Iteration-3 Synthesis: Sharper Obstructions, Same Partial Coherence

Status: Active synthesis — coherence verdict UNCHANGED at PARTIAL Last updated: 2026-06-08 Iteration: 3

This note records the iteration-3 investigation and folds in the synthesis-verdict review (one substantive technical correction was forced; the conclusion is unchanged). The honest one-line headline: the program got more rigorously understood and more honestly hedged, the obstructions got sharper, and the central bet got no closer to being either confirmed or empirically decidable. INFERENCE, high confidence

---

Updated coherence verdict (still-partial)

The coherence verdict is PARTIAL, unchanged from iteration 2, and the iteration-3 work confirms this is the correct verdict rather than a placeholder. INFERENCE, high

Iteration 3 ran four moves: (1) a de Sitter entanglement first-law attempt (OP-41); (2) a formal encode-vs-generate criterion (OP-46); (3) a Program-A $+$ Program-B merge attempt; (4) a red-team pass over the synthesis page. Three of the four sharpened obstructions rather than removing them; the fourth forced the synthesis page to retract over-claims back to what its own source notes already supported.

What survived every attack — the project's best work — and was strengthened, not weakened:

• The deflationary core [ESTABLISHED-as-argued]: the catalogued GR/QM "contradictions" are domain-mismatches or unsolved-but-consistent problems, not logical inconsistencies (referee verdicts in ../GAPS_AND_CONTRADICTIONS.md are sound).
• The central diagnosis INFERENCE, high: the algebraic/holographic program ENCODES geometry, it does not GENERATE it — strengthened by the verified Giddings results (arXiv:2505.22708, 2510.24833 [VERIFIED]: dressed operators generically fail to commute spacelike at leading order in $G_N$).
• The Born form/meaning split, linearity-as-a-derivative-of-no-signaling, the refusal of an "it's all information" fusion, the dS negative meta-result, and BMV-as-class-falsifier-not-quantization-proof.

The wager — algebra/entanglement before geometry is true of our $\Lambda>0$ universe — remains SPECULATIVE and, by the wiki's own anti-crank standard, still not yet physics: it has no distinguishing near-term test (see Empirical falsifiability portfolio).

---

Did Program A move?

SHARPENED, NOT ADVANCED — and that distinction is the honest headline. INFERENCE, high

Program A (geometry-from-algebra) did not graduate from "encoding" to "generating" geometry at $\Lambda>0$. The dS-first-law attempt (OP-41, see the attempt note) produced a precisely-located obstruction, not a derivation.

(1) Jacobson 2015 re-read: a real but different result

Jacobson 2015 (arXiv:1505.04753, PRL 116 201101) [VERIFIED] is correctly re-read as already a local, dS-valid derivation of the full Einstein field equation from entanglement equilibrium.

Synthesis-verdict correction (the one substantive technical fix). Iteration-3's draft claimed Jacobson is "the NONLINEAR semiclassical Einstein equation ... in one respect (nonlinearity) STRONGER than the AdS FGHMVR linearized result." This conflates two senses of "nonlinear" and is [CONTESTED-as-stated]. The corrected characterization:

Jacobson derives, from a first-order entanglement-equilibrium (stationarity at fixed volume) condition imposed pointwise in small geodesic balls, the full Einstein equation $G_{\mu\nu} + \Lambda\, g_{\mu\nu} = 8\pi G\, \langle T_{\mu\nu}\rangle$ as a local tensor equation, including $\Lambda$, for conformal matter (nonconformal matter requires the unproven $\delta X$ / entanglement-entropy-variation conjecture). The derivation is first-order in the state variation — exactly like FGHMVR. The genuine contrast is not nonlinear-vs-linearized but: Jacobson reconstructs a complete local tensor equation pointwise because Einstein gravity is algebraic in curvature (no derivatives of curvature), so a small-ball first-order condition at every point reconstructs the full tensor equation. Bueno–Min–Speranza–Visser (arXiv:1612.04374) show the same recipe yields only the linearized equations for higher-curvature gravity — isolating that the "full equation" is a property of Einstein gravity's local curvature-algebraic structure, not of the derivation's perturbative order. FGHMVR instead reconstructs the bulk equation explicitly linearized about pure AdS. ESTABLISHED, corrected

Crucially, Jacobson presupposes geometry: the area–entropy density $\eta = 1/(4G\hbar)$ is [ASSUMED, not derived]; the maximal (not merely stationary) vacuum-entanglement hypothesis at fixed volume is required (fixed-radius spoils the coefficient by $(d+1)/3$); and the matter is conformal. It is non-holographic / local-thermodynamic, answering a different question than the holographic/algebraic OP-41. This is a reconciliation, not closure — iteration 2 already conceded Jacobson "presupposes geometry." ESTABLISHED

(2) The CLPW Type II$_1$ route breaks — and we now know exactly where

The CLPW static-patch route (arXiv:2206.10780 [VERIFIED]: Type II$_1$, empty dS $=$ maximum-entropy state) was pushed toward linearized Einstein and broke at a newly-identified structural step, not merely "stalled." The break is at the conversion-to-tensor step (Step 5 of the attempt), and it breaks for a structurally different reason than in AdS:

• (i) One Killing vector, not a family of anchoring regions [decisive]. AdS FGHMVR obtains a local equation $\delta G_{\mu\nu}(x) = 8\pi G\,\delta T_{\mu\nu}(x)$ because the timelike boundary supplies a continuum of ball regions (all positions, sizes, boosts) $\Rightarrow$ a continuum of independent first-law constraints inverting to a pointwise field equation. The dS static patch supplies exactly one distinguished region and one Killing vector $\xi$ (the static-patch boost). So $\delta S_{\rm gen} = \delta\langle K\rangle$ yields only the single scalar constraint $\int_\Sigma \xi\cdot\delta C = 0$ — the $\xi$-contracted, slice-integrated linearized Hamiltonian constraint $=$ horizon-area stationarity $=$ the dS GSL — not the tensor $\delta G_{\mu\nu}$. There is no boundary to slide the region along.
• (ii) No timelike conformal boundary $\Rightarrow$ no boundary stress tensor, and the "surface" is the cosmological horizon (a bifurcate Killing horizon), fixed by background isometry, not an extremized RT min-cut. There is no min-cut/extremization to linearize. ESTABLISHED
• (iii) The variational principle is sign-REVERSED. Banihashemi–Jacobson–Svesko–Visser (arXiv:2208.11706, JHEP 01(2023)054 [VERIFIED]) establish $dE_\xi = -T_{\rm dS}\,dS_{\rm hor}$: adding Killing energy decreases horizon area/entropy. With empty dS maximizing $S_{\rm gen}$, the dS variational principle is a concave maximization ($\delta^2 S_{\rm gen} < 0$), structurally opposite to AdS RT minimization (Hollands–Wald positive-canonical-energy). The AdS machinery cannot be transcribed. INFERENCE, high — rests on established facts
• (iv) The modular Hamiltonian is geometric only for Bunch–Davies. $K_{\rm BD} = 2\pi H_\xi$ holds for Bunch–Davies; for a perturbed state, modular flow is non-geometric (no Bisognano–Wichmann), so $\delta\langle K\rangle$ is not a clean boost energy. Geometry is presupposed, not generated (the iteration-2 algebraic-background-independence verdict, now in explicit dS dress; see ../notes/2026-06-08-algebraic-background-independence.md). INFERENCE, high

Net on Program A

Genuine technical progress of the negative kind. Iteration 2 said "no CLPW-rooted dS first-law $\Rightarrow$ Einstein chain exists." Iteration 3 says why (one region / one Killing vector vs AdS's continuum; sign-reversed maximization) and identifies one concrete, citable escape: a family of static-patch modular Hamiltonians over the $SO(d+1,1)$ observer orbit (Chen–Xu covariant observers, arXiv:2511.00622 [VERIFIED], which builds exactly this $L^2(SO(1,d))$ QRF family) or sub-patch CHM causal diamonds (concretely, the dS-diamond modular Hamiltonian of Fröb, arXiv:2308.14797, JHEP 12(2023)074), to manufacture the missing continuum of constraints.

Tagging correction (forced by the verdict review). The draft tagged the obstruction "near-no-go (HYP-dS-NOMIN)." This overstates. No theorem forbids manufacturing the missing continuum; and Fröb cautions that the dS-diamond modular Hamiltonian is geometric only in the full-static-patch limit and is non-local for finite diamonds. The route is non-trivial but UN-foreclosed. Correct tag: a sharpened-but-OPEN structural obstruction with one identified, unexecuted escape route — OPEN, not [near-no-go]. Whether the $SO(d+1,1)$ family yields a pointwise $\delta G_{\mu\nu} = 8\pi G\,\delta T_{\mu\nu}$ is untested; calling it "the missing continuum of constraints" should be hedged to "a candidate mechanism whose ability to produce a pointwise field equation is untested." OPEN

Verdict: Program A advanced in understanding but did not advance toward geometry-generation.

This is reinforced by the encode-vs-generate criterion (OP-46, see the criterion note): the encode/generate frontier coincides exactly with the presence of a hand-chosen tensor-factorization $\mathcal H = \bigotimes_i \mathcal H_i$ or graph $G=(V,E)$ — which is the Type III$_1$ no-factorization fact in disguise. No surveyed construction generates Lorentzian spacetime from factorization-free, region-free, symmetric-state-free data. Cao–Carroll(–Michalakis) "space from Hilbert space" comes closest, but outputs only a Riemannian spatial metric and only after positing a factorization (a surrogate for input-geometry $g_5$). INFERENCE, high on the classification; medium on the criterion being uniquely right

---

The A+B merge attempt

SUBSTANTIVE PARTIAL MERGE at one sublayer; thin methodological bridge at the deeper sublayer; NOT empty re-description, NOT a full merge. A real but bounded upgrade to iteration-2's "thin bridge." INFERENCE, medium (See the merge note.)

The shared object is a von Neumann algebra with faithful normal state $(\mathcal M, \omega)$ under the no-signaling / normal-steering (GHJW) constraint.

What merged — the Born-form $+$ thermal/entropic sublayer

On $(\mathcal M, \omega)$ one can non-circularly co-derive:

• (a) Pillar B's Born-rule FORM $+$ exact linearity, lifted to infinite-dim normal state spaces by the 2026 causal-rigidity theorem (arXiv:2602.09056 [VERIFIED], published 7 Feb 2026: recovers Born on vN-algebra normal state spaces via GNS, under no-signaling $+$ normal steering $+$ $\sigma$-affinity); and
• (b) Pillar A's KMS/thermal-time (Tomita–Takesaki) $+$ emergent modular/causal flow $+$ (via an internal crossed-product clock) a Type II trace, finite generalized entropy, GSL.

Neither pillar presupposes the other on this sublayer, and the "observer" is demoted to an internal gauge-invariant quantum reference frame (CLPW $+$ covariant-observer 2511.00622 [VERIFIED]), genuinely dissolving the agent/no-agent category clash here.

What did NOT merge — the Hilbert-space-FRAMEWORK sublayer

The GPT axioms that single out Hilbert space / the complex field / tensor-product composition have no Type III formulation and provably fail in infinite dimensions. Three located breaks (candidate-no-go OP-GPT-TYPE3):

• BREAK 1 — composition not even statable. Local tomography presupposes $\mathcal H_{AB} = \mathcal H_A \otimes \mathcal H_B$; Type III$_1$ has no tensor factorization and no density matrices (Connes–Haag–Araki) ESTABLISHED. The 2026 theorem is honest: it lifts only the Born-rule form, not composition.
• BREAK 2 — load-bearing axioms false in $\infty$-dim. Hardy continuous-reversibility and CDP purification fail (the shift operator is an isometry, not unitary; orthogonal projections are not sequentially compact) ESTABLISHED.
• BREAK 3 — the crossed-product trace re-imports geometry. The III $\to$ II crossed-product trace exists only with a clock Hamiltonian bounded below in a fixed-background semiclassical perturbative window — re-importing the preferred-state/fixed-geometry that Pillar A should generate (Bisognano–Wichmann/CHM circularity), and may fail nonperturbatively (Giddings 2505.22708/2510.24833 [VERIFIED]). INFERENCE, high

Net on the merge

YES for a single non-circular substrate on the Born-form $+$ thermal/entropic sublayer; NO (with the precisely-located candidate-no-go OP-GPT-TYPE3) for the Hilbert-space-framework sublayer. The shared no-signaling principle does real non-tautological work on both sides (so not empty re-description), but composition/factorization cannot be posed on Type III (so not a full merge). The merge "succeeded narrowly, failed deeply" — and crucially it did not merge in the cosmological closed-universe / no-external-observer regime that was the prompt's crux, because that is exactly where the BREAK 3 perturbative window closes. INFERENCE, medium

---

Red-team impact (what we weakened)

The red-team found 0 fatal errors, 4 serious cracks, 2 cosmetic; all load-bearing 2025–2026 citations re-verified exactly. The cracks are the synthesis page over-compressing better-hedged source notes. Four forced weakenings: INFERENCE/CONTESTED, high that the weakenings are warranted

1. "Fixed cores" re-tagged from flat to HETEROGENEOUS. The five "fixed cores" are not the same kind of object, and three are not actually fixed:
   • (i) Type III$_1$ "for any $\Lambda$" is ESTABLISHED for QFT on a fixed background but CONTESTED/OPEN as a fixed core of the gravitational theory — Giddings shows gravitational dressing alters the algebra at leading order in $G_N$. The synthesis page listing it as ESTABLISHED-fixed directly contradicts the wiki's own algebraic-background-independence note (which tags it OPEN/CONTESTED, and Giddings backs the note). This is a genuine internal inconsistency that must be resolved by propagating the note's qualification up.
   • (ii) Area law is NOT a universal input — derived in AdS (RT), output in Sakharov; load-bearing only for the Jacobson–Padmanabhan–Verlinde lineage specifically.
   • (iii) No-signaling is genuinely shared but NOT quantum-forcing (PR-boxes are no-signaling yet non-quantum) — demote from "core that forces structure" to "one of several GPT axioms."
   • (iv) Born form is forced only conditional on non-contextuality $+$ effect-additivity.
   • (v) Unitarity is conditional on standard QM being final and is deliberately violated by live GRW/CSL.
   • Honest header: "a small set of robust-but-non-homogeneous anchors, each conditional on a stated background or premise."
2. "Demotion-and-derivation" demoted from "falsifiable program" to "research HEURISTIC." As a meta-strategy it forbids nothing and is unfalsifiable by scoping; its only falsifiable content lives in the object-level deliverables (OP-dS-FIRSTLAW, OP-CP1, BMV $m^2$-vs-$m^3$, collapse bounds) and in the bracketed SPECULATIVE wager it quarantines. "Falsifiable" was a category error applied to a procedure.
3. "Independent convergence of six lenses" weakened to "convergence of methods sharing load-bearing inputs" (a streetlight artifact). H0/H4 share Killing-symmetric vacua $+$ AdS/large-$N$; H1/H3/H6 share the continuum Malament/HKM theorem $+$ the unproven Hauptvermutung; H7 is AdS/large-$N$. Agreement of methods sharing their assumptions is weaker evidence for the wager than genuine independent convergence.
4. CLPW Type II$_1$ kept as a $\Lambda>0$ entropy foothold but stripped of "$\Lambda$-DIAGNOSTIC" billing. Trace normalizability (II$_1$ vs II$_\infty$) is set by the horizon/observer-Hamiltonian spectrum, not by $\Lambda$: Schwarzschild–de Sitter ($\Lambda>0$) is II$_\infty \times$ II$_\infty$ (KLS, PRD 111 025013 [VERIFIED]). So II$_1$ is a feature of the static patch's bounded area, not of $\Lambda>0$ as such — a persistent emphasis error.

What survived every attack (the wiki's best work): the deflationary core; the "encodes-not-generates geometry" verdict (STRENGTHENED by Giddings); the Born form/meaning split; linearity-as-derivative-of-no-signaling; the PARTIAL-coherence verdict; the refusal of the "it's all information" fusion; the dS negative meta-result; BMV-as-class-falsifier-not-quantization-proof.

---

Empirical falsifiability portfolio

Every live channel tests the program's logical GATE (is gravity non-classical?), its correct TARGET (is the universe really dS?), or the status of its FIXED CORES (is linearity/Lorentz exact?). None tests the geometry-from-algebra wager itself. INFERENCE, high

---

Net progress

Iteration 3 produced real but almost entirely NEGATIVE / STRUCTURAL progress, and the coherence verdict is UNCHANGED at PARTIAL — correctly so. INFERENCE, high

Three of the four moves sharpened obstructions; the fourth forced retractions back to the source notes:

1. The dS-first-law attempt did NOT advance Program A toward geometry-generation. It converted iteration-2's bare "no dS first-law chain exists" into a located structural obstruction — one region / one Killing vector vs AdS's continuum, plus a sign-reversed concave maximization — leaving exactly one un-foreclosed escape (an $SO(d+1,1)$ family of modular Hamiltonians, concretely instantiated by Chen–Xu covariant observers and dS-diamond CHM constructions). Tagged OPEN, not "near-no-go."
2. The encode-vs-generate criterion is a genuine deliverable. It makes the central distinction decision-usable (two graded axes $\Delta_{\rm geo}$, $\kappa_\Phi$ $+$ a 5-item smuggling checklist $g_1$–$g_5$), classifies six constructions, and isolates that the encode/generate frontier coincides exactly with a hand-chosen factorization/graph — the Type III$_1$ no-factorization fact in disguise. Net finding: NO surveyed construction generates Lorentzian spacetime from factorization-free, region-free, symmetric-state-free data.
3. The A+B merge succeeded only on the Born-form $+$ thermal sublayer and failed (candidate-no-go OP-GPT-TYPE3) on the Hilbert-space-framework sublayer — notably failing in the cosmological no-observer regime that was the crux.
4. The red-team found no fatal errors but forced four weakenings, including resolving a genuine internal contradiction (Type III$_1$ as "fixed core" once gravity is on).

The deflationary core and the central diagnosis ("encodes, does not generate") survived every attack and were STRENGTHENED by the verified Giddings results. The wager remains SPECULATIVE and, crucially, without any distinguishing near-term test: every live channel tests the gate, the target, or the cores — none tests geometry-from-algebra. By the wiki's anti-crank standard, the wager is therefore still not yet physics.

Honest one-line summary: more rigorously understood, more honestly hedged, obstructions sharper, central bet no closer to confirmation or empirical decidability.

---

Next highest-leverage targets

1. [Re-aimed OP-dS-FIRSTLAW-R2 — single highest-leverage theory target.] Attempt the one un-foreclosed escape: build a FAMILY of static-patch modular Hamiltonians over the $SO(d+1,1)$ observer orbit (Chen–Xu, arXiv:2511.00622, $L^2(SO(1,d))$ QRF) or sub-patch CHM diamonds (Fröb, arXiv:2308.14797), and test whether the family supplies enough independent first-law constraints — each with geometric modular flow in a common state — to invert to a local $\delta G_{\mu\nu}$. If it cannot (constraints not independent, or modular flow geometric only in Bunch–Davies), HYP-dS-NOMIN is promoted from conjecture to a true no-go and the holographic dS route is closed. OPEN
2. [Settle OP-GPT-TYPE3 — converts the partial merge into a full merge or a true no-go.] Decide whether the GPT framework-fixing axioms (local tomography/tensor-composition, purification, continuous reversibility) admit ANY formulation on a Type III$_1$ factor or its normal state space. Concretely: can the 2026 causal-rigidity theorem's $\sigma$-affinity $+$ normal-steering be strengthened to fix composition, not just the probability rule? If composition provably needs factorization (which III$_1$ lacks), this is a no-go for the Hilbert-space-framework half. OPEN
3. [Break the Bisognano–Wichmann/CHM circularity — deepest reorganization-vs-generation test.] Recover a Killing/boost structure and a horizon as the OUTPUT of an $(\text{algebra, state})$ pair whose modular flow is NOT a known symmetric (BW/Hadamard) vacuum. This is the criterion's clause (iii), satisfied by no surveyed construction; succeeding here is the only thing that would turn "encode" into "generate." OPEN
4. [Resolve the nonperturbative fate of the crossed product — OP-CP1.] Determine whether the III $\to$ II crossed-product Type II structure survives beyond all-orders-in-$1/N$, or is replaced by Giddings' non-algebraic structure (arXiv:2510.24833). Bears on BREAK 3 of the merge and on whether "Type III$_1$ for any $\Lambda$" can be a fixed core at all once gravity is on. Either answer is decisive. OPEN/CONTESTED
5. [Propagate the red-team weakenings into the synthesis page — documentation, not research.] Re-tag the "fixed cores" list as heterogeneous with per-item status; demote "falsifiable program" to "research heuristic"; add the shared-input/streetlight caveat to the convergence claim; strip "$\Lambda$-diagnostic" billing from CLPW II$_1$. Required for the wiki to stop contradicting its own better-hedged notes.
6. [Mature the shared enabling technology — highest-leverage EMPIRICAL target.] Push matter-wave / levitated-optomechanics mass scale upward: it simultaneously gates the BMV gate-test of Program A AND tightens objective-collapse bounds on the linearity assumption — the two live channels of Program B — and is the prerequisite both ultimately depend on. No theory target is empirically decidable without it.

---

See also

• 2026-06-08-iter3-de-sitter-first-law-attempt.md — the OP-41 dS entanglement first-law attempt (where and why Step 5 breaks)
• 2026-06-08-iter3-encode-vs-generate-criterion.md — the formal OP-46 encode-vs-generate criterion and verdict table
• 2026-06-08-iter3-merge-attempt-operational-substrate.md — the Program-A $+$ Program-B merge on a shared operational substrate
• 2026-06-08-unified-program-synthesis.md — the iteration-2 synthesis note this one updates
• 2026-06-08-algebraic-background-independence.md — the encode-not-generate / Type III$_1$ background note
• ../FINDINGS.md — consolidated findings register
• ../ROADMAP.md — iteration roadmap and open-problem queue
• ../GAPS_AND_CONTRADICTIONS.md — the deflationary-core referee verdicts
• ../OPEN_PROBLEMS.md — OP-41, OP-46, OP-CP1, OP-GPT-TYPE3
• ../HYPOTHESES.md — H0–H7 and HYP-dS-NOMIN / HYP-GIE-DISCRIM