SP-006 — Solar-Wind Magnetic Helicity Replication: Evidence Pack

Decision and exact target

2026-10-22. Expands section 5 of the earlier triage. Research and bounded access precheck only. No replication experiment was approved, launched, or completed. No paid compute was provisioned.

WORTH AN EXPERIMENT. Specifically, a capped raw-reconstruction gate followed only on success by a preregistered replication. The reason is now stronger than “public data exist”: two official calibrated L2 CDFs were actually retrieved and parsed, and the archived implementation exposes concrete, testable reproducibility questions. This is not a finding that the physical claim is false.

Version-pinned target: Iizawa et al., arXiv 2507.13213v2, dated 2026-10-13, arxiv.org/html/… . It reports opposite-signed spectral peaks in single-spacecraft reduced radial magnetic polarization, with lower-frequency migration across distance. It is not a universal solar-wind handedness rule. Its data window is 2021-09-29 00:00 to 2021-11-22 00:00 UTC. At 0.1 AU it describes negative and positive peaks near 10 and 20 Hz. Its exponential curve is explicitly a rough guide, not a fitted parameter estimate.

The measured quantity is sigma(f) = 2 Im[C_TN(f)] / (P_TT(f) + P_NN(f)), with C_TN = mean(conj(FFT(B_T)) × FFT(B_N)), and exp(-2 pi i f t) Fourier convention. Retain the transverse denominator, not P_RR + P_TT + P_NN. It describes rotation around the radial axis; it is not full 3D helicity or a directly measured helicity flux.

Two separate questions:

• Primary: do opposite-signed lobes occur within the same qualified held-out hour more often than under the registered noise/sampling null?
• Secondary: do their frequencies migrate relative to local scales after correcting time joins, sampling and plasma-condition confounds?

Finding a pair answers only the first. Neither outcome establishes a causal inverse cascade, a helicity barrier, a unique wave mode, or evolution of the same plasma parcel. A second-order single-spacecraft statistic cannot by itself establish those mechanisms.

Raw CDFs actually retrieved

Direct downloads from the official FIELDS archive on 2026-10-22. Both are calibrated L2 fluxgate MAG RTN products at their recorded cadence, not uncalibrated telemetry, SCM, a merged MAG/SCM product, or plotting arrays. “Full-cadence” names the product, not a guarantee of its maximum rate throughout. The source preprocessing explicitly selects v02, so the sample used v02. All main magnetic vectors and timestamps in both downloaded files were read.

A. psp_fld_l2_mag_RTN_2021110106_v02.cdf

• 774,297 bytes; CDF 3.8.0; data version 02.
• Actual first/last record: 2021-11-01 11:05:49.490204160 to 11:59:59.958573568 UTC. Despite the 06 filename, this is only about 54 minutes of magnetic coverage, not six hours.
• 31,088 vectors; 30,450 valid and 638 nonfinite. No fill-value or out-of-range rows in this sample.
• Median dt 0.109226752 s; rate switches between 8 and 64 samples/cycle. One delta >1.5× median; largest 1.761183872 s. No non-increasing timestamps.
• 360 minute quality records, all zero. Zero flags clearly do not guarantee complete coverage or valid vectors.
• SHA-256: 6d7a10e1550113022e06f67fb7a16f6950f2928b04a6c8eebebf59ec2f53f542
• Source: research.ssl.berkeley.edu/data/…/psp_fld_l2_mag_RTN_2021110106_v02.cdf

B. psp_fld_l2_mag_RTN_2021111600_v02.cdf

• 73,118,730 bytes; CDF 3.8.0; data version 02.
• Actual first/last: 2021-11-16 00:00:00.004527104 to 05:59:59.994375680 UTC.
• 3,164,053 vectors, all finite, within declared valid range, and not fill values.
• Median dt 0.006826624 s, about 146.5 Hz sampling; 128 samples/cycle throughout. Delta range 0.006690048–0.006935168 s. No non-increasing timestamps or gaps >1.5× median.
• 360 minute quality records: 346 zero and 14 flag=1, antenna bias sweep. Their effect on MAG still requires product-specific evaluation; primary proposed eligibility conservatively masks all flagged minutes.
• SHA-256: 25668de10e78e03daffe14664a3864e94281c118b9be627095c715f3e9a60355
• Source: research.ssl.berkeley.edu/data/…/psp_fld_l2_mag_RTN_2021111600_v02.cdf

Total CDF payload: 73,893,027 bytes. Both dates lie inside the source study window. These are access/cadence test cases, not asserted matches to a published one-hour spectrum. Exact radius and

published-hour matching remain untested. The sample does not establish availability or quality of every needed date, the 293 Hz regime, SWEAP, ephemeris, or held-out encounters.

Readability and provenance

Read using cdflib 1.3.16. Main variables: epoch_mag_RTN (CDF_TIME_TT2000, nanoseconds on the TT/J2000 basis) and psp_fld_l2_mag_RTN (N×3 float32, nT). Epochs were read as int64; cadence differences were formed as integer nanoseconds before division by 1e9. UTC conversion used cdflib, not a float Unix epoch. Four sampled TT2000 breakdown/reconstruction round trips per file returned exactly zero nanosecond difference. Stored epochs themselves lie on a 128 ns grid; the printed digits preserve stored values, not nanosecond physical accuracy. This check does not establish the origin or physical significance of the source quantization. Labels actually read: B_R, B_T, B_N; indices 1, 2, 3. Additional variables expose mission elapsed time, range, sensor mode, sample rate, packet index, and separate quality epochs/flags.

Both CDFs identify outboard MAG (sensor mode 0). This mode variable identifies the sensor, not survey/burst cadence; rate is recorded separately. The paper names the FIELDS fluxgate and describes a survey-mode limitation near 0.2 AU; exact paper-hour cadence remains a reconstruction question. Metadata declares FILLVAL=-1e31 and valid component range ±65536 nT. Both record v02 corrections to coordinate-transformation timing, sensor nonorthogonality, and downsampling-filter phase. The manifest preserves calibration and attitude-parent filenames. There is no embedded CDF checksum in these two files; SHA-256 above was calculated locally after download, so it is a reproducibility fingerprint, not an independently published checksum.

Quality flags are a bit field on their own 60 s time axis. Preserve and join by time, never by equal row index. The inspected files describe values 1 through 512; current official documentation adds 1024. Their declared VALIDMAX=255 does not encompass every documented bit. Decode documented bits rather than treating the bound as a complete whitelist. Record unknown bits and stop for review if present. This sample contains only 0 and 1.

The CDF confirms RTN labels. The standard right-handed convention is R Sun→spacecraft, T proportional to solar-spin-axis × R, N=R×T. A future velocity/ephemeris transformation must explicitly use that same convention. The authors' velocity helper instead constructs its normal from r×v; do not silently equate an orbital triad with heliographic RTN. Source for standard convention: cdaweb.gsfc.nasa.gov/pub/data/…/coordinate_systems.html . Dataset specification: spase-metadata.org/NASA/…/… .

Access route: the canonical research.ssl.berkeley.edu archive succeeded with ordinary HTTPS. An initially incorrect fields.ssl.berkeley.edu file-directory URL returned 403; that was not evidence that the public data are unavailable. No browser/security warning was bypassed.

Environment: Python 3.12.14, NumPy 2.3.5, SciPy 1.17.0; cdflib 1.3.16 and h5py 3.15.1 installed as binary wheels from PyPI into the bounded workspace. No authors' script was executed. No spectral estimation on raw data, synthetic validation, scientific hypothesis test, held-out download, or paid service was run. Parsing and archive inspection are the completed tests.

Official product and quality documentation: fields.ssl.berkeley.edu/data . Official MAG v02 correction release: fields.ssl.berkeley.edu/2021/06/release-notes-for-psp-fields-data-release-8 .

Zenodo archive actual inventory

Downloaded the actual Zenodo 22779133 v0.1.5 ZIP, not just its landing page: 20,769,258 bytes. Published MD5 754ac787f81044a4608196d1c51ab182 matched locally. Local SHA-256: 8c83d874bd53153c6a42d3b70da0ea5bd61e47aa0b3dfa0095df5be9c3186a7c. ZIP root identifies commit d370ba5. Archive contains 28 files plus seven directory entries, 29,525,451 uncompressed bytes.

Actual content:

• Four plot data files: fig1_data.csv (68,522 bytes), fig2_data.h5 (14,254,268), fig3_data.h5 (233,760), fig4_data.h5 (10,014,980).
• Nine rendered figures: one PDF and eight PNGs.
• calculate_plot_data.py; plot_figure1.py through plot_figure4.py; prep_step0_plasma_and_position.py through

prep_step3_generate_isotemporal.py; run_all_preprocessing.sh.

• One text SPICE meta-kernel; README, CITATION.cff, LICENSE and .gitignore.
• No raw CDF, IPC, Parquet, timestamp-list file, binary SPICE kernel, requirements lockfile, or raw-to-figure checksum manifest.

Successfully opened all three HDF5 files and the CSV, without running supplied code. CSV: 271 rows, 15 columns. Fig2: 271×4097 top spectral array and 8×4097 representatives. Fig3: eight distances with 1025 frequency/sigma/CI values each. Fig4: 271 distances with 1025 spectral values each in three panels. Archive arrays are useful comparison targets, not independent evidence of raw-data reconstruction.

Inputs missing from the archive, mapped to source stage: 1. Step 0: raw SWEAP SPC L3, SPAN-I spi_sf00 L3 v04, and psp_helio1hr_position_20180813_v01.cdf. Outputs absent: psp_swp_spi_sf00_L3_mom_v04.parquet, psp_sweap_spc_l3_vp_moment.parquet, psp_sweap_spc_l3.ipc, psp_helio1hr_position_20180813_v01.ipc. 2. Step 1: raw MAG RTN v02 files. Outputs absent: psp_fld_l2_mag_RTN_2021_####_v02.parquet chunks. 3. Step 2: psp_fld_l2_mag_RTN_2021_v02_cat.ipc in both intermediate directories and the corresponding ..._cat_position.ipc. 4. Step 3: all 01au_isotemporal_*.ipc variants and num_timestamp_starttimestamp_list.txt. Fig3 additionally requests 01au_isotemporal_0.1_3415.ipc, whereas the wrapper's discrete 0.1 call uses the suffixless path at 0.109227 s. That required high-cadence filename is not generated by the shown wrapper. 5. Meta-kernel dependencies absent: naif0012.tls, pck00011.tpc, spp_nom_20180812_20300101_v043_PostV7.bsp, de442s.bsp. Code references kernels/mk while the ZIP directory is kernel/mk. Paths need reconciliation. Source archive: zenodo.org/records/22779133 . Pinned repository: github.com/Institute-of-Theoretical-Physics/helicity-evolution_psp/tree/0.1.5 . Its README openly describes the intermediate-file gap and reference-only, non-turnkey preprocessing. The September release predates October paper v2, so exact paper-to-release lineage is itself a gate requirement.

Material reproducibility findings

These are static code findings plus direct checks of archived arrays, not a raw-data test of the physical conclusions. The source claims to generate these figures, but its exact execution lineage into the deposited arrays is not verified. The array dimensions and CI pattern corroborate specific observations, not every code path.

1. Welch length mismatch. The paper's Fig3 caption says 8192 points; archived calc_mag_helicity_normalized_welch uses 2048 with 50% overlap. HDF5 frequency grids contain 1025 one-sided bins, consistent with 2048, not 8192. Magnitude PSD uses 8192. Calculation lines 412–425; reconcile rather than choose the favorable variant. 2. Significance problem. The archived signed-sigma confidence bounds multiply sigma by positive chi-square ratios. Therefore a nonzero estimate cannot obtain an interval crossing zero under that construction. Direct array check: all 8,184 finite nonzero Fig3 entries have bounds on the same side of zero. This is a mathematical defect for using “CI excludes zero” as a helicity sign/null test. It does not establish whether the visible peaks survive a valid test. Chi-square power intervals can be appropriate for nonnegative autospectral power under stated assumptions; that rationale does not justify applying positive scale factors to a signed cross-spectral ratio. Use independently validated uncertainty. Calculation lines 427–441. 3. Cadence-path mismatch. Fig3's 0.2 AU array stops at 4.577622749 Hz, while the Fig4 0.2 AU array reaches about 73.24 Hz. Fig3 and Fig4 use separate IPC loaders. Also the 0.1 AU Fig3 grid corresponds to 0.003415 s, versus 0.003414 s in the main grid. Calculation loader lines 152–180. The paper also prints 3441 microseconds near 0.1 AU, versus 3414/3415 microseconds in source paths. Do not silently correct or equate these. Unsupported frequencies are not physical nondetections. 4. Plasma-row indexing risk. In calc_figure4 the density and speed arrays span the full concatenated distance grid, but the index i restarts within each cadence group. Later groups use mean_densities_interp[i] and mean_flow_speeds_interp[i] without the group offset. That code path pairs some spectra with earlier plasma rows. Calculation lines 149–150, 466–494, 655–670 and

677–696. Join by UTC/interval ID. We have not established whether the archived arrays were generated by precisely this path or how much the claimed trend changes.

5. Normalization mismatch. The archived middle panel divides by f_ci×(1+Vsw/VA), not merely f_ci. Calculation lines 443–454 and 663–670. Register raw cyclotron and propagation-model normalizations separately. Correcting a mislabeled axis is not automatically a physical refutation. 6. Quality/selection risks. The supplied MAG converter retains magnetic values and timestamps, not the quality bit field. The spline step has no explicit maximum-gap guard. The radial threshold selector needs its exact UTC provenance recovered. Adjacent radius bins may use overlapping hours and cannot be treated as independent observations. MAG converter lines 25–41; spline extraction lines 90–106. Rerendering arrays cannot resolve them. Pinned calculation source: github.com/Institute-of-Theoretical-Physics/helicity-evolution_psp/blob/…/calculate_plot_data.py#… . Preprocessing: github.com/Institute-of-Theoretical-Physics/helicity-evolution_psp/blob/…/prep_step3_generate_isotemporal.py#… .

The promising publishable question is now a raw-lineage and valid-inference audit plus a bounded out-of-sample occurrence test, not a generic “does handedness flip?” plot exercise.

Proposed access gate

Proposed cap: two analyst-days = 16 analyst-hours total, not two elapsed days with unlimited staffing. This gate has NOT been executed. The completed precheck above does not count as passing it.

Hours 0–4: build a frozen raw manifest and recover exact original one-hour windows for 0.1, 0.2 and 0.3 AU. Pin raw versions, URL, retrieval date, byte size, SHA-256, actual first/last UTC, sample count/cadence regimes, units, RTN transform, fill/valid bounds, sensor/range, quality bits, calibration parents and ephemeris/plasma versions. Recover and document the source's interval-selection logic, not approximate calendar substitutes. Retrieve only necessary files, not the wrapper's multi-year mirror.

Hours 4–8: independently implement timestamp-safe reading and the documented reference spectrum. Run both 2048 and 8192 reconstructions to localize the mismatch. Keep the reference branch separate from corrected inference. Do not import or execute unreviewed authors' scripts. Check dimension, units, row joins, normalization and the Fig3-specific high-cadence path. Fix the uncertainty algorithm below and select its block length using original-period pilot data only. Full calibration follows under the later experiment cap, before held-out unblinding; the access gate alone does not establish statistical validity.

Hours 8–12: synthetic convention and reconstruction checks:

• With B_T=cos(2 pi f0 t), B_N=-sin(2 pi f0 t), the stated FFT/CSD convention must give sigma=+1 at a resolved pure tone; replacing -sin by +sin must give -1. Require absolute error ≤0.01 for noiseless resolved tones.
• Swapping T/N or reversing time must negate sigma within 1e-6; flipping both transverse components or a proper transverse-axis rotation must preserve it within 1e-6. These are mathematical tests, not assumptions about physical field polarity.
• Inject two separated opposite-rotation tones: recover both signs and peak frequencies within one native frequency bin. For linearly polarized noiseless input, require |sigma|≤0.01 in energized bins. Validate scale invariance and |sigma|≤1 up to 1e-6 numerical tolerance.

Hours 12–16: compare independently reconstructed spectra to archived targets at the exact matched hours. Proposed tolerances over predeclared finite, non-artifact bins: median absolute sigma difference ≤0.03; 95th-percentile difference ≤0.10; both dominant lobe signs agree and peak centers within max(one comparison bin, 5% frequency). Also require transverse-power scale agreement within 10% where applicable. Archive values are targets only, never raw inputs. Different cadence grids must be compared over matched supported frequencies, not padded.

PASS only if all three original intervals have reconciled provenance; the 2048/8192 discrepancy has a documented explanation; raw reconstruction satisfies tolerances; deterministic synthetic sign tests pass; quality/gap handling is explicit; and remaining plasma/normalization defects are localized with a correct independent

route. A corrected branch need not reproduce an erroneous original CI, but must explain the difference.

FAIL/STOP at 16 hours for unreconciled timestamps, unsupported versions/coordinates, missing essential files, failing sign tests, unexplained spectrum disagreement, or an unresolved path to correct plasma normalization. Record access/reconstruction failure, NOT refutation. No extra download campaign or author outreach is automatic. Re-entry requires the missing provenance/data or a separately approved amended gate. Source contact could help, but no external contact has been sent.

Frozen estimator and primary peak statistic

Everything below is a proposed preregistration, not a completed analysis. Numeric tolerances are study design choices, not established instrument standards. Freeze software, configuration and cohort hashes before examining held-out helicity.

Use independent Hann Welch PSD/CSD, 64-second segments, 50% overlap, constant detrending, one-sided density scaling, and C_TN=conj(T)×N. Explicitly construct the periodic Hann array instead of inheriting a version-sensitive default. Retain native cadence and round segment length to the nearest integer sample, recording the actual duration. Use sigma's transverse denominator. The reference reproduction branch separately retains the authors' 2048/8192 alternatives; do not mix that branch with held-out inference.

Primary scope: 0.100≤r<0.120 AU, fixed spacecraft-frequency bands L=[6,14) Hz and H=[14,30] Hz. These bands deliberately test a narrow near-Sun recurrence claim. A feature shifted outside them is a failure of this bounded criterion, not proof that no bihelical feature exists anywhere.

Exact peak statistic:

• Freeze candidate log-window centers c_j=6×2^(j/6) Hz for integer j≥0. Each window spans [c_j/2^(1/6), c_j×2^(1/6)]. Retain only windows wholly inside L or H. No data-dependent smoothing or band adjustment.
• For each candidate window w calculate a_w=2×sum(Im C_TN×df)/sum((P_TT+P_NN)×df). This is a ratio of integrated spectra, not an unweighted average of noisy normalized bins.
• For every low/high window pair compute q=max(min(-a_L,a_H), min(a_L,-a_H)); Q_i is the maximum q for hour i. The peak locations are the winning fixed-window centers; ties go to the lower frequency, then earlier center. Q_i>0 requires opposite signs in the SAME hour, with size equal to the weaker lobe. Record the raw signed amplitudes and all candidate scores.
• For encounter e, T_e=max_i Q_i. This is an occurrence endpoint, not an estimate that all hours or all solar wind are bihelical.

Proposed primary success: in EACH of E11 and E12, at least one eligible hour has Q_i≥0.20 and simultaneous confidence intervals for the two winning lobe amplitudes lie on opposite sides of zero. Use the simultaneous all-window intervals and Bonferroni allocation over all selected hours in BOTH encounters specified below, targeting nominal familywise error 5%. Their coverage protects both orientations and any selected pair without independence assumptions between hours. Repeat the complete maximization in every synthetic calibration realization. No uncorrected peak picking.

A broad band-average version is one declared sensitivity, not a replacement endpoint if this fails. Frequency bins are not independent replicates. Report failures and unsupported windows explicitly. Inward/outward sector signs remain visible; no post hoc sign flip to improve agreement.

Cross-spectrum convention source: docs.scipy.org/doc/scipy/…/scipy.signal.csd.html .

Held-out cohort and exclusions

Proposed held-out bounding windows, UTC:

• E11: [2022-02-10 00:00, 2022-03-14 00:00).
• E12: [2022-05-17 00:00, 2022-06-18 00:00).

These analyst-defined windows include the February 25 and June 1 perihelia. They are not assertions of complete usable data coverage. Mission references: svs.gsfc.nasa.gov/4970 and science.nasa.gov/blogs/…/parker-solar-probe-completes-12th-perihelion . No data from these windows were analyzed in this precheck.

Form nonoverlapping, UTC-aligned one-hour candidates with all valid ephemeris radii inside 0.100–0.120 AU. Keep inbound/outbound as separate legs. Select at most 12 per encounter-leg, 48 primary hours total. If more exist, choose 12 evenly spaced chronological ranks including endpoints,

floor fractional ranks and break ties earlier. Candidate selection uses timestamps, product versions, ephemeris and instrument-state eligibility only, never helicity. Freeze the list and an ordered reserve list before spectral inspection; use reserves only for a predeclared metadata/data-integrity failure, not weak peaks or failed stationarity.

Minimum interpretable cohort: 12 eligible primary hours overall, at least four per encounter and two per leg. This is an adequacy floor, not proof of statistical power. Do not silently widen radius bands, switch encounters or replace unfavorable hours. If the floor or synthetic power gate fails, stop as insufficient coverage/sensitivity.

Proposed primary exclusions:

• No complete support for both bands below 80% of Nyquist, mixed sensor/range/cadence in an accepted segment, timestamp reversal, fill/nonfinite/out-of-range vector, or unresolved calibration state.
• Reject any segment crossing a missing sample or delta >1.5× local median. Resample only within otherwise contiguous runs to correct measured timing jitter; validate that operation with exact recorded timestamps. No interpolation across missing-data gaps and no extrapolation. No anti-alias-free downsampling.
• Mask every minute with a nonzero or unknown FIELDS quality bit and ±60 s at each boundary for the conservative primary. Product-specific relevance is a declared sensitivity, not a reason to ignore flags after seeing a peak.
• Retain an hour only if ≥80% of nominal Welch segments remain and both peak-search bands have integrated transverse power at least 10 times a documented, applicable instrument noise baseline. If the noise baseline cannot be established, the criterion is unresolved, not passed.
• Clean stationary cohort: quarter-hour mean-field directions differ by ≤30 degrees and maximum/minimum quarter-hour transverse variance ≤2. Show the full quality-valid cohort separately so selection does not disappear from the report.

Flag current sheets and switchbacks using field rotation alone: reversal >90 degrees relative to a 30-minute running vector median, plus ±60 s boundaries. Present primary clean segments and the same selected hours with/without this mask. Do not define a switchback from helicity. Treat long sector changes separately from isolated returns. Publish a complete inclusion/exclusion waterfall by encounter and leg.

Frozen uncertainty and calibration

Proposed, NOT validated. Freeze this algorithm, code, seeds and block length at the access gate using original-period pilots; then validate under the follow-on cap before held-out unblinding. Failure stops inference or requires an amended, refrozen plan, not silent method switching.

For every accepted 64 s Welch segment, retain jointly all candidate-window integrated imaginary cross-power numerators and transverse-power denominators. Resample these complete vectors together, preserving component/window relationships and overlap dependence. Use a moving-block bootstrap within each hour: uniformly draw all eligible overlapping blocks of L consecutive segment vectors, concatenate draws and truncate to the original number of segments. Blocks cannot cross a rejected segment, gap or regime boundary. Recompute each ratio a_w and the selected pair for every resample; never bootstrap already selected peaks alone.

Pilot rule: from original-period segment numerator/denominator series, find the first lag after which absolute autocorrelation is below 0.1 for three successive lags, taking the largest across nonconstant windows/series. Choose the smallest L in {4,8} at least twice that lag and freeze it globally. No such L, fewer than eight block-equivalents in an hour, or too few eligible contiguous blocks means inadequate uncertainty support, not a smaller convenient block. This finite-sample recipe must pass calibration; it is not an established instrument standard.

Use B=19,999 bootstrap draws per hour, seed 6006. For draw b calculate D_b=max_w |a*_w−a_w| across ALL candidate windows. Allocate alpha_h=0.05/48 to each hour, using the hard 48-hour maximum across BOTH encounters even if fewer survive. Sort D; take the order statistic at rank ceil((B+1)(1−alpha_h)), capped at B. Every window gets [a_w−c_h,a_w+c_h], clipped to [−1,1]. This nonstudentized simultaneous-band construction plus Bonferroni over hours is the one primary method. No

assumption of independent hours is needed for the union bound, but bootstrap coverage remains an assumption to validate. Cross-hour descriptive intervals use fixed six-hour clusters, with 12/24-hour sensitivity where supported; two encounters do not establish population universality.

Calibration uses synthetic multicomponent signals and real allowed timestamps/masks. Include colored zero polarization, one polarized band, two same-sign bands, opposite-sign lobes, instrumental lines, clipping, current-sheet steps, low SNR, jitter/interpolation and edge drift. Maximum four frozen representative cadence/mask classes. For each, run 2,000 composite-null cases and 1,000 injected alternatives: 500 at true weaker-lobe amplitude 0.30 and 125 each at 0, 0.10, 0.20 and 0.40. Each realization is a full mock cohort of 48 hours: alternatives place one pair in a fixed hour per encounter, all others null. Inject at c_4 and c_10 with the registered window width, stronger-lobe amplitude 0.50, and balance both sign orientations. Freeze noise levels/masks on original-period pilots. Truth means integrated window amplitude, not raw sinusoid amplitude.

Repeat the full endpoint, including Q≥0.20 and simultaneous sign intervals. Require empirical false positives ≤0.05 and their one-sided 95% Clopper–Pearson upper bound ≤0.065 under every null class. Require the one-sided 95% Clopper–Pearson lower bound on power ≥0.80 at the 0.30 alternative and median amplitude bias ≤0.03. Report the complete power curve by noise/coverage class. There is NO 80% power requirement at true amplitude 0.20, which is the observed-effect decision boundary. Monte Carlo tolerances are proposed checks, not proof of exact finite-sample error control. Stop if calibration or compute caps fail.

Controls: preserve raw polarity, leg, sampling angle and Vrel; audit noise/lines and sensor dependence (merged MAG/SCM is not independent MAG validation); show gaps, jitter, stationarity and switchback cuts; join quality-screened SWEAP by UTC, not row index.

Spacecraft-frequency inference needs no Taylor conversion. For spatial claims use Vrel=Vplasma−Vspacecraft in verified RTN, VA and epsilon=delta-u/(sqrt(2) Vperp). Proposed strict subset: epsilon≤0.3 and VA/Vperp≤0.3 with measured nonzero Vperp. These screens do not prove validity for dispersive modes. Report phase-speed/angle sensitivity or stop at frequency-only conclusions. arxiv.org/abs/1406.5470 .

Bounded migration extension

Secondary, separately gated question: does an estimated spectral scale retain radial dependence after local normalization and sampling/geometry controls? It is not automatically answered by primary pair occurrence.

Within the same held-out date windows allow up to 48 additional one-hour intervals, giving a hard total cap of 96. Freeze a metadata-selected, chronological-rank-balanced grid across [0.10,0.20), [0.20,0.30), [0.30,0.40] AU, each encounter and leg. Retain all failures/censoring; no new encounter or radial expansion. Require at least three populated radial strata and eight uncensored paired spectra per encounter, otherwise classify this extension as inadequate.

Freeze a broad absolute-frequency search from 0.5–30 Hz using the same fixed logarithmic-window estimator, identical quality rules and full-search null correction. The primary near-Sun bands remain unchanged. Do not slide the search windows along the paper's exponential guide. The secondary extrema must be separated by at least one full window width and not touch the supported-band boundary. Report candidate ambiguity, no detection, and one-lobe-only cases instead of forcing two frequencies.

For each retained lobe, report spacecraft frequency; f/f_ci with f_ci=q|B|/(2 pi m_p); and k di with k=2 pi f/|Vrel| only in the qualified Taylor subset. Keep the authors' f_ci×(1+Vsw/VA) normalization as a distinct, labeled model sensitivity. Do not conflate these definitions. Preserve uncertainty in B, density, velocity, angle and mapping.

Compare three prespecified trend descriptions: constant normalized scale, log(f)=a+b log(r), and log(f)=a+b r. Report predictive error by held-out leg and encounter, coefficient intervals and residual association with plasma/geometry strata. Evaluate raw local-normalized scales separately; a density/B/flow imbalance across radius can mimic evolution. With a small cohort do not fit an

unconstrained many-covariate model. If matched strata lack overlap, say the radial interpretation is unidentified.

Primary multiplicity is handled by the simultaneous full-search procedure. Secondary lobe × normalization × model contrasts form a separate declared family, controlled by Holm at 0.05; show all tested contrasts. The exploratory geometry/cut sensitivities cannot become the new confirmatory result. Compare attenuation on paired retained intervals, while also showing how exclusions change the cohort.

The explanatory bar is high: an exponential model must predict held-out data better than alternatives, with uncertainty, before describing exponential rather than generic monotone migration. No pre-fixed guide coefficient counts as an estimated result. Even then, this is a radial/time association across sampled plasma, not proof of transport in a tracked fluid parcel.

Effort and cost caps

The earlier “one to two weeks” is a planning envelope, not measured effort or a delivery promise. Proposed sequential authorization: first the 16-hour access/reconstruction gate; only on a pass and explicit experiment approval, a maximum of 10 additional analyst-days (80 hours). Parallel staffing does not increase the total cap.

Follow-on work allocation, caps that sum to 10 analyst-days: 1. Cohort/manifest completion, exact UTC joins and preregistration: 1 day. 2. Independent estimator, reference-versus-corrected branches and reproducibility tests: 1 day. 3. Composite-null, real-mask artifact tests, power calibration and uncertainty validation: 2 days. 4. Frozen held-out ingestion, exclusions and primary result: 1.5 days. 5. Polarity/geometry/stationarity/noise/switchback controls and paired attenuation analysis: 1.5 days. 6. Secondary migration analysis, only if eligibility permits: 1 day. 7. Discrepancy review, figures, executable artifact, write-up and release checks: 2 days. Stop at each phase's cap if its acceptance condition is not reached. Do not convert a blocked phase into a larger unpriced project. A five-day core-only outcome is possible only by dropping the secondary and limiting the deliverable, not by claiming the full ten-day plan is complete.

Proposed compute/storage caps: CPU only; 8 CPU threads, 16 GB RAM working envelope, 100 GB temporary storage, 25 GB raw transfer, 100 CPU-hours for the complete registered analysis matrix. Stream one hour/file at a time; no GPU or cluster requirement assumed. Benchmark one representative raw hour and one null batch before projecting cost. These are protective caps, not predicted usage or proof that every workload fits.

Measured here: 73.893 MB of CDFs plus 20.769 MB ZIP transferred; the two primary CDF array inspections took about 0.02 s and 0.51 s after download in this environment. These parser timings are not spectral, bootstrap or end-to-end benchmarks. No runtime extrapolation is presented as established.

Price accounting: public payload access had no purchase charge. No new paid compute was provisioned. Existing assistant/workspace consumption is not represented as free or costed from these timings. Cloud VM, storage/egress, analyst labor and specialist review prices are UNPRICED. Dollar total UNKNOWN. Before any paid run, obtain an actual provider quote, resource-hours/egress estimate and separate spending authorization. The analysis caps do not authorize a charge.

FIELDS publication practice: instrument citation and acknowledgement, appropriate data provenance, and scientific review are part of the release checklist. Its rules request early PI contact and sharing manuscripts at submission/publication. No such outreach was performed; it would need the user's communication authorization. fields.ssl.berkeley.edu/rules .

Results, publication bar and final call

Keep these outcomes separate:

ACCESS/RECONSTRUCTION FAILURE: original window identity, raw/calibration versions, exact convention, or reference spectrum cannot be reconciled within the gate. Produce the manifest and discrepancy log. No physical refutation claim.

INADEQUATE/INCONCLUSIVE: access succeeds, but qualified coverage, noise documentation, synthetic false-positive control or power does not pass. Report the missing sensitivity and do not search extra encounters until something turns positive.

ATTENUATION: valid reprocessing/control comparisons reduce the

same measured lobe amplitudes or detection evidence. Quantify paired changes and uncertainty. A point estimate dropping is not enough to claim statistically established attenuation, and nuisance removal can remove real shared signal.

PRIMARY NONREPLICATION: access and sensitivity pass, but the preregistered paired-peak criterion fails in one or both held-out encounters. State the exact radius/frequency/amplitude scope and uncertainty. This does not refute the existential observation that a pair occurred in the original data.

PRIMARY REPLICATION: the frozen occurrence criterion passes independently in both held-out encounters with simultaneous error control and documented controls. This supports recurring reduced radial polarization pairs, not ubiquity.

EXTENDED REPLICATION: primary occurrence plus separately supported local-scale-normalized migration under valid mapping and geometry controls. If primary passes but migration does not, report that split explicitly.

NARROW REFUTATION: reserve this stronger word for a specific registered quantitative assertion that is incompatible with a valid reconstruction or a powered/equivalence-bounded result. For example, if a new preregistration asserts an original-period lobe ≥0.20, a valid upper bound below 0.20 could contradict that quantitative assertion. This 0.20 threshold is ours, not a reported paper threshold. Do not attach “refuted” to the whole paper from code inspection, one null encounter, a failed download, or a p-value above 0.05. The currently inspected CI construction is invalid for the proposed zero-exclusion inference; the underlying physical claim has not been tested here.

Publishable unit: a versioned raw-file manifest, minimal independently implemented reader/estimator, immutable cohort/configuration, raw-to-spectrum lineage, synthetic validation and power surfaces, selected and failed spectra, valid uncertainty, control/exclusion tables, corrected-versus-reference discrepancies, and an encounter-specific report. Release code and compact derived results with source/archive attribution, not an undocumented cache of figures. Retain raw provider links and hashes so others can retrieve exact inputs.

Publication bar: a technically reviewed result that changes confidence in the physical inference, demonstrates a reproducibility defect with its actual effect, or gives a sufficiently powered independent replication. Merely rerendering the authors' HDF5 or listing suspected bugs is not automatically a journal paper. A negative result is informative only with access, sensitivity, scope and exclusions documented. Specialist interpretation is especially important for noise and Taylor/geometry issues. Publication acceptance and novelty are not guaranteed.

Sources actually inspected: pinned paper v2; checksum-verified Zenodo v0.1.5 archive and all three data HDF5s; pinned source/preprocessing text; two official v02 raw CDFs and their internal metadata; FIELDS product, correction and usage documentation; NASA mission timing references; SciPy CSD convention. File-level inventory, parser script, software versions and SHA-256 manifest accompany this pack. No authors' code or full science experiment was executed.

The broader study is not ready to run until the reconstruction gate passes and Johnny approves it. A gate failure stops the study until the specific provenance/reconstruction problem is resolved. No experiment is authorized by this recommendation.

Final classification for the specified 16-hour access/reconstruction gate: WORTH AN EXPERIMENT.