UNNS SUBSTRATE RESEARCH PROGRAM · TURBULENCE BRANCH · JHTDB PILOT-B REPLICATION · 2026-09-07

JHTDB Turbulence — Pilot B Replication

A preregistered, protocol-frozen independent-sample replication of the Pilot-A scale–time routing result on Johns Hopkins Turbulence Database forced-isotropic data (isotropic1024coarse). Pilot B uses a spatially and temporally disjoint 256³ cutout processed through the unmodified STRUC-ROUTE-I v0.1.2 → ROUTE_LADDER_EXTRACT → STRUC-I v1.0.4 / STRUC-PERC-I v2.5.0 → STITCH-MECH v0.1.1 pipeline, scored against the eight preregistered R1–R8 criteria.

SOURCE ACQUIRED · HASHED · ZERO OVERLAP WITH PILOT A R1 PASS — CONSTRAINED_ROUTING R2 PASS — D□ ≪ N0, p=0.0099 R3 FAIL — D_STITCH Structural Boundary (Aκ=0.852) R4 PASS — FULL_PERCOLATION, 336 gaps R5 PASS — survives N1, p=0.0099 R6 PASS — survives N2, p=0.0099 R7 FAIL — tail enrichment 1.0×, p=1.0 R8 PASS — branching localization CLASSIFICATION: STRUCTURAL PARTIAL REPLICATION MECHANISM: SURVIVES_LOCAL_GEOMETRY_CONTROL RECORD STATUS: FINAL_FROZEN_RESULT

Instruments: STRUC-ROUTE-I v0.1.2 · ROUTE_LADDER_EXTRACT · STRUC-I v1.0.4 · STRUC-PERC-I v2.5.0 · STITCH-MECH v0.1.1 Pilot B: 256³ cutout, x/y/z 513:768 (1-based) · frames 501:510 · t=1.000–1.018 Pilot A reference: x/y/z 1:256 · frames 1:10 · t=0.000–0.018 Objects: 10,515 · Relations: 14,169 supplied / 14,118 eligible Status: R1–R8 frozen · protocol locked · no post-hoc threshold changes

§0 PILOT-B PIPELINE — ALL STAGES COMPLETE

SOURCE
ACQUISITION

→

PHYSICAL
ADAPTER

→

ROUTE-I
PRIMARY

→

ROUTE-I
SENSITIVITY [1,2,4,8]

→

ROUTE LADDER
EXTRACTION

→

STRUC-I
v1.0.4

→

STRUC-PERC-I
v2.5.0

→

STITCH-MECH
N1 / N2

→

R1–R8
SYNTHESIS

INTEGRATED PRINCIPLE

The route mechanism reproduces; the full cross-chamber signature does not.
An exceptionally low scale–time stitching defect (D□) reappears in a spatially and temporally disjoint turbulent sample and survives both a distance-controlled (N1) and a distance-plus-feature-controlled (N2) graph-local null. But the scalar STRUC-I admissibility of that same ladder falls short of Geometric Persistence, and the frozen upper-decile physical-intensity localization does not survive a zero-heavy D□ distribution. Route-mechanism reproducibility and full cross-chamber/localization reproducibility are separable questions — Pilot A alone could not distinguish them.

§1 INDEPENDENT PHYSICAL SAMPLE — ZERO OVERLAP WITH PILOT A

2.0GB

SOURCE SIZE (2,015,302,728 B)

SPATIAL GRID-POINT OVERLAP

TIME-FRAME OVERLAP

256³

CUTOUT SHAPE (BOTH PILOTS)

STORED FRAMES (BOTH PILOTS)

0.982

PHYSICAL-TIME GAP FROM PILOT A END

PASS

HDF5 VALIDATION STATUS

JHTDB getCutout QUERIES (Z-SLABS)

PILOT A ↔ PILOT B — SOURCE SELECTION · isotropic1024coarse · same forced-isotropic regime

Property	Pilot A (frozen baseline)	Pilot B (this record)
spatial cube, 1-based	x/y/z 1:256	x/y/z 513:768
spatial cube, 0-based	x/y/z 0:255	x/y/z 512:767
stored frames, 1-based	1:10	501:510
stored dt	0.002	0.002
physical time window	0.000 – 0.018	1.000 – 1.018
selection basis	—	half-domain translation of Pilot-A origin in x/y/z; 256-gridpoint unselected band each axis
source SHA-256	e32c9225af656a2f0fa0a704be7dcd78fa12efc1a01b880af23eb45e02108a46	977e6ab3c437252395dc7f7185af1829619fe1eec754f59f5f3ceae2ca0b969f
selection-lock SHA-256	—	385cca0a0b28ac3b17f757ef3b7fcfe2c1954423ceb5376b927b67633a40875e
acquisition method	Hugging Face mirror (ArielLubonja)	authenticated JHTDB getCutout, serial, 30-depth z-slabs

Independence rule satisfied exactly: 0 shared spatial grid points, 0 shared stored frames. Pilot B is a valid preregistered replication sample, not an internal subdivision of Pilot A.

§2 PRIMARY STRUC-ROUTE-I RESULT · run jhtdb_pilot_b_20260907_145640

10,515

STRUCTURAL OBJECTS

14,169

RELATION CANDIDATES

14,118

ELIGIBLE RELATIONS

4,830

SCALE SOURCE NODES

8,806

TIME SOURCE NODES

3,962

STITCH-ELIGIBLE SOURCES

100/100

NULLS ACCEPTED

REAL-GRAPH NULL MATCHES

VERDICT · 2/3 SUPPORTIVE · 0 OPPOSING

CONSTRAINED_ROUTING tags: PERSISTENT_TIME · LOW_STITCH_DEFECT

Test	Real	Null mean	z_fav	p_fav	Read
scale persistence	0.346636	0.346954	−0.590	0.742574	not significant
time persistence	0.500681	0.488256	12.520	0.009901	supported
D□ stitch defect	0.014172	0.996336	868.885	0.009901	strongly supported

Null model: directed layer-degree-preserving endpoint swaps · 7,047,947 / 7,056,000 swap attempts accepted · mean mobility 0.998859 · 100 unique null graphs (fraction 1.0).

STITCH DEFECT BY SCALE INDEX · STITCH_BY_SCALE.csv

scale idx	eligible	mean D□	mean enstrophy	mean diss. proxy
0 (×1→×2)	2,609	0.014390	1585.76	336.50
1 (×2→×4)	1,234	0.015058	950.92	170.91
2 (×4→×8)	114	0.000195	488.31	86.84
3 (×8→×16)	5	0.0	208.01	58.87

Population collapses sharply above the ×4 scale gap (114 → 5 eligible) — consistent with the sparsest layer contributing negligibly, as in Pilot A.

STITCH DEFECT BY TIME INDEX · STITCH_BY_TIME.csv · 9 consecutive frame transitions

t01→t02 (n=455)

D=0.01126

t02→t03 (n=442)

D=0.00848

t03→t04 (n=438)

D=0.01387

t04→t05 (n=438)

D=0.01486

t05→t06 (n=430)

D=0.01028

t06→t07 (n=444)

D=0.01751

t07→t08 (n=439)

D=0.02063

t08→t09 (n=435)

D=0.01313

t09→t10 (n=441)

D=0.01752

Bar width ∝ mean D□ relative to the largest transition (t06→t07, D=0.0206). No single frame transition dominates — the low stitching defect is distributed evenly across the 10-frame window, not concentrated at one time boundary.

§3 ROUTE-I SENSITIVITY [1,2,4,8] · factor-16 layer excluded · run jhtdb_pilot_b_20260907_154708

PRIMARY vs SENSITIVITY · qualitative verdict unchanged after removing the sparsest scale layer

Metric	Primary (factors 1,2,4,8,16)	Sensitivity (factors 1,2,4,8)
verdict	CONSTRAINED_ROUTING	CONSTRAINED_ROUTING
objects	10,515	10,487
relations supplied / eligible	14,169 / 14,118	14,139 / 14,088
scale persistence (real / null / p_fav)	0.346636 / 0.346954 / 0.742574	0.462341 / 0.462809 / 0.742574
time persistence (real / null / p_fav)	0.500681 / 0.488256 / 0.009901	0.500709 / 0.488242 / 0.009901
D□ (real / null / p_fav)	0.014172 / 0.996336 / 0.009901	0.014190 / 0.996365 / 0.009901
stitch-eligible sources	3,962	3,957
null mean mobility / unique fraction	0.998859 / 1.0	0.998869 / 1.0

Removing factor-16 raises absolute scale-persistence values (fewer, coarser scale gaps) but leaves every qualitative read — verdict, time-persistence significance, D□ significance — unchanged. The primary Pilot-B result is not an artifact of the sparsest scale layer.

§4 ROUTE LADDER EXTRACTION · ROUTE_EXTRACT_PB_v013 · frozen deterministic descendants

3,962

D_STITCH POPULATION

8,806

P_TIME POPULATION

4,830

P_SCALE POPULATION

DEDUPLICATION / JITTER / SMOOTHING APPLIED

Rules preserved from the frozen extractor: duplicates preserved, no jitter, no smoothing, no normalization or rescaling, stable ordering only, full stored numeric precision. Canonical STRUC-I inputs: ladders\struc_i\pilot_b\. Canonical STRUC-PERC-I inputs: ladders\struc_perc_i\pilot_b\.

§5 STRUC-I v1.0.4 — ADMISSIBILITY RESULTS · 2,000-MC primary · 10,000-MC precision rerun

STRUC-I RESULTS · 40 κ steps · κ ∈ [0.01, 1]

Ladder	MC	Regime / State	mean Aκ	min Aκ
D_STITCH	2,000	STRUCTURAL BOUNDARY / TRANS.	0.851863	0.7925
P_SCALE	2,000	STRUCTURAL INSTABILITY / RANDOM	0.564837	0.5395
P_TIME	2,000	STRUCTURAL INSTABILITY / RANDOM	0.597188	0.562
P_TIME (precision)	10,000	STRUCTURAL INSTABILITY / RANDOM	0.598235	0.5784

D_STITCH sits 0.078 below the 0.93 Geometric-Persistence boundary (not boundary-adjacent; no rerun triggered). P_TIME sits only 0.0028 below the 0.60 Structural-Instability/Boundary line, so the frozen precision rule triggered a 10,000-MC rerun — final distance 0.0018, regime unchanged.

Aκ vs κ — D_STITCH (2,000 MC) · domain 0.75–1.00 · dashed = 0.93 GP boundary

D_STITCH oscillates between Aκ≈0.79 and Aκ≈0.92 across the full κ range without ever reaching 0.93 — a stable Structural-Boundary read, not a transient dip.

Aκ vs κ — P_TIME (10,000-MC precision rerun) · domain 0.55–0.70 · dashed = 0.60 SI/SB boundary

P_TIME sits just below the 0.60 line through most of the κ range and only crosses it at the largest perturbation scales (κ>0.55) — the boundary-adjacency that triggered the frozen 10,000-MC precision rerun is visible directly in the curve.

§6 STRUC-PERC-I v2.5.0 — CONNECTIVITY RESULTS

STRUC-PERC-I BATCH · 3 ladders · all FULL_PERCOLATION

Ladder	Verdict	n	Giant ratio	κ_conn	Tail dominance
D_STITCH	FULL PERCOLATION	336	1.000000	10	0.3067
P_SCALE	FULL PERCOLATION	3	1.000000	0.010	0
P_TIME	FULL PERCOLATION	9	1.000000	1	0

D_STITCH clears the preregistered R4 minimum (giant ratio ≥ 0.95, gap vertices ≥ 100) decisively at 336 gap vertices post-deduplication — the substantive connectivity condition passes without qualification.

§7 STITCH-MECH v0.1.1 — HARDER NULL HIERARCHY

SURVIVES

MECHANISTIC VERDICT

0.01417

REAL D□ (EXACTLY REPRODUCED)

98.87

z_fav vs N1

35.97

z_fav vs N2

0.0099

p_fav (N0 · N1 · N2, ALL)

100/100

N1 · N2 UNIQUE NULLS

NULL HIERARCHY — D□ COLLAPSE · N0 → N1 → N2 → REAL

N0 — degree-preserving

0.99634

N1 — +distance-stratified

0.32620

N2 — +feature-stratified

0.21145

REAL

0.01417

Bar width ∝ mean D□ on a 0–1 domain. Local graph geometry (N1: distance-stratified; N2: +feature-similarity-stratified) explains part of the N0 contrast but leaves a large residual — the real defect is 93.3% below N2, not merely below N0.

N1 / N2 QUALITY GATES

Null	Mean D□	real/null	mean mobility	unique graphs	real match
N1 (distance-stratified)	0.326204	0.04344	0.044413	100/100	0
N2 (+feature-stratified)	0.211452	0.06702	0.014257	100/100	0

All thresholds passed with no quality flags: min mean-mobility 0.01, min unique-graph fraction 0.1, max real-match fraction 0.9. N1/N2 remain graph-local geometry controls, not exact field-level overlap/IoU-preserving surrogates.

FINDING §7.1

The route mechanism reproduces in an independent sample and survives a harder null than Pilot A used alone

Pilot B's real D□ (0.014172) is 46.8% lower than Pilot A's real D□ (0.026628) — the independent sample shows an even stronger stitching-defect suppression than the original. Critically, this low value is not a fragile threshold effect: it clears the same N1/N2 quality gates Pilot A cleared, with comparable null means (N1: 0.3262 vs Pilot A's 0.3291; N2: 0.2115 vs Pilot A's 0.2236). The N1/REAL and N2/REAL ratios are in fact larger in Pilot B (23.0× and 14.9× respectively) than in Pilot A (12.4× and 8.4×), meaning the separation between the real defect and its local-geometry-controlled null is more pronounced in the replication sample, not less.

§8 R7 — PHYSICAL-INTENSITY LOCALIZATION · frozen selector fails; alternative tail definitions are descriptive

R7 status: FAIL. The frozen 90th-percentile D□ threshold evaluates to exactly 0.0 in Pilot B because more than 90% of the 3,962 eligible objects sit at D□=0 — the nominal "top-decile" tail is therefore the entire eligible population (3,962/3,962), producing degenerate enrichment (1.0×) and p=1.0 for both physical-intensity measures. This is recorded as a genuine FAIL, not UNDERRESOLVED — the frozen protocol reserves that label for unusable N1/N2 tests, not for a degenerate quantile selector. No post-hoc tail repair is applied.

FROZEN 90TH-PERCENTILE SELECTOR · degenerate at D□=0

Measure	Enrichment	Permutation p	topD n / eligible n
enstrophy	1.0×	1.0	3,962 / 3,962
dissipation proxy	1.0×	1.0	3,962 / 3,962

Ordinary (unconditioned) rank association remains positive: ρ(D□,enstrophy)=0.1021, ρ(D□,diss.proxy)=0.1266. Geometry-conditioned partial-rank: ρ_partial(D□,enstrophy)=0.1027, ρ_partial(D□,diss.proxy)=0.0494 — both weaker than Pilot A's 0.179 / 0.120, but still positive.

ALTERNATIVE FIXED-THRESHOLD TAILS · descriptive only — do not override R7

Tail def.	n	frac.	D median	enstrophy ratio	diss. ratio
top10pct (frozen)	3,962	100%	0.0	n/a	n/a
top5pct	199	5.02%	0.2286	1.187×	1.290×
top1pct	41	1.03%	0.5887	1.129×	1.272×
D□ ≥ 0.1	144	3.63%	0.3341	1.173×	1.275×
D□ ≥ 0.5	33	0.83%	0.6516	1.151×	1.321×

Every fixed-value tail definition (not conditioned on the corpus's rank quantiles) shows consistent, modest positive enrichment (1.13×–1.32×) in both physical-intensity measures. This is a genuine descriptive signal — but it is not the preregistered test, and it does not convert the frozen R7 FAIL into a pass.

FINDING §8.1

R7 fails on a selector artifact, not on an absence of physical association

Pilot B's D□ distribution has an unusually large atom at exactly zero (p90 = 0.0), which breaks the frozen quantile-based tail selector rather than reflecting a genuine absence of physical localization. Fixed-value alternative thresholds all recover the same qualitative direction seen in Pilot A — elevated D□ co-occurs with elevated enstrophy and dissipation proxy — at roughly half Pilot A's effect size (Pilot A: 2.50× / 1.87×; Pilot B alternative tails: ~1.13–1.32×). The preregistered protocol correctly records this as FAIL rather than rescuing it, since no alternative tail definition was specified in advance. The honest reading is: physical-intensity localization is present but weaker and its measurement is selector-sensitive in this sample.

§9 R8 — RESTRUCTURING LOCALIZATION

BRANCH / MERGE CONTRASTS · D□ median difference, flagged vs unflagged

Flag	Flagged n	Unflagged n	Median Δ	p	Replicated positive?
scale branching	237	3,725	+0.0002180745	0.00019996	YES
time branching	81	3,881	+0.0000111951	0.00019996	YES
scale merging	178	3,784	0	1.0	NO
time merging	60	3,902	0	1.0	NO
any restructuring	428	3,534	0	1.0	NO

R8 requires at least one preregistered restructuring class to show significantly higher D□ (p<0.05); Pilot B has two. Pilot A additionally showed significant time merging and any-restructuring contrasts that Pilot B does not reproduce — the replicated statement is narrower: noncommutation localizes around branching events specifically, not merging or aggregate restructuring.

§10 FROZEN R1–R8 CRITERION RECORD — FINAL

CLASSIFICATION: STRUCTURAL PARTIAL REPLICATION · R2 passes and R5+R6 pass, but R3 (cross-chamber) fails — not CORE REPLICATION

ID	Frozen test	Pilot-B result	Status
R1	ROUTE-I organization	CONSTRAINED_ROUTING; 100/100 nulls; mobility 0.998859; unique 1.0; match 0	PASS
R2	Primary N0 stitching contrast	real D□ 0.0141716 ≪ N0 mean 0.9963364; z 868.885; p 0.009901	PASS
R3	STRUC-I D_STITCH → Geometric Persistence or stronger	mean Aκ 0.851863; Structural Boundary	FAIL
R4	STRUC-PERC-I substantive connectivity	FULL_PERCOLATION; giant ratio 1.0; gaps 336; isolated 0	PASS
R5	N1 survival	N1 usable; mean 0.326204; real/N1 0.043444; z 98.874; p 0.009901	PASS
R6	N2 survival (principal mechanism target)	N2 usable; mean 0.211452; real/N2 0.067021; z 35.966; p 0.009901	PASS
R7	Physical-intensity localization	enstrophy enrichment 1.0, p 1.0; dissipation enrichment 1.0, p 1.0	FAIL
R8	Restructuring localization	scale branching p=0.00020; time branching p=0.00020	PASS

Additional preregistered robustness result (not itself an R-criterion): the [1,2,4,8] scale sensitivity run also returns CONSTRAINED_ROUTING.

§11 DIRECT PILOT A ↔ PILOT B COMPARISON

CRITERION-BY-CRITERION COMPARISON · Pilot-A column maps the frozen baseline onto the later R1–R8 definitions for direct comparison

ID	Pilot A baseline	Pilot B replication	Comparison
R1	PASS — CONSTRAINED_ROUTING	PASS — CONSTRAINED_ROUTING	route verdict replicates
R2	PASS — real 0.026628 vs N0 0.996786, p 0.009901	PASS — real 0.014172 vs N0 0.996336, p 0.009901	primary stitching contrast replicates
R3	PASS — Aκ 0.971708, Geometric Persistence	FAIL — Aκ 0.851863, Structural Boundary	cross-chamber admissibility does not replicate
R4	PASS — giant 0.996933, gaps 652, isolated 0	PASS — giant 1.0, gaps 336, isolated 0	substantive connectivity replicates
R5	PASS — N1 0.329091, p 0.009901	PASS — N1 0.326204, p 0.009901	distance-controlled survival replicates
R6	PASS — N2 0.223649, p 0.009901	PASS — N2 0.211452, p 0.009901	harder mechanism target replicates
R7	PASS — enstrophy 2.50×, diss. 1.87×, p≈0.0002	FAIL — both 1.0×, p=1.0	tail localization does not replicate
R8	PASS — branching + time merging + any restructure	PASS — scale + time branching	replicates in narrower form

QUANTITATIVE STITCHING HIERARCHY · D□ across N0 → N1 → N2 → REAL

Quantity	Pilot A	Pilot B	B relative to A
REAL D□	0.026628	0.014172	46.8% lower
N0 mean	0.996786	0.996336	0.045% lower
N1 mean	0.329091	0.326204	0.88% lower
N2 mean	0.223649	0.211452	5.45% lower
N1 / REAL	12.36×	23.02×	larger separation in B
N2 / REAL	8.40×	14.92×	larger separation in B

The independent null hierarchy (N0/N1/N2) stays numerically close between pilots while the real Pilot-B stitching defect is substantially smaller — the replicated feature is not a fragile threshold effect but the persistence of an exceptionally low real D□ relative to closely comparable constrained-null ensembles.

FINDING §11.1

Two separable questions: route-mechanism reproducibility vs. full cross-chamber signature reproducibility

Across two independent forced-isotropic samples: CONSTRAINED_ROUTING reappears, real D□ is far below N0 in both, low real D□ survives both N1 and N2 in both, N1/N2 pass all frozen quality gates in both, substantive D_STITCH connectivity appears in both, and branching-associated elevated D□ appears in both. Not established across both pilots: STRUC-I Geometric Persistence of D_STITCH, upper-decile physical-intensity enrichment under the frozen selector, time-merging/aggregate restructuring localization, Reynolds-number generality, or generality beyond forced isotropic turbulence. The strongest two-pilot statement is a reproducible, largely commuting scale–time route regime with sample-dependent scalar admissibility strength and sample-dependent localization details of the rare noncommuting population.

§12 EPISTEMIC STATUS — SUPPORTED / NOT CLAIMED

SUPPORTED BY THE COMPLETED TWO-PILOT RECORD

✓ Pilot B is a valid preregistered independent sample — zero spatial/time overlap with Pilot A

✓ CONSTRAINED_ROUTING reappears under the frozen ROUTE-I null in an independent sample

✓ Real D□ is far below its degree-preserving null (N0) in both pilots

✓ Low real D□ survives both the distance-controlled (N1) and distance+feature-controlled (N2) nulls in both pilots

✓ N1/N2 pass all frozen quality gates in both pilots — no flags

✓ D_STITCH reaches FULL_PERCOLATION with substantive gap-vertex support in both pilots

✓ [1,2,4,8] scale-sensitivity preserves CONSTRAINED_ROUTING in both pilots

✓ Elevated D□ localizes around branching events in both pilots

✓ Pilot-B real D□ is 46.8% lower than Pilot A's — replication strengthens, not weakens, the core defect suppression

NOT CLAIMED BY THIS RECORD

✗ CORE REPLICATION — R3 fails, so the strict R1–R6-all-pass bar is not met

✗ That D_STITCH reaches Geometric Persistence in Pilot B (it reaches Structural Boundary)

✗ That the preregistered upper-decile physical-intensity localization replicates (frozen selector degenerate; FAIL is genuine)

✗ That time-merging or aggregate-restructuring localization replicates

✗ Reynolds-number generality

✗ Generality beyond forced isotropic turbulence

✗ Survival under a field-level surrogate preserving exact newly evaluated voxel overlap/IoU

✗ Equivalence of D□ breakdowns with a standard intermittency observable

Correct claim: the route mechanism (constrained routing + low, null-surviving D□) is reproducible; the full cross-chamber admissibility and localization signature is sample-dependent.

§13 NEXT WORK — STOPPING RULE ENGAGED · PILOT C DECISION PENDING

PILOT B IS FROZEN AS OBSERVED — NEXT DECISION POINTS

STOPPING RULE

Pilot B is one preregistered physical sample and must not be replaced under the same label after seeing its outcome. Any further independent sample requires a separately preregistered Pilot C.

PILOT C DECISION

The R3 and R7 divergences from Pilot A are the open question. A third independent sample, preregistered before acquisition, would test whether Structural-Boundary D_STITCH and degenerate tail-selector behavior are Pilot-B-specific or recur.

SELECTOR REFINEMENT

R7's frozen 90th-percentile selector is degenerate whenever D□ has a large zero-atom. A preregistered alternative (e.g. a fixed absolute threshold) could be specified for Pilot C in advance — not retrofit onto Pilot B.

No threshold, selector, or criterion is altered post hoc for Pilot B. Any change is a new preregistration version for Pilot C or later.

FINAL INTEGRATED STATEMENT

The preregistered Pilot-B replication is complete and frozen. Its final classification is STRUCTURAL PARTIAL REPLICATION: R1, R2, R4, R5, R6, and R8 pass; R3 and R7 fail.

Pilot B independently reproduces the central route-mechanism result of Pilot A — CONSTRAINED_ROUTING, an exceptionally low real scale–time stitching defect, and survival of that low defect under both the distance-controlled (N1) and distance-plus-feature-controlled (N2) graph-local nulls — in a spatially and temporally disjoint sample of the same forced-isotropic dataset. It does not reproduce Pilot A's complete cross-chamber signature: the scalar STRUC-I admissibility of D_STITCH falls short of Geometric Persistence, and the frozen upper-decile physical-intensity localization degenerates against a zero-heavy D□ distribution.

The two-pilot record separates a question that Pilot A alone could not: route-mechanism reproducibility is strongly supported; full cross-chamber/localization signature reproducibility is not established. Pilot B is now frozen as observed. The next decision — whether the R3/R7 divergences justify a separately preregistered Pilot C — is made from this frozen comparison, not by altering Pilot B or retuning its thresholds.

Data sources: Johns Hopkins Turbulence Database (JHTDB), isotropic1024coarse forced-isotropic dataset, authenticated getCutout acquisition · Pilot-B source SHA-256 977e6ab3c437252395dc7f7185af1829619fe1eec754f59f5f3ceae2ca0b969f · frozen instruments STRUC-ROUTE-I v0.1.2 / ROUTE_LADDER_EXTRACT / STRUC-I v1.0.4 / STRUC-PERC-I v2.5.0 / STITCH-MECH v0.1.1 · records: R1_R8.json/.md, SYNTHESIS.json/.md (analysis/replication/jhtdb_pilot_b/, analysis/synthesis/jhtdb_pilot_b/) · record status FINAL_FROZEN_RESULT, created 2026-09-07T22:11:59Z · All claims scoped to the tested two-pilot corpus.