Testing the comparison across pair macrograins

6 Executed official intrinsic-grain comparison

6.1 Frozen domain and environment

IIT's intrinsic-unit framework explicitly addresses causal grain [14, 15]. Its existence should not be confused with deriving a micro-level substrate from a bare transition table. We restored the archived Python 3.13.15 environment and official PyPhi wheel 2.0.0rc3.dev6+g2b98c1d5b, revision 2b98c1d5bba246c042fd1c65618ebad86196342a. The complete canonical presets, not hand-mixed options, specify the 2023 and 2026 analyses. The intrinsic-units article's publication year does not determine the integration preset used by its examples [19].

The primary contract permits at most two micro constituents per macro unit, one update per unit, one hierarchy level, all nonconstant binary maps modulo output complementation, no background apportionment, and all official disjoint assemblies and admissibility/exclusion rules. There are seven pair maps for each two-constituent footprint and five native units, or 75 raw unit candidates before validity. The exact signed-control transport preserves constituent stakes and map partitions, not merely the same command-line flags. The cost estimate was 2,061 systems and 36,261 partition sweeps per cell as upper bounds; pruning was disabled for the primary comparison.

The pair bound is the smallest spatial extension beyond individual units, while exhaustive pair mappings avoid choosing a particular favorable aggregation rule. It is a bounded stress test, not a claim that larger searches are impossible or unnecessary for an all-grain conclusion. The native CCI-1 outcomes were already known. Local records place the contract lock at 20:26:13 UTC and the first recorded primary completion at 20:26:51 UTC on 30 September 2026. This is a same-author local protocol, not an externally preregistered or blinded test; the timestamps and hashes are provenance records, not independent evidence of preregistration.

All 32 matched original states are evaluated for the original, recoded, and signed-control systems under both presets. Each of the 192 cells runs in its own process. Three local subprocesses ran concurrently; computation was not delegated to another model. Official results are serialized and reloaded with equality checks. A separate fresh-process replay repeats all six state-zero cells. This is same-workflow repeatability, not independent replication.

6.2 Results and zero-valued returned objects

Table 2 gives the positive selected organizations. Official validity retains 19 units per 2023 cell: five native units and fourteen pair-macro candidates. In the 2026 cells, the constituent-positivity gate leaves only the five native units. Thus the two presets do not have equally extensive surviving macro competition. Within this domain, no positive macro selection replaces the inherited native selections: both positive original 2023 complexes remain, the recoded selection remains all five native units, and the 2026 selection remains empty.

Preset	Original	Recoded
IIT 2023	$\{a_1,a_2\}$ and $\{j_0,j_1\}$, each $\varphi_s=2$	All five native units, $\varphi_s=2$
IIT 2026	No positive complexes	No positive complexes

The completed primary run contains 17,800 evaluation records. All 64 matched state–preset comparisons with the signed control agree on the retained unit domain, evaluation records, positive selections, and ties. The 3,136 native evaluation records encountered in the new search match the inherited CCI-1 table. All 192 official serialization round trips pass, and six fresh-process state-zero repeats reproduce the compared scientific fields. There are no unresolved tie groups in this bounded result. Raw output includes 160 zero-valued returned objects across the primary cells, retained separately from the 160 positive complexes.

The macro interface can return zero-valued objects in its complexes collection. The official complex object's truth value tests positive integration. We preserve those objects in raw output and do not count them as positive complexes. This reporting distinction is essential: interpreting the raw collection length as a subject count would contradict the package's own positivity semantics. The adapter does not change the official selection code or numerical values.

The direct comparison checks selected unit partitions, values, valid-unit domains, evaluation records, and ties under the signed control. It does not assert that an arbitrary $E$ maps native subsets to native subsets. Nor does matching positive supports and values alone prove equivalence of all unselected structures. CCI-1's earlier detailed structure records remain separately preserved.

Preserved pilot-summary schema.

The first archived 2023/original/state-zero summary omits the later adapter field official_truth_value on its three returned objects. Their recorded integration values and the official object agree with the later schema; this is a metadata difference, not a numerical correction. The original remains unchanged. A separately labeled derivative adds the three Boolean values, with a machine-readable change log and a comparison to the saved hostile-review replay. No original value, selection, or failed comparison is removed [20].

6.3 Scope of the deterministic gate

The newer preset incorporates the intrinsic-information requirement, and its deterministic-zero consequence is already explained by the authors [16]. Official intrinsic-unit admissibility requires positive integration of the constituent system before a candidate's own mapping or update grain is used [15]. Therefore, for this construction with fixed native backgrounds, every hierarchy whose first new unit must pass that gate on a zero-valued native constituent system fails at its first step. Induction excludes such bottom-up additions while those premises remain unchanged.

This is a source-level gate observation, not a completed unrestricted macro search. It does not establish that every deterministic description at every coarse grain has zero intrinsic information. Coarse-graining, histories, or background treatment can change the relevant probabilistic model. In particular, the unrestricted 2023 temporal and larger-spatial-grain question remains outside the executed domain.