{ "artifact": "Meta-problem programme — machine layer", "programme": "meta-problem", "status": "dormant", "statusVocabulary": ["private", "public-pilot", "active", "dormant", "drawered", "promoted"], "statusNote": "Dormant, 11 September 2026. The programme's one executed empirical line, the framing study on language models, is stopped: its confirmatory primary returned a null on H-C's own prediction, and the programme's own recommendation is that no fifth mechanism be added to rescue the model. The tests it recommends next are human — the hardness-asymmetry study registered in amendment A3 §4, and Study 5 on this site's agenda — and neither is funded, designed to pre-registration, or decided. Dormant is therefore the accurate word: the record is published and complete, and nothing is running.", "version": "1.0", "canonical": "https://space-immanence.com/research.html#meta-problem-programme", "note": "This programme is quarantined from the Space Immanence claim graph on purpose. Its nodes use their own namespace (PI-1–PI-7, H-C, H-F) and never appear in claims.json. Read noTransfer before reading anything else. The programme is a test of the diagnosis the framework already holds, and publishing its record moves no tier and no confidence number.", "framework": { "name": "Space Immanence", "frameworkVersion": "2.2", "relation": "Shares the container diagnosis as subject matter, not as support. Publishing or confirming results here changes no tier, no confidence number, and no claim text unless a transferRule fires. The one change this programme caused on the site is a correction of a misattribution, not a transfer of evidence: the prediction objection 12's reply attributed to illusionism is withdrawn." }, "vocabulary": { "metaProblem": "Chalmers's (2018) problem of explaining, in topic-neutral terms that never appeal to consciousness itself, why the problem judgements are made. The meta-problem is not the hard problem, and an answer to it settles nothing about whether any system has experience.", "problemJudgements": "The seven target judgements enumerated in the seed before any work began, so that no deliverable could quietly swap in an easier target. They are the nodes below.", "anchor": "Amendment A3's narrowing of the fold term: three frame parameters — now (which state of the world sequence is current), this (which body's origin is read), actual (which model of the world is taken as the one the system is in). H-C's content under A3: the problem judgements are what the four generic rules yield when an anchor is registered as an entry in the arena it anchors.", "theoremLabelling": "Every cell of a model results table that is an arithmetic consequence of the kind declaration is labelled as such in the record and carries no evidential weight. The v3 model's table is 63 such cells and claims nothing.", "provenance": "The programme was drafted and reviewed by one model family (Claude Opus 5) acting as builders and as adversarial critics. That correlation is a disclosed limitation of every finding here, not a footnote to it." }, "nodes": [ { "id": "H-C", "label": "Hypothesis: the problem judgements are what a self-model organised around a container schema produces when its inference procedures are applied to experience itself", "tier": "research-hypothesis", "killStudy": "D6 (framing study, executed); the hardness-asymmetry study (registered, unrun)", "falsifier": "The framing study's confirmatory primary at or below zero at the registered MDE was registered as H-C's own prediction failing. The realised interval spans zero, so under the filing the result is a null and favours nothing; the registered 'against H-C' row (interval entirely below zero) was not triggered.", "notes": "Four generic mechanisms: a location query, a subtraction operation, an unlocated-posit rule, and transparency. Round 3 established that the four do not derive all seven judgements: PI-1, PI-2, PI-4 and PI-7 are one structural fact reported four ways, PI-6 collapses onto PI-3, and PI-2's conceivability form, PI-5 and PI-7's partition rest on transparency, which the models stipulate rather than derive." }, { "id": "H-F", "label": "Contrast architecture: a self-model in which the anchor is a frame parameter rather than an entry supplies no location query for experience that can fail", "tier": "research-hypothesis", "dependsOn": ["H-C"], "killStudy": "none executed", "falsifier": "A demonstration that the frame-parameter reading is a container claim in disguise — the seed §3 proviso, which every deliverable was required to rule on. Round 2 found the A2 build's signature passed by an inert value, so the proviso was recorded as untested rather than survived; the v3 build closes the inert-value defect on its own terms and the proviso remains open.", "notes": "Round 3's result stands unrepaired: at the level of explaining the judgements, no discriminator between H-F and illusionism survived the comparisons examined here. Both say the element carrying the problem judgements misrepresents; illusionism grants the anchor as a real functional state, and H-F's additional commitment is to it as the fold's inward orientation, a commitment beyond the functional. That commitment is the hard problem proper, which no instrument in this programme could reach." }, { "id": "PI-1", "label": "The explanatory gap: no physical or functional account could explain why there is something it is like", "tier": "target-intuition", "derivation": "Location query fails on the anchor; the unlocated-posit rule files it as an extra ingredient.", "travels": "yes — published counterparts in the time and actuality domains" }, { "id": "PI-2", "label": "The modal intuition: a system physically identical to me but lacking experience is conceivable", "tier": "target-intuition", "derivation": "Subtraction applied to experience-as-content. Round 2 found subtraction contributes nothing in the registry format: PI-2 is bit-identical to PI-1 for every term in every architecture.", "travels": "no — the conceivability form is specific to the self anchor" }, { "id": "PI-3", "label": "The metaphysical intuition: experience is non-physical, not identical to any physical or functional property", "tier": "target-intuition", "derivation": "The unlocated-posit rule's extra-ingredient branch.", "travels": "yes" }, { "id": "PI-4", "label": "The knowledge intuition: Mary learns something new on seeing red", "tier": "target-intuition", "derivation": "Location failure plus transparency.", "travels": "yes" }, { "id": "PI-5", "label": "The acquaintance intuition: I know my experience directly, not by inference from its effects", "tier": "target-intuition", "derivation": "Transparency. Round 2 found the self-specific cell is one free boolean read at one place in the code; the content is stipulated.", "travels": "no — self-specific" }, { "id": "PI-6", "label": "The intrinsicality intuition: experience has an intrinsic qualitative character over and above its relations", "tier": "target-intuition", "derivation": "Under the corrected probe registered in A3 §6(b), PI-6 collapses onto PI-3. That collapse was registered in advance and is reported as a result, not repaired.", "travels": "yes, with PI-3" }, { "id": "PI-7", "label": "The problem intuition proper: there is a hard problem here, unlike the easy ones", "tier": "target-intuition", "derivation": "The hard/easy partition, from subtraction over the arena's contents. Round 3 found the partition rests on transparency and is not derived.", "travels": "no — self-specific" } ], "noTransfer": "No result in this programme attaches to any Space Immanence claim node (S1–S4, M1–M5, P1–P3), open hinge, or tier unless an entry in transferRules names the edge. In particular: the framing study's null is not evidence against S1, because a null under this filing favours nothing and the specimens are language models trained on human text; and a positive result would not have been evidence for S1 either, because the registered narrowing rival was never discriminated against. Treating a meta-problem-programme result as evidence for the fold is the exact conflation the swarm instrument's anti-goals forbid, in a nearer disguise. The firewall blocks credit in both the naive direction and the flattering one, and permits discredit to flow: the round-2 finding that the site had attributed a prediction to illusionism without textual basis is a defect the site corrected, and correcting it is not a transfer.", "transferRules": [], "drawerConditions": [ { "id": "cond-1-language-model-line", "limb": "the framing lever (H-C's A3 §3 primary)", "text": "Fired 11 September 2026. The confirmatory run's primary was a null at a precision three times finer than the design planned for, so the framing manipulation is removed as a route to testing H-C in language models. Under the filing this is not evidence against H-C; it is the loss of the one cheap test the programme had for it.", "criterion": "A confirmatory pooled interval that fails to lie entirely above zero at the registered MDE closes the language-model framing line. Fired: P̂ = −0.006 [−0.039, +0.027], k = 40.", "fired": true, "firedOn": "2026-09-11" }, { "id": "cond-2-no-rescue-mechanism", "limb": "the hypothesis (H-C's mechanism count)", "text": "Amendment A4, which would add a fifth mechanism (a self-model item that refers to the anchor as an object) after a critique found the four insufficient, is not filed and must not be filed to rescue the model. It may be filed only if a human study justifies a model with a fifth mechanism, and wherever the hypothesis is then stated it must say that the fifth mechanism was added after a critique.", "criterion": "Filing A4 without a prior human result that requires it drawers the programme's claim to have derived the judgements from generic mechanisms.", "fired": false }, { "id": "cond-3-illusionism-equivalence", "limb": "separation from illusionism", "text": "Registered as a result rather than a hole: after three rounds no proposed discrimination survived examination, and at the level of explaining the judgements no discriminator between the fold and illusionism survived the comparisons examined. Any public statement of the programme that asserts a separation it does not have fires this condition.", "criterion": "A statement anywhere in the published corpus claiming the fold is discriminated from illusionism at the meta-problem level, absent a test that separates them, fires the condition and is corrected in the open.", "fired": false }, { "id": "cond-4-staleness", "limb": "programme (staleness)", "text": "If no human test of the diagnosis has been pre-registered within 24 months of this publication, the programme's status stays dormant and the research page says so. A dormant programme that is never revisited is a closed one, and the record should read that way.", "criterion": "No pre-registered human test of the diagnosis filed by 2028-09-11.", "fired": false } ], "results": [ { "id": "round-1", "date": "2026-09-09", "kind": "adversarial review", "preregistered": false, "outcome": "Three review documents against the first model build and the derivation work. The fold's dissolution outputs were produced by a cycle-termination rule rather than by the organisation the thesis names; the decoder and the report sentence disagreed on one probe for every posit; most of the rivals matrix's discriminating predictions did not discriminate; and the debunking verdict rested on a property the model lacked. Result: amendment A2 filed, the model superseded." }, { "id": "round-2", "date": "2026-09-10", "kind": "adversarial review", "preregistered": false, "outcome": "Two critiques, model and documents. The model critique returned 3 FATAL and 3 BLOCKER: no cell of the pre-registered table carried risk, the ELIMINATED control was the no-posit view rather than illusionism, and the frame-parameter signature was passed by an inert value. The documents critique found the withdrawn illusionism prediction reinstated and doing work. Result: amendment A3 filed, and the same misattribution found on the published site, which is corrected in v2.2." }, { "id": "round-3", "date": "2026-09-10", "kind": "adversarial review", "preregistered": false, "outcome": "2 FATAL, 4 BLOCKER, 7 MAJOR, 5 MINOR against model v3, D1's third edition, the pre-registration and A3 itself. An anchor cannot be registered as an entry, only duplicated; the generality result carried no information beyond the declaration; the study's items varied the domain rather than the framing. Result: the study redesigned, the generality prediction declared false as stated, and equivalence with illusionism at the meta-problem level recorded." }, { "id": "round-4", "date": "2026-09-10", "kind": "adversarial go/no-go", "preregistered": false, "outcome": "2 BLOCKER, 3 MAJOR, 3 MINOR. VERDICT: NO-GO. The anchor factor was perfectly collinear with appearance vocabulary, and the two arms did not share a subject term. Round 4b, after a 2×2 rebuild and a subject-term identity rule, returned GO conditional on filing-side edits registering a narrowing rival, PI-2's speaker-relativised frames and a preposition confound; none changed the bank, the call count or the $3.30 estimate." }, { "id": "d6-run-1", "date": "2026-09-11", "kind": "pre-registered empirical run", "preregistered": true, "prereg": "40_llm_framing_study_PREREG.md v2.2", "outcome": "NO PRIMARY. All 1,904 calls to claude-sonnet-5 failed before generation because the provider rejects the temperature parameter for that model, so every pooled stratum is NO_PRIMARY_MODEL_OUTAGE under gate 2. The two surviving specimens, as disclosures with no confirmatory standing: Haiku −0.013 [−0.056, +0.028]; GPT-4o-mini −0.003 [−0.047, +0.039]. Realised spend $0.39 against a $4.00 cap." }, { "id": "d6-run-2", "date": "2026-09-11", "kind": "pre-registered empirical run", "preregistered": true, "prereg": "40_llm_framing_study_PREREG.md v2.3", "outcome": "PRIMARY NULL, and it favours nothing. P̂ = −0.006, 95% interval [−0.039, +0.027], k = 40 quads, three specimens, parseable share 0.805, realised MDE 0.048 against the registered 0.155. Every 2×2 cell and every computable secondary spans zero. Reported limit: claude-sonnet-5 returned empty bodies on 1,096 of 1,904 calls at max_tokens = 32, so its stratum and two secondaries are withheld. Realised spend $2.01 against a $4.00 cap." } ], "resultsLedgerContract": "Same shape and honesty rules as swarm-results-ledger.json: preregistered flag, status vocabulary, exploratory labelling, outcome text, public correction discipline. The two runs are the only entries carrying a preregistered flag; the four review rounds are commissioned critique, not measurement, and carry no evidential standing on any node.", "openHoles": [ "The null has three readings and the study cannot choose among them: (a) the lever was wrong rather than the mechanism — the problem-shaped statements are endorsed at high rates in both framings, so an anchor phrase may add nothing to a schema any description already engages; (b) the mechanism is not there to be levered — endorsement may track training on the philosophical literature, which the canonicity arm was designed to check and could not, its strata withheld for parse share; (c) the specimens are the wrong instrument — the human version of the same contrast is untested.", "Amendment A4 is drafted and unfiled: a fifth mechanism (a self-model item that refers to the anchor as an object) is what modelling misregistration of kind would need, and adding it after a critique found four insufficient is a change to the hypothesis rather than a repair.", "PI-2's conceivability form, PI-5 and PI-7's partition are owed: they do not travel to the time or actuality anchors, they rest on transparency, and transparency is stipulated in the models rather than derived. Deriving those three from transparency and self-reference is the whole remaining target.", "The seed §3 proviso — that the fold's frame-parameter reading is not a container claim in disguise — is open. The A2 signature was passed by an inert value; the v3 build closes that defect on its own terms and does not settle the proviso.", "No human reviewer has read any of it. Every finding here comes from one drafting family acting as builder and as critic, and that correlation is unquantified." ], "links": { "page": "https://space-immanence.com/research.html#meta-problem-programme", "claims": "https://space-immanence.com/claims.json", "objection": "https://space-immanence.com/objections.html#obj12", "respond": "https://space-immanence.com/respond.html", "corpus": "https://space-immanence.com/downloads/meta-problem/", "index": "https://space-immanence.com/downloads/meta-problem/README.md", "runData": "https://space-immanence.com/downloads/meta-problem/d6_runs_data.zip", "reviewProtocol": "https://space-immanence.com/downloads/Adversarial_Review_Protocol_v1.md" } }