quads, of which 40 form the registered primary; 5,712 attempted call records per run and 11,424 across two runs, failures included), did not detect the predicted positive effect. Run 1 yielded no pooled primary, because one model rejected a request parameter on every call. Run 2 supplied the confirmatory result: an anchor effect of −0.006 with a 95% interval of [−0.039, +0.027], at a precision three times finer than the design planned for. Under its own filing a null favours nothing and is no equivalence claim. Two limits sit beside it: one model returned empty output on 1,096 of its 1,904 calls, so its own stratum and two secondaries were withheld, and the primary's parseable share was 0.805 against a floor of 0.80.

What it changed. No claim node, tier or confidence moves. The reply to objection 12 is qualified: the prediction it attributed to illusionism is withdrawn as unsupported by illusionist texts, and the matching sentence in the academic edition is corrected. The reception ledger gains four commissioned autonomous-agent critique rounds. The programme's own recommendation is recorded with its result: the language-model line is stopped, no fifth mechanism is added to rescue the model, and the next tests worth running are human, the hardness-asymmetry study the programme registered and, on this site's agenda, Study 5.

The record. The seed with its amendments, the models' descriptions (their Python source stays in the repository's research record), all four critique rounds, the filed pre-registration, both runs' analyses and the status documents are published under downloads/meta-problem/, with a machine layer at meta-problem.json under the same non-transfer clause as the container programme. The multi-agent adversarial review protocol, run with a single model family (a disclosed limitation), identified the recorded defects before any human reviewer saw the programme; it is published separately as a reusable artefact.

Where the method is exercised outside the paper. The same discipline — state the type of thing being made, state the standard it is held to and the cost before starting, have it read from outside the drafting lineage, and end by saying what would count against it — is applied to shorter work in the Lab: something made, something tested, something written down and dated. Nothing in the lab changes a claim on the site; anything there that amounts to an objection goes through the objections ledger like any other reader’s.

Study concepts #

Study A — Comparative phenomenology corpus

Build a database of first-person reports from Zen, Advaita, Dzogchen, Christian apophatic mysticism, Sufism, contemporary nondual teachers, and modern meditation studies. Goal: see whether two orientations appear without forcing them.

Study B — Meditation route comparison

Compare practices aimed at open awareness, self-inquiry, objectless absorption, analytic deconstruction, and devotional surrender. Measure report structure, neural dynamics, and self/world boundary markers.

Study C — AI aperture audit

Create a scoring rubric across eight aperture dimensions and apply it to current LLMs, memory-enabled agents, embodied robots, and future recurrent systems.

Study D — Extraction versus emergence design experiment

Compare products designed to maximize engagement with products designed to increase user self-transparency. Measure agency, reflective clarity, compulsive use, preference stability, wellbeing, and post-interaction self-understanding.

Study 5 — The dissociation study (a meta-problem probe)

Chalmers’ meta-problem asks why people judge consciousness to be a hard problem. Space Immanence proposes that misapplied container imagery generates those judgments. Study 5 is the first proposed test of this diagnosis, rather than of the fold, and distinguishes three outcomes.

Design: Measure two intuition families in long-term practitioners and matched non-practitioners: location and production intuitions (“consciousness is in the head,” “the brain generates experience”) and phenomenal-realism intuitions (“there is something it is like to see red”). This proposal predicts a dissociation: location and production intuitions weaken with practice depth while phenomenal-realism intuitions hold steady. The null, in which neither moves, counts against this proposal's diagnosis. Whether any outcome counts against a particular illusionist account depends on that account committing to a prediction before the data; the joint-weakening prediction an earlier version of this paragraph attributed to illusionism has been given no textual basis (see objection 12, September 2026).

Controls, required before pre-registration: the doctrinal-conformity confound. Practitioners in no-self traditions may answer no-self-flavoured probes from teaching rather than insight. Therefore: sample across traditions with different doctrinal loads (Zen/vipassanā, concentration-heavy, secular MBSR); use indirect scenario-judgment probes in the style of the zombie-intuitions literature (Fischer and Sytsma 2021) rather than doctrine-adjacent agree/disagree items; and pre-register the dissociation as tracking practice-depth measures, not doctrinal exposure. If the signature follows insight rather than ideology, it is evidence.

Status: design stage. Nothing about this study is pre-registered yet, and no result should be read into its description. The September 2026 meta-problem programme bears on this study in one way: its claim to discriminate the diagnosis from illusionism now rests on an illusionist committing to a prediction in advance (see objection 12's qualification), and no discriminator between the two survived the comparisons the programme examined, at the level of the judgements.