Paper 1 — Proposed Methodology

Status and decision boundary

This document defines the proposed scientific method for Paper 1. It is a protocol, not evidence that the method has been completed or that any model has passed verification or validation.

The accessible-source Evidence Review supports, and the approved D1 decision fixes, a Paper 1 scope of conventional-LINAC converted-X-ray shielding at incident-electron energies from 5 to 7.5 MeV. The owner has confirmed that institutional Scopus, Web of Science, INSPEC, and IEEE full-text access is unavailable; this is a declared coverage limitation, not a blocker. Four full-text records remain unresolved, so the review does not support a comprehensive database-coverage or absolute-novelty claim. The broader owner intention—direct-EB and converted-X-ray prediction from 0.2 to 10 MeV—remains a longer research programme, not the present Paper 1 claim.

No PHITS reference-dataset generation or surrogate performance claim should begin until the pre-study decisions below are approved and recorded in the Project.

Study design

Paper 1 will use a staged computational study to develop and evaluate a surrogate model of a defined PHITS input-to-field mapping. PHITS will provide reference calculations; it will not be treated as physical truth. The study will proceed through the following stages:

systematic evidence review and scope confirmation;
physical and numerical specification;
component and limiting-case verification;
preliminary PHITS calculations;
physical validation where applicable measurements are available;
PHITS reference-dataset generation;
surrogate development and model selection;
evaluation on an untouched final-test partition; and
independent PHITS verification of selected surrogate predictions.

A later stage may use only Results that satisfy the predefined criteria of all preceding stages. Failed or inconclusive Results will remain traceable records but will not silently enter the Dataset.

Current RAMAL-EBX boundary

The implemented GUI already supports the Case-to-Study-to-Dataset-to-Model-to- Prediction workflow, whole-configuration partitions, and independent Prediction Verification. Its current trainer accepts scalar continuous or binary features and dose fields normalized per primary source electron on one exact grid, and it compares three deterministic reduced-order baselines. If the approved Paper 1 specification requires a different target representation or a new spatial model, that reusable GUI capability must be implemented and verified before the method uses it. Available software will not determine the scientific specification.

Systematic evidence review and scope confirmation

Before finalizing the simulation domain, the existing narrative review will be extended using a reproducible review protocol. The protocol will record:

databases and official-source collections searched;
complete search strings and search dates;
publication years and languages covered;
inclusion and exclusion criteria;
duplicate-removal and screening procedures;
reasons for full-text exclusion;
source-quality and applicability assessment;
citation-chain searches; and
the final evidence table linking each methodological choice to its support.

The review will separately map evidence for industrial accelerator types, direct-EB and converted-X-ray source modes, incident-electron energies, converter systems, shielding geometries, radiation-protection quantities, PHITS verification, measurement-based validation, photonuclear transport, and radiation-transport surrogates. Evidence from medical LINACs, Rhodotrons, homogeneous slab studies, or product-dose studies may support a component method but will not establish applicability to the complete Paper 1 system.

The completed accessible-source review supports the narrower scope approved in D1. D2 fixes the field, and the 2026-09-11 D4 amendment fixes the primary closed-room orientation as vertically downward. D3 and D5 are amendment-open after the owner required an industrially representative broad scanned source; D6 fixes photonuclear comparisons and neutron retention; D7–D11 remain open.

Physical scope

Accelerator and source

The D3/D5 amendment target is a conventional industrial electron LINAC represented at the converter boundary. Each primary electron begins at the upstream face of a broad planar tantalum-water-steel assembly. The active tantalum plate is 100 cm x 10 cm; final tantalum, coolant, and cover thicknesses require revised component and sensitivity evidence. A normally incident circular Gaussian local spot with 2 cm FWHM and declared 3 cm support is distributed along the 100 cm direction by a normalized one-dimensional scan-position density. A uniform continuous line scan, approximated with normalized trapezoidal quadrature, is the initial explicit reference assumption, not a universal scanner law.

Each Case declares one monoenergetic primary-electron energy in the approved 5–7.5 MeV interval. The reference calculation explicitly transports the electron and all converter descendants. A reusable phase-space source may be used for room calculations only after its particle types, energy, position, direction, weights, normalization to original primary electrons, and converter conditions have been independently verified against the coupled calculation.

The accepted converter construction, local beam footprint, and scan profile will remain fixed conditions, not surrogate predictors. Endpoint-only sampling is insufficient for an interval claim: justified intermediate energies and at least one complete intermediate-energy condition excluded from model fitting remain required. D5 acceptance now requires scan sampler, normalization, and finite-edge verification before Dataset generation.

Shielding geometry

The D4-approved family is an axis-aligned rectangular closed room. x is the scan/lateral axis, y is vertical, and z is the second horizontal axis. The source is fixed at the room centre and points downward along -y; the Gaussian footprint and converter face therefore lie in the x-z plane. The five geometry predictors are interior width 4–10 m, length 6–16 m, height 3–6 m, primary- floor thickness 100–180 cm, and one shared 50–150 cm thickness for the roof and four side walls.

Every barrier uses PNNL ordinary NBS-03 concrete at 2.35 g/cm³; interior and exterior regions use PNNL dry air at 0.001205 g/cm³. The reportable domain includes interior air and a 100 cm exterior air shell with 30 cm reference surfaces. Doors, mazes, ducts, penetrations, reinforcement, joints, products, equipment, and other nearby structures are excluded. Openings require a separate geometry family; D8 sensitivity checks must justify omitted nearby structures before complete-room Dataset generation.

Predicted quantities and normalization

The approved D2 decision below defines the Paper 1 target as ambient dose equivalent, $H^*(10)$, calculated with the PHITS -200 coefficients and stored intrinsically in pSv/primary-source-electron. Photon, electron, positron, and neutron contributions remain separately inspectable before their predefined sum.

Dose rate is derived, not learned:

$$ \dot H^(10)\,[\mathrm{mSv/h}] = H^(10)\,[\mathrm{pSv/primary\ e^-}] \times \frac{I_{\mathrm{avg}}}{e} \times (3600\,\mathrm{s/h})(10^{-9}\,\mathrm{mSv/pSv}), $$

where $I_{\mathrm{avg}}$ is the time-averaged beam current and $e$ is the elementary charge. For pulsed operation, $I_{\mathrm{avg}}=I_{\mathrm{peak}}t_{\mathrm{pulse}}f_{\mathrm{pulse}}$. Exposure time, workload, and operating schedule are recorded outside the intrinsic transport field and applied only when the reported scenario requires them. The normalized scan-position density is part of the intrinsic source model; frequency, current, duty cycle, and workload remain operational provenance and are not additional transport multipliers.

Coordinates use the Design's right-handed Cartesian x,y,z system in centimetres. Values are cell quantities on an explicit rectilinear voxel grid; they are not point samples or interpolated values. D4 fixes the spatial bounds and D7 fixes the grid resolution and convergence criteria. A separately named ICRU Report 95 $H^*$ calculation using PHITS -204 may be reported as a secondary comparison, but it is not mixed into the primary Dataset target.

Photonuclear treatment

Neutron generation and transport will be evaluated for converted-X-ray configurations throughout the working 5–7.5 MeV interval because applicable IAEA guidance requires consideration at 5 MeV and above [1]. Photon and neutron contributions will be calculated separately.

The Project will record the photonuclear data library, reaction model, particle cutoffs, and transport settings. Sensitivity calculations will compare applicable data or model choices near threshold, informed by the published PHITS–MCNP6–TRIPOLI-4 comparisons [37,40,41]. D6 selects a conditional JENDL-5 reference, ENDF/B-VIII.1 and built-in-model comparisons, and retention of neutron dose in every primary combined field. No small-fraction omission rule applies.

Activation and post-shutdown dose are outside the present method. The study concerns prompt radiation during operation.

Parametric design

Only parameters varied in the PHITS calculations will be surrogate inputs:

incident-electron energy;
interior width, length, and height;
primary-floor thickness; and
shared secondary-barrier thickness.

The source placement, source orientation, converter, concrete, air, and exterior margin remain fixed conditions. Spatial position belongs to the target voxel grid rather than the Case-level predictor vector.

The final parameter table will state, for every variable, its physical meaning, unit, lower and upper bounds, sampling method, constraints, and whether it is continuous, categorical, or fixed. Published facility values will not be adopted as universal bounds.

Sampling will be incremental. Preliminary calculations will first vary one physical factor at a time around a reference configuration to expose numerical failures, strong interactions, and implausible bounds. The subsequent multivariable design will cover the approved domain without assuming that arbitrary combinations of geometry are physically valid. Constraints will be applied before Case creation so that every immutable Case is a coherent physical configuration.

PHITS reference calculations

Transport specification

Each immutable Case will record:

PHITS version and input;
converter and source definition;
material compositions and densities;
geometry parameters;
transported particles;
electromagnetic and photonuclear settings;
nuclear-data libraries;
energy cutoffs;
tallies and spatial grid;
dose-conversion multipliers;
source normalization;
variance-reduction settings;
histories and batches; and
random seed.

The same physical quantity, unit, normalization, particle definition, grid, and coordinate convention will be used across all Cases in one Study. A change to one of these definitions creates a different Study rather than an undocumented mixture.

Statistical reporting

Every scored field will retain the PHITS statistical uncertainty. An unscored voxel is missing information, not zero dose. Relative statistical error will not be interpreted without the corresponding scored value and history count.

The calculation-completion criteria will be set after preliminary PHITS calculations and before Dataset generation. They will include, at minimum:

a maximum relative statistical error for reportable positive voxels;
a minimum reportable voxel count;
a minimum reportable fraction within each evaluation region;
grid-convergence criteria;
repeated-run stability criteria using independent random seeds; and
a rule for calculations that remain unscored or poorly converged.

These numerical values are unresolved. They will be justified from preliminary calculation behaviour, intended use, and relevant statistical literature rather than selected to retain convenient Results.

Variance reduction

Variance reduction may be introduced for deep shielding after an unbiased reference calculation has established the expected behaviour. Each technique and parameter will be recorded. Checks will demonstrate that variance reduction does not introduce a detectable change in the expected tally result while improving statistical efficiency. Calculation time, uncertainty, and figure of merit will be compared using the same hardware and output definition.

Verification

Verification will test whether the implementation correctly represents its specification and numerical method. It is distinct from physical validation. The verification programme will include:

Input and record integrity. Reconstruct the generated PHITS input from each Case and verify source, geometry, materials, tallies, normalization, output roles, and hashes.
Electron transport. Compare stopping power and CSDA range for relevant materials with NIST ESTAR [15].
Photon interactions. Compare attenuation behaviour and material definitions with NIST XCOM [16].
Converter transport. Compare photon yield, residual-electron transport, spectra, and angular distributions with sufficiently specified published calculations [3,4,10,14,35].
Shield attenuation. Reproduce applicable simple concrete and angular attenuation calculations before using a complete room [3,33].
Opening transport. If a later scope includes a maze or penetration, reproduce an applicable fixed-geometry comparator before varying it [34].
Photonuclear sensitivity. Compare applicable data libraries or reaction models and selected inter-code results near threshold [37,40,41].
Dimensional and limiting cases. Check inverse-square behaviour where its assumptions hold, monotonic attenuation in simple slabs, energy and particle conservation diagnostics, grid refinement, source discretization, and broad symmetric geometries away from edge effects.
Independent implementation. Compare selected Cases with a separately configured transport code or independently prepared calculation where practical.

Agreement criteria for each comparison will be defined before observing the final comparison. A published benchmark will be used only when its source, geometry, material, quantity, normalization, and uncertainty can be reproduced closely enough for the stated comparison.

Physical validation

Validation will test whether the PHITS model adequately represents reality for the declared purpose and domain. It requires relevant measurements. Published calculations alone provide verification evidence, not physical validation.

Candidate validation evidence includes measured converter spectra, angular fields, component transmission, barrier attenuation, or facility survey data at applicable energies and configurations. For each comparison, the method will record:

measurement geometry and operating conditions;
detector type, response, calibration, and uncertainty;
background and environmental corrections;
spatial registration and averaging volume;
PHITS representation of the measurement; and
combined experimental and simulation uncertainty.

Validation will be reported only for the configurations and quantities actually compared. If no sufficiently applicable measurement is available, Paper 1 will state that limitation and will not describe agreement with published Monte Carlo calculations as validation.

Preliminary PHITS calculations

Preliminary PHITS calculations will establish the numerical method before the reference Dataset is generated. They will determine:

practical spatial resolution;
histories and batches required by region and field level;
occurrence of unscored voxels;
relative-error and reportable-coverage behaviour;
repeated-run stability;
source discretization convergence;
useful variance-reduction settings;
neutron contribution and photonuclear sensitivity; and
approximate runtime and storage cost.

These calculations will not be mixed silently into the final Dataset. A preliminary Result may be retained only if it uses the finalized specification, is independently reassessed, and meets the same inclusion criteria as every other Dataset member.

Approved Project and Study organization

The owner approved this organization on 2026-09-05T04:57:05Z:

Keep component verification, preliminary numerical calculations, and final Dataset generation inside the same Paper 1 Project.
Use separate Studies for D8 verification, D7 preliminary numerical work, and final Dataset generation.
Separate preliminary Studies whenever the grid, physics definition, or other Study-level field contract changes. Independent seeds and increased histories may remain within one comparison Study where its contract permits.
Preserve Case and Result traceability. Preliminary Results do not enter the final Dataset automatically; reuse requires explicit reassessment against the finalized D7 criteria and D10 partition rules.
Store Paper 1 settings and scientific decisions in Project records. RAMAL-EBX supplies reusable GUI operations, not hardcoded Paper 1 policy.

This approval fixes organization only. D7 numerical criteria remain open, and this record does not establish that the Studies or calculations already exist.

Study inclusion criteria

The inclusion criteria will be predefined in the Study before final Dataset generation. A Result may enter the Dataset only when:

its immutable Case, Job, Attempt, and Result records pass integrity checks;
it uses the Study's exact physical quantity, units, normalization, grid, and particle definitions;
the PHITS process completed successfully;
required output files and uncertainties are present and finite;
each required evaluation region meets the predefined reportable-count and reportable-fraction criteria;
applicable grid, repetition, and variance-reduction checks have passed; and
the configuration lies within the approved physical domain.

Exclusion reasons will be retained. Excluded Results will not be converted to zeros, repaired by undocumented interpolation, or replaced by a different seed under the same record identity.

Dataset construction

The Dataset will contain complete shielding configurations, not independently shuffled voxels. Every sample will retain its Case and Result traceability, physical feature values, target field, uncertainty field, reportable mask, spatial coordinates, and material representation.

Whole configurations will be assigned before model fitting to:

a training partition;
a model-selection partition, serialized by the current software as validation; and
an untouched final-test partition.

All replicas and paired source conditions belonging to one physical configuration will remain in one partition. No tally voxel, profile point, replica, or paired condition from a final-test configuration will appear in training or model selection. Partition counts, proportions, random seed if used, and grouping identifiers will be recorded before training.

The final-test design will include scientifically difficult conditions relevant to the claim, such as predefined domain boundaries and at least one held-out intermediate energy if energy interpolation is claimed. A geometry or material family will be withheld only if generalization to that family is an explicit research question and the representation can express it.

Surrogate development

The prediction target will be the reportable three-dimensional field on the predefined grid. The surrogate will receive only the approved physical inputs and geometry/material representation. It will not learn beam current when the field can be scaled analytically from the intrinsic per-source result.

Model development will follow increasing complexity:

nearest training configuration;
interpolation or a fitted attenuation response where applicable;
conventional regularized regression;
a Gaussian-process or other compact nonlinear baseline; and
a spatial neural model only if the simpler methods fail and the Dataset size and geometry representation justify it.

The currently implemented reduced-order nearest, quadratic POD-ridge, and radial-basis Gaussian-process models are baselines. Their presence in RAMAL-EBX does not predetermine the final Paper 1 model. Architecture, transformations, hyperparameters, loss, and stopping criteria will be chosen using only training and model-selection data.

The selected Model will be finalized before the final-test partition is opened. No architecture, feature, transformation, or hyperparameter will be changed in response to final-test performance. Any later change creates a new Model and requires a new untouched final test.

Evaluation

Evaluation will compare every surrogate and simple reference method against the same PHITS Results. Metrics will be reported overall and separately by incident- electron energy, geometry condition, barrier role, spatial evaluation region, field level, and particle contribution where applicable.

The evaluation will include:

signed error and bias;
mean and median absolute error;
relative error where the reference value is reportable;
logarithmic field error for the multi-order-of-magnitude range;
frequency and magnitude of underprediction and overprediction;
maximum and high-quantile error;
spatial error maps and one-dimensional error profiles;
comparison with PHITS statistical uncertainty;
empirical coverage if the surrogate reports an uncertainty interval; and
performance at predefined domain boundaries and held-out energies.

If a Project defines a design dose criterion, the analysis will report incorrect classification relative to that predefined criterion. It will not call this regulatory compliance unless the applicable jurisdiction, quantity, occupancy, workload, averaging period, and legal basis are explicitly established.

A favorable average score will not compensate for failure in a declared subdomain. Underprediction will be analyzed separately because it is non-conservative for shielding assessment.

Independent PHITS verification of predictions

Selected final-test predictions will be compared with independent PHITS runs using newly created immutable Cases or new Attempts with independent random seeds, as appropriate. Selection will include:

the largest underpredictions;
the largest overpredictions;
high-uncertainty predictions;
representative low-, middle-, and high-energy conditions;
domain-boundary conditions; and
conditions close to any predefined design dose criterion.

The independent run will use the same declared physical specification but will not reuse a stochastic Result as if it were independent. Differences will be interpreted together with both PHITS statistical uncertainties. A Prediction will remain a surrogate record and will never be promoted into a PHITS Result.

Computation-time comparison

“Rapid” will be evaluated operationally. The study will report, separately:

PHITS wall-clock and CPU time per configuration;
hardware and parallel-process count;
histories, batches, variance reduction, and achieved uncertainty;
Dataset generation and preprocessing time;
surrogate training and model-selection time;
final-test evaluation time;
single-configuration and batched inference time; and
end-to-end speedup for the same output definition.

One-dimensional profiles and two-dimensional maps

The primary scientific artifact remains the immutable three-dimensional rectilinear VTK field. One-dimensional profiles and two-dimensional maps are derived views of that same field; they must not become independent, manually edited Result records. RAMAL-EBX already extracts an X, Y, or Z profile through a declared physical position and reports the corresponding relative standard errors. D7 will predefine the profile axes and fixed coordinates used for component, room, and surrogate comparisons.

For two-dimensional reporting, Analyze must extract XY, XZ, or YZ slices at a declared physical coordinate. Each map must support photon, neutron, and combined-field pages; linear and logarithmic colour scales; a fixed numeric range reusable across compared Cases; the corresponding relative-error map; and image plus tabular export. Every export must identify the Case, Result, quantity, unit, particle page, plane, slice coordinate, grid coordinates, colour range, and whether cells were scored. The existing sectioned 3D voxel view remains an exploratory display and is not the publication map.

D7 will select the actual profile lines, slice planes, coordinates, colour ranges, and reportable-RSE treatment from preliminary PHITS Results. These values must not be guessed during UI implementation. D8 verification must confirm slice indexing, orientation, values, units, and uncertainty alignment against the source VTK field before any map is used as paper evidence.

Training-data generation will not be hidden inside inference speedup. Timing comparisons will use repeated measurements after a predefined warm-up procedure where applicable and will report the summary statistic chosen before testing.

Uncertainty and applicability

The uncertainty discussion will distinguish:

Monte Carlo statistical uncertainty;
source, geometry, material, and operating-condition uncertainty;
transport-model, cross-section, and nuclear-data uncertainty;
measurement and detector-response uncertainty; and
surrogate approximation uncertainty relative to PHITS.

A surrogate interval calibrated only against PHITS labels will be described as surrogate uncertainty relative to the PHITS model, not total physical uncertainty.

The final range of applicability will state the approved accelerator type, source mode, electron-energy interval, converter version, material and geometry families, spatial domain, quantity, normalization, and known exclusions. Inputs outside that range will be identified as outside the model domain and will require a PHITS calculation and professional assessment.

Reproducibility and data traceability

All scientific work will be performed through the RAMAL-EBX GUI. Repository scripts are implementation backends, not an alternative research workflow. The Project will retain non-overwriting Design, Case, Job, Attempt, Result, Study, Assessment, Dataset, Model, Prediction, and Prediction Verification records.

For every reported result, the retained records will identify the software and PHITS versions, input, parameter values, source and material definitions, transport settings, random seed, output files, statistical uncertainties, Dataset membership, partition, Model definition, and evaluation record. Derived figures and tables will identify the source records from which they were made.

Decision justification requirement

Every scientific or methodological decision must be recorded with:

the exact decision and its owner;
the evidence supporting it;
the credible alternatives considered;
why those alternatives were rejected;
assumptions, applicability limits, and unresolved uncertainty;
the observable verification or acceptance criterion; and
the approval date.

Evidence may constrain a choice but does not make the choice automatically. Missing evidence remains an explicit uncertainty; the application must not invent scientific policy. A decision is complete only after its justification is recorded in the Project and approved by the project owner.

D1 — Paper 1 scope decision

Decision owner: project owner
Status: approved; the overall scope remains in force while D3/D5 are amended
Approval date: 2026-09-04T02:44:40Z

Approved decision

Industrial-representativeness amendment (2026-09-10). The project owner clarified that Paper 1 is intended to represent industrial radiation-processing X-ray systems. The circular 5 cm-radius, unscanned assembly below remains a Petwal component-benchmark definition, but it no longer defines the production source. The replacement under verification is a planar rectangular tantalum-water-steel converter normal to local +z, mapped to world -y in the primary D4 room, with a 100 cm x 10 cm active tantalum plate and a normalized one-dimensional scanned electron-beam distribution along the 100 cm direction. The local normally incident spot is circular Gaussian with 2 cm FWHM and declared 3 cm support. A uniform continuous time-integrated line scan, approximated with normalized trapezoidal quadrature, is the initial explicit reference assumption, not a universal scanner law. Converter construction and scan profile remain fixed source-family properties; electron energy remains a Case variable. Final coolant and cover dimensions remain unresolved: Petwal's 2 mm water plus 2 mm steel may be used only as a named preliminary transport assumption and must pass sensitivity checks before Dataset generation.

The historical decision follows for benchmark provenance.

Paper 1 will develop and independently evaluate a surrogate for one predefined three-dimensional radiation-protection field produced by verified PHITS reference calculations for conventional industrial electron LINACs operating in converted-X-ray mode from 5 to 7.5 MeV. The reference model will include the incident electrons, one explicitly specified finite converter family, one declared beam or scan distribution, and one parameterized family of closed shielding rooms. Photonuclear production will be assessed and photon and neutron contributions will remain separately inspectable. Model fitting, model selection, and final evaluation will use disjoint complete configurations. Surrogate predictions are for preliminary analysis inside the declared domain; selected predictions and final shielding decisions require PHITS confirmation and professional assessment.

Justification

Boundary	Supporting evidence	Alternative rejected and reason	Required check or limitation
Industrial converted-X-ray shielding	The owner identifies this as the immediate application need. Industrial X-ray processing and shielding are documented by Meissner et al., Cleland and Stichelbaut, and Peri and Orion. Converted X-rays provide greater penetration than direct electrons but add inefficient conversion and a demanding bremsstrahlung-shielding problem.	Product-dose prediction and medical radiotherapy answer different questions and use different quantities and source heads. Direct EB has distinct electron-range, scattering, beam-window, and interception physics and therefore requires its own verified mapping.	Claims remain limited to radiation protection around industrial converted-X-ray facilities; direct EB remains in the broader programme.
Conventional LINAC only	Conventional-LINAC source and shielding evidence exists, while the IAEA industrial-accelerator review confirms that materially different industrial architectures also exist.	Combining LINACs, Rhodotrons, dynamitrons, DC accelerators, and betatrons would mix source timing, beam delivery, and converter conditions without common verification.	The final applicability statement must name the accelerator architecture.
Coupled electron-to-converter source	Converter material, thickness, layers, beam footprint, and scan width change bremsstrahlung efficiency, spectrum, angular distribution, and residual electrons (Hua et al.; Lazurik et al.).	A nominal-energy monoenergetic photon source discards the source physics that the surrogate is intended to represent.	A reusable phase-space source is allowed only after comparison with the coupled calculation for the same converter and beam conditions.
`5–7.5 MeV` incident-electron interval	Direct evidence exists at `5`, `5.5`, `6`, and `7.5 MeV`, including industrial source, converter, and attenuation studies; the 1993 attenuation study reports measurements spanning `5.5–25.1 MeV` (DOI).	The broader `0.1–10 MeV` programme crosses different electron-range, converter-efficiency, shielding, photonuclear, and activation regimes and is not supported as one first-paper domain. Endpoint-only cases would not support an interval claim.	Include justified intermediate energies and hold out at least one complete intermediate-energy condition from model fitting.
One converter family	Converter-specific studies show dependence on construction and beam conditions (Tuan and Tao; Hua et al.), but no publication validates the final Paper 1 converter and room together.	Multiple unrelated converter families would enlarge the physical domain and confound source and shielding effects before one source model is verified.	D3 records every transport-relevant layer and fixes which dimensions and conditions vary.
Closed-room geometry family	Peri and Orion support industrial barrier calculations, while Cleland, Galloway, and Brown show that maze and opening transport depends on successive scattering and geometry.	Doors, mazes, ducts, and penetrations would require separately parameterized topology, sampling, verification, and failure analysis.	D4 records the approved complete walls, floor, roof, materials, densities, fixed source placement, and exterior evaluation regions. Streaming geometries remain out of domain.
One three-dimensional radiation-protection quantity	The research need is a spatial shielding field, and PHITS supports particle-resolved fluence and dose-conversion tallies.	Predicting several interchangeable dose quantities would obscure the target contract; product absorbed dose and activation are different studies.	D2 records the exact quantity, unit, coefficients, normalization, coordinate system, spatial representation, and particle combination.
PHITS reference with photonuclear sensitivity	The PHITS manual documents the required coupled transport and tally capabilities. IAEA SSG-8 and the Sari et al. benchmark show that neutron production cannot be dismissed from nominal energy alone and depends on converter composition, evaluated data, and reaction models.	A surrogate without verified Monte Carlo labels cannot establish physical correctness; photon-only labels could hide a converter-specific neutron contribution.	D6–D9 must define code/data sensitivity, component verification, measurement applicability, and the neutron reporting rule.
Complete-configuration final testing	Geometry and source conditions are shared by all voxels from one calculation. Splitting voxels across training and test sets would leak configuration information and exaggerate generalization.	Random voxel-level partitioning does not test prediction for a new room configuration.	D10 must freeze grouped partitions before fitting; the final-test configurations remain unopened until model selection is complete.
Preliminary-use claim	Both the owner intent and evidence package require PHITS confirmation, professional assessment, commissioning evidence, and regulatory review for final shielding decisions.	Treating surrogate output as final design evidence would exceed its validation and applicability.	The GUI and paper must label out-of-domain inputs and require PHITS confirmation of selected predictions.

Assumptions and unresolved decisions

Institutional Scopus, Web of Science, and INSPEC coverage is unavailable, so the evidence supports a bounded contribution claim rather than absolute novelty.
Four evidence records remain awaiting-full-text and cannot support stronger claims until resolved.
D1 did not choose the dose quantity, converter construction, room bounds, scan distribution, photonuclear settings, numerical acceptance thresholds, Dataset partition, or surrogate metric. D2 and D4 approve the field and room geometry; D3 and D5 are amendment-open for the industrial scanned-plate source; D6 fixes photonuclear comparisons and neutron retention; D7–D11 retain the remaining decisions.
A physical-validation claim is permitted only for components or configurations matched to applicable measurements and their uncertainties.

D1 acceptance record

D1 is complete: the project owner approved the exact decision on 2026-09-04T02:44:40Z; the scope matches the Research Plan and Evidence Review; and values owned by the other decisions were not silently promoted into D1.

D2 — Radiation-protection field decision

Decision owner: project owner
Status: approved
Approval date: 2026-09-04T03:18:08Z

Approved decision

The primary Paper 1 target is the three-dimensional ambient dose equivalent $H^*(10)$ field:

Contract item	Approved value
Target ID	`ambient-dose-equivalent-h10`
Quantity	Ambient dose equivalent, $H^*(10)$
PHITS coefficients	Built-in multiplier `-200`; coefficient data and PHITS version retained with every Result
Intrinsic unit	`pSv/primary-source-electron`
Intrinsic normalization	Per primary electron launched at the D3 source plane
Transport particles	Photon, electron, positron, and neutron
Primary field	Predefined sum of separately converted particle contributions
Coordinate system	Design right-handed Cartesian `x,y,z`, in `cm`
Spatial representation	Rectilinear voxel cells; values are cell quantities, not point interpolation
Derived operational unit	`mSv/h`, calculated from the intrinsic field and recorded operating conditions
Secondary quantity	ICRU Report 95 ambient dose, $H^*$, using PHITS `-204`, only as a separately named comparison

Justification

Choice	Supporting evidence	Alternative rejected and reason	Required check or limitation
$H^*(10)$ primary	ICRU Report 95 states that the existing ambient-dose-equivalent quantities remain established operational quantities and that adoption of the replacement quantities requires a gradual, prudent transition. $H^*(10)$ also preserves comparability with existing survey instruments and historical shielding evidence.	Making the newer $H^*$ primary now would reduce direct comparability with legacy measurements and current practice before facility-specific instrument and regulatory adoption is established.	$H^*(10)$ is not claimed to be explicitly mandated by Malaysian P.U. (A) 46. A licence condition or facility measurement programme overrides this research convention where applicable.
PHITS `-200` coefficients	The PHITS 3.37 multiplier manual identifies `-200` as $H^*(10)$ and states that its radiation-dose coefficients use `pSv cm²`.	Custom coefficients would add an unnecessary unverified data path.	Retain the actual multiplier file, hash, PHITS version, interpolation behaviour, particle applicability, and energy bounds with each Result.
Per-electron intrinsic field	Linear particle transport permits source-strength scaling while geometry and material state remain unchanged. Separating the intrinsic field from beam operation prevents the surrogate from learning a known linear current conversion.	Training directly on one machine's `mSv/h` embeds current and schedule assumptions and makes physically identical fields appear different.	Linearity must not be assumed if material state, source distribution, or other physics changes with operating intensity.
Separate particle contributions	The PHITS multiplier mechanism supports particle-selective conversion. IAEA guidance and photonuclear evidence require neutron assessment in the selected energy domain.	A combined-only output could hide a small but uncertain neutron component or a particle-definition error.	Preserve photon, electron, positron, and neutron fields and uncertainties before calculating the declared sum. D6 defines the neutron rule.
$H^*$ secondary only	ICRU Report 95 recommends $H^*$ as a better estimator of protection quantities over broader particle and energy ranges; PHITS provides it as `-204`.	Mixing $H^(10)$ and $H^$ in one target would create a quantity with no valid interpretation.	Any $H^*$ output is a separate Result/field with its own identifier and coefficients; it is not required for the primary training Dataset unless a later justified amendment says otherwise.
Derived `mSv/h`	Malaysian P.U. (A) 46, regulation 21 requires workplace measurements of external radiation levels and documented monitoring methods, but does not name $H^*(10)$ in the reviewed text. Operational rate is therefore useful for comparison but must retain its scenario assumptions.	Treating a rate as intrinsic transport output conflates physics with current, pulsing, scan utilization, workload, and schedule.	Report every scaling input and compare only with measurements using a compatible quantity, calibration, averaging period, location, and uncertainty.
Cartesian voxel contract	PHITS `[T-Track]` supports an explicit `xyz` mesh, and RAMAL-EBX binds Results to their immutable grid and coordinate arrays.	Unlabelled points, arbitrary interpolation, or mixed grids would make Dataset labels ambiguous.	D4 sets bounds; D7 sets resolution, convergence, and reportable coverage. Changing either creates a different Study.

Assumptions and unresolved decisions

The primary field is a calculation quantity for area shielding assessment, not personal dose equivalent and not an individual's effective dose.
D4 fixes the spatial domain; D7 retains grid resolution and convergence.
D6 fixes the photonuclear comparison strategy and neutron retention; installed isotope coverage and D7/D8 verification remain prerequisites.
Measurement comparison is valid only when the instrument quantity, calibration field, energy and angular response, location, averaging period, and uncertainty are sufficiently matched.
If the applicable Malaysian licence conditions or approved monitoring procedure specify another quantity, D2 must be amended before Dataset generation.

Implementation boundary

RAMAL-EBX already represents the -200 target, particle list, intrinsic unit, reported unit, coordinate grid, and normalization metadata. Its current tally generation scales the PHITS field directly to mSv/h from beam current. That does not yet satisfy the approved primary-Dataset contract: the application must retain the per-electron field as the immutable primary Result and derive operational rate separately before Paper 1 reference-Dataset generation.

D2 acceptance record

D2 is complete: the owner approved the decision on 2026-09-04T03:18:08Z; the quantity, coefficients, units, normalization, particles, combination rule, coordinate convention, and secondary quantity are explicit; D4 fixes the spatial domain, D6 fixes photonuclear comparisons and neutron retention, and D7 retains grid/statistical decisions.

D3 — Converter and incident-electron decision

Decision owner: project owner
Status: approved
Approval date: 2026-09-04T03:39:33Z

Approved decision

Paper 1 will use one fixed planar converter assembly based on the RRCAT industrial 5/7.5 MeV study:

primary-electron source plane
       ↓ local +z, normal incidence
1.2 mm natural tantalum
2.0 mm liquid-water cooling layer
2.0 mm Type 304 stainless-steel downstream layer
            ↓ converted field enters room

All three layers are coaxial disks of 5 cm radius. The primary source plane is coincident with the upstream tantalum face. Each source history begins as one electron at that plane. The local beam footprint is a centred circular Gaussian with $\sigma=6.25$ mm, truncated at radius 12.5 mm (25 mm diameter, ±2σ), and normal incidence.

Each Case uses a declared monoenergetic primary-electron energy $E_e\in[5,7.5]$ MeV. The endpoints and at least one intermediate condition must be verified; Dataset sampling and the untouched intermediate-energy test remain owned by D10. D5 fixes the footprint as static and centred with no spatial translation.

Material definitions use natural tantalum at 16.654 g/cm³, liquid water at 0.997 g/cm³, and the PNNL-15870 Rev. 2 Type 304 stainless-steel composition at 8.03 g/cm³. The exact elemental or isotopic records and source publication identifiers are stored with every Design and Case.

Justification

Choice	Supporting evidence	Alternative rejected and reason	Required check or limitation
Tantalum converter	Petwal et al. selected tantalum as a compromise among photon yield, residual radioactivity, physical and mechanical properties, fabrication, and cost for a `5/7.5 MeV`, `10 kW` industrial LINAC. Peri and Orion also used tantalum for the applicable concrete-attenuation calculations.	Gold offers little shielding-field benefit for its cost; tungsten has similar photon output but the cited industrial design identifies brittleness and fabrication disadvantages.	This is a research reference assembly, not a certified thermal or mechanical design. D6 must assess tantalum photonuclear behaviour near `7.5 MeV`.
Fixed `1.2 mm` tantalum	Petwal et al. report photon energy transmitted after tantalum of `9.76%` at `5 MeV` versus the `10.14%` maximum at `0.9 mm`, and `15.20%` at `7.5 MeV` versus the `15.24%` maximum at `1.4 mm`. Thus `1.2 mm` retains approximately `96.3%` and `99.7%` of the respective reported maxima while keeping one physical converter across the interval.	Energy-specific `0.9/1.4 mm` optima would change converter construction with energy and confound the effect Paper 1 intends to learn. Making thickness a predictor would expand the first Dataset before one assembly is verified.	Reproduce the cited thickness curves in PHITS and test neighbouring thicknesses. The fixed choice is rejected if PHITS materially changes the spectral, angular, residual-electron, or neutron field relative to the cited behaviour.
`2 mm` water plus `2 mm` stainless steel	Petwal et al. explicitly model these layers for both energies: water removes heat and stainless steel forms the channel and reduces residual-electron contamination. Their results also show that the layers absorb photon output, so they are transport-relevant and cannot be omitted.	A bare tantalum slab would exaggerate the emerging field and would not represent the cited high-power assembly.	Preserve downstream electrons and photons rather than assuming complete electron removal. Thermal-hydraulic adequacy is outside Paper 1 and must not be claimed.
`5 cm` disk radius	This is the finite radius used by Petwal et al. and is large relative to their `12.5 mm` beam truncation radius.	An infinite slab removes edge escape; a smaller disk introduces an unsupported edge regime.	D5 keeps the footprint centred; any future scan family requires separate converter-edge verification.
Gaussian `25 mm` footprint	Petwal et al. use a normally incident circular beam with `25 mm` diameter (`±2σ`) and $\sigma=6.25$ mm. Reusing the published distribution gives a traceable source rather than an invented spot.	A pencil beam simplifies transport but changes local heating, lateral electron transport, and near-field emission. An arbitrary uniform disk lacks direct support.	Verify the implemented radial sampler statistically against the declared truncated Gaussian.
Monoenergetic $E_e$ parameter	Petwal et al., Peri and Orion, and the accessible converter literature model declared monoenergetic incident energies, enabling direct endpoint and intermediate-energy comparisons. No measured energy spectrum for the eventual Paper 1 machine is available.	Inventing a Gaussian energy spread would add an unsupported width and tail. Treating only `5` and `7.5 MeV` would not support interpolation across an interval.	Applicability is to the declared monoenergetic computational source, not every installed LINAC. If a validation machine is selected, its measured energy distribution must be tested and may require a new Study or justified D3 amendment.
Source plane at tantalum face	The cited converter calculation defines the incident beam at the target. This makes one primary source electron unambiguous and preserves the D2 normalization through all converter descendants.	Starting with photons downstream would discard converter yield, angular distribution, residual electrons, and photonuclear production. Modelling an unspecified accelerator exit window would invent machine geometry.	Any real upstream window, air gap, or beamline component lies outside this source boundary until specified and tested. A phase-space source must normalize back to the original primary-electron count.
PNNL material records	PNNL-15870 Rev. 2 provides traceable radiation-transport compositions and densities for tantalum, liquid water, and Type 304 stainless steel.	Unnamed “stainless steel,” default water, or software-library material aliases are not reproducible scientific definitions.	Actual fabrication certificates supersede these representative materials for facility-specific validation. D6 owns isotope/data sensitivity.

Required verification and limitations

Before complete-room Dataset generation:

reproduce Petwal et al.'s emerging photon-energy fractions versus tantalum thickness at 5 and 7.5 MeV, including the 0.9, 1.2, and 1.4 mm cases;
compare photon spectrum, angular distribution, residual-electron energy, and particle-resolved fluence immediately after each layer;
verify the truncated-Gaussian radial sampler and normal incidence;
compare applicable 5, 6, and 7.5 MeV source or attenuation benchmarks, without treating agreement in one observable as validation of the assembly;
perform the D6 photonuclear data/model sensitivity for natural tantalum, stainless steel, and water; and
retain PHITS version, electromagnetic and nuclear data, material records, cutoffs, source weights, geometry, random seed, and statistical uncertainty.

Paper 1 does not claim that this converter is thermally optimized, manufacturable, licensed, or representative of every commercial LINAC. It is a fixed, literature-based transport reference whose range of applicability is the declared source boundary and material definition.

D3 historical acceptance record

D3 was approved for the compact component model: the owner approved the fixed 1.2 mm tantalum assembly and incident-electron definition on 2026-09-04T03:39:33Z. That acceptance is superseded by the owner direction of 2026-09-10; its calculations may be retained only as named Petwal component benchmarks and are excluded from D7 and the production Dataset. Revised D3 acceptance requires plate-size, finite-edge, coolant/cover-thickness, scan-distribution, source-field, and D6 checks.

D4 — Closed-room geometry and material decision

Decision owner: project owner
Status: approved
Approval date: 2026-09-04T04:00:51Z; orientation amended 2026-09-11

Approved decision

Paper 1 will use one axis-aligned rectangular closed-room family. In the Design coordinate system, x is the scan/lateral axis, +y is vertical/up, and z is the second horizontal axis associated with product or conveyor travel. The D3 converter emits vertically downward along -y, so its Gaussian footprint and converter face lie in the world x-z plane. The clear interior is:

$$ -W/2 \le x \le W/2,\qquad 0 \le y \le H,\qquad -L/2 \le z \le L/2. $$

The extended converter is centred at $(0,H/2,0)$ and its downstream normal is -y. Its PHITS s-type = 13 source remains local X-Y with local +Z direction and is transformed as one source definition into the room frame. Its plate dimensions, scan direction, and scan-position density are fixed source-family properties rather than first-paper surrogate predictors. Every room-boundary check must use distances from the full finite scanned source, not only its centre.

Geometry parameter	Approved domain	Role
Interior width, $W$	`4–10 m`	Equal `x` clearances of `2–5 m`
Interior length, $L$	`6–16 m`	Equal `z` clearances of `3–8 m`
Interior height, $H$	`3–6 m`	Equal source-to-floor and source-to-roof clearances of `1.5–3 m`
Primary-floor thickness, $t_p$	`100–180 cm`	Floor intersected by the `-y` converter axis
Uniform secondary-barrier thickness, $t_s$	`50–150 cm`	Roof and four side walls

The shielding is one concrete shell: the outer box

$$ [-W/2-t_s,W/2+t_s]\times[-t_p,H+t_s]\times [-L/2-t_s,L/2+t_s] $$

minus the clear interior. This constructive definition fills edges and corners without gaps or overlapping cells. Concrete at y ≤ 0 is the primary-floor region; the remainder is secondary barrier. There are no doors, mazes, ducts, penetrations, joints, voids, reinforcement, embedded services, or streaming paths. The complete converter is inside the room; no separate accelerator head, product, conveyor, support, beam stop, or other nearby scattering structure is included in this first family.

All shielding uses PNNL-15870 Rev. 2 material 105, Concrete, Ordinary (NBS 03), at 2.35 g/cm³. Its elemental or isotopic weight fractions are copied exactly into each Design rather than referenced through an unstable software alias. Interior and exterior air use PNNL material 4, Air (dry, near sea level), at 0.001205 g/cm³. The D3 converter materials remain unchanged.

The computational universe contains a 100 cm air margin beyond every outer shield surface. The primary reportable regions are the interior air and the entire exterior air shell. Exterior cells retain their nearest barrier face; edge and corner cells remain explicit rather than being duplicated between face regions. Reference surfaces 30 cm outside each barrier are retained for directional comparisons; D7 will choose the voxel grid and reportable-coverage criteria. Air below the floor is a leakage diagnostic, not a claim that a real facility has occupied air beneath its foundation; in the vertical family it is the exterior region downstream of the primary barrier.

Justification

Choice	Supporting evidence or design reason	Alternative rejected and reason	Required check or limitation
One rectangular closed room	The owner requires the first family to include complete walls, floor, roof, primary and secondary barriers, and exterior evaluation regions while excluding streaming openings. A rectangular box is the smallest topology satisfying that requirement.	A slab or spherical shell is easier but cannot represent corners, distinct barrier roles, or a room field. A maze or penetrated bunker adds a separate streaming problem before the closed-room model is verified.	Applicability is limited to this box topology; no result applies to openings or other room families.
Five geometry predictors	Width, length, height, and two barrier thicknesses vary the main distance, angular, attenuation, and room-scatter effects without multiplying weakly supported construction variables. In the vertical family, height controls the primary source-to-floor distance while width and length control transverse boundaries. The numerical room bounds are an owner-approved study-design domain, not values inferred from a particular facility.	Six independent surface thicknesses, arbitrary source coordinates, and free rotations would greatly enlarge the Dataset before any complete-room calculation establishes tractability. One fixed room would not support a geometry surrogate.	Preliminary calculations must confirm useful response variation and feasible statistics across every boundary and interaction; otherwise D4 is amended before Dataset generation.
Fixed centred vertical source placement	The D3 field is strongly forward-biased. A centred source keeps source-to-floor and source-to-roof distances equal while the primary and secondary barrier thicknesses remain distinct, isolating orientation and attenuation from an additional source-height predictor.	Freely moving or rotating the source introduces additional interactions not required for the first family; an arbitrary offset would add a confound without facility evidence. A horizontal source is supported by RAMAL-EBX but is not mixed into the primary vertical Paper 1 Dataset.	Paper 1 cannot claim prediction for other source placements or orientations. A later separately verified geometry family may use horizontal or offset delivery.
`100–180 cm` primary and `50–150 cm` secondary concrete	Peri and Orion calculate ordinary-concrete attenuation from `10` to `180 cm` at `5` and `7.5 MeV` and show strong energy-angle dependence. HJ 979-2018 tabulates broad-beam concrete TVLs of `32.5 cm` at `5 MeV` and `36.8 cm` at `7.5 MeV`; the candidate domains therefore span multiple attenuation decades while remaining inside the published Monte Carlo thickness range.	Treating one wall thickness as safe would require beam current, workload, occupancy, jurisdiction, and facility details that D4 does not possess. Thicknesses below `50 cm` add many obviously weak barriers without strengthening the intended deep-shield surrogate problem.	These are simulation bounds, not safe or compliant construction dimensions. HJ 979 applies its X-ray requirements only through `5 MeV` and is used here as a transport comparator, not Malaysian regulatory authority.
One ordinary-concrete record	PNNL-15870 Rev. 2 gives material 105 a complete transport composition at `2.35 g/cm³`. Holding composition and density fixed isolates geometric effects.	A generic “concrete” label is irreproducible. Varying density independently while retaining one composition would imply unsupported physical concretes and expand the first Dataset.	Results do not apply to as-built concrete, reinforcement, moisture states, aggregates, cracks, joints, or voids. Actual facility data require a new material definition and validation.
Dry-air interior and exterior	PNNL material 4 supplies a traceable composition and density. Using one air record avoids an unsupported environmental parameter.	Vacuum removes room and exterior scatter; variable temperature, pressure, and humidity are not part of the first research question.	Applicability is to the declared reference atmosphere. Environmental sensitivity may be added only if preliminary transport shows material effect.
`100 cm` exterior margin and `30 cm` references	The owner requires exterior fields. HJ 979 evaluates accessible areas at `30 cm` or more from shielding surfaces, while a full metre preserves spatial falloff rather than one control point.	A surface-only tally cannot train the approved three-dimensional field; an unbounded exterior domain adds cost without a defined use.	D7 owns voxel size and statistical criteria. Outer-boundary and air-margin convergence must show that truncation does not materially alter reportable cells.
Empty-room baseline	No specific product, conveyor, beam stop, support, or accelerator-head geometry is available. Inventing one would turn an unsupported facility example into application scope.	Omitting structures silently would understate their possible scattering effect; hard-coding a speculative proxy is not more scientific.	Before complete-room Dataset generation, D8 must test credible bounding structures one at a time. If any changes the D2 field beyond the D7 materiality criterion, D4 must be amended to include or explicitly parameterize it.

Required verification and limitations

Before complete-room Dataset generation:

reconstruct every inner and outer surface analytically and prove positive volumes, exact wall contacts, and no unintended gaps or overlaps at all parameter bounds;
reproduce selected Peri–Orion ordinary-concrete attenuation cases before relying on the room calculations;
compare simple inverse-square and slab-attenuation limits where their assumptions hold;
test source-to-wall distances and the primary/secondary barrier assignment at every domain boundary;
establish grid and outer-air-margin convergence under D7;
test omitted nearby structures under D8 and amend D4 if the predefined materiality criterion is exceeded; and
preserve exact material fractions, densities, geometry parameters, region identities, and source-to-surface distances in every Design and Case.

Paper 1 will not describe any sampled geometry as safe, compliant, buildable, or representative of all industrial irradiation facilities. D4 defines a bounded computational geometry family for method development and evaluation.

D4 acceptance record

D4 is complete: the owner approved the topology, parameter domain, fixed source placement, materials, and evaluation regions on 2026-09-04T04:00:51Z and amended the primary orientation on 2026-09-11 to +y up with the beam along -y, scan along x, and the primary barrier at the floor. Horizontal delivery remains an application-supported alternative outside this primary family. Verification remains mandatory; D5 retains the scanned-source amendment, and D7 and D8 retain grid/statistical and benchmark details.

D5 — Beam-scan decision

Decision owner: project owner
Status: amendment open; the 2026-09-04 static-source decision is superseded
Approval date: 2026-09-04T08:54:45Z

Approved decision

Industrial-representativeness amendment (2026-09-10). Paper 1 now uses a normalized one-dimensional time-integrated scan-position distribution over the broad D3 plate. The local Gaussian spot is sampled at each scan position. The initial reference density is a uniform continuous line scan approximated by normalized trapezoidal quadrature and must be labelled as a study-design assumption; measured or manufacturer-supplied current-versus-position weighting supersedes it when available. Frequency, current, duty cycle, and workload are recorded as operational provenance and used only for derived dose rates; they do not alter the per-primary transport field when the normalized spatial distribution is fixed. The historical static definition below remains only as a comparison condition.

The historical decision follows for provenance.

Revised scan representation and precedent

PHITS does not require the scanning magnet trajectory to be transported for a steady, time-integrated shielding field. RAMAL-EBX represents the declared scan-centre probability density as a weighted mixture of ordinary PHITS Gaussian sources. For an inclusive uniform interval, N odd equally spaced centres use normalized trapezoidal weights: each endpoint has $1/[2(N-1)]$, each interior node has $1/(N-1)$, and the weights sum to one. The generated PHITS [Source] card uses s-type = 13, one <source> = weight subsection per centre, and totfact = 1, following the official PHITS multi-source normalization mechanism [8]. This is a RAMAL-EBX quadrature construction; no reviewed paper is claimed to use this exact PHITS card.

The closest directly relevant industrial precedent is Ziaie and Tahami's EGS4 study of a Rhodotron X-ray facility [42]. Their measured system used a 100 cm scanned electron beam and a tantalum-water-stainless converter, and their dynamic calculation integrated stationary dose distributions over spatial intervals. That establishes the same linear-superposition principle for an industrial scanned electron/X-ray system, although it is EGS4 rather than PHITS and includes product motion. Ma et al. independently showed that scanned electron beams can be represented by spatial and multiple-source Monte Carlo models, with comparison against full phase-space calculations [43]. These references justify the numerical strategy, not the Paper 1 node count or scan density; those remain subject to normalization, quadrature-convergence, finite-edge, and static-versus-scanned verification.

Historical superseded static-source decision

Paper 1 previously excluded beam scanning. Every primary history used the D3-approved circular Gaussian in the converter's local x-y plane:

$$ \Delta x_{\mathrm{scan}}=\Delta y_{\mathrm{scan}}=0. $$

The footprint centre was at the D4 room centre $(0,H/2,0)$, its axis was +z, and its radial distribution was $\sigma=6.25$ mm truncated at 12.5 mm. The source condition identifier is static-centered. Scan amplitude, waveform, frequency, phase, dwell, speed, utilization, and timing are absent, not zero-valued surrogate predictors.

Justification

Choice	Supporting evidence or design reason	Alternative rejected and reason	Required check or limitation
Static centred footprint	D3 already adopts the finite normally incident Gaussian used by Petwal et al.. D4 fixes the converter at the room centre. Keeping this source unchanged isolates incident energy and the five approved room variables.	Adding a scan without a selected scanner would require invented axes, amplitude, waveform, dwell, frequency, and converter-edge rules.	Paper 1 applies only to the static source; it cannot claim prediction for scanned industrial systems.
No scan predictor or placeholder fields	A constant input contains no learnable information. Omitting it keeps the Case contract smaller and prevents a future nonzero scan from being mistaken for an in-domain value.	Retaining `scan_amplitude=0` and other dormant fields would preserve speculative schema without scientific use.	Every Case still records the explicit source condition ID so static and future scanned Studies cannot be mixed.
Scanning deferred to a separate source family	Hua et al. report dependence on beam sweep width. Lazurik et al., Salehi et al., and Zhang et al. treat scan delivery as hardware- and facility-specific.	Treating all scan systems as a generic translated Gaussian would erase time weighting, path, edge interception, and converter thermal constraints.	A future scanned family requires its own evidence, exact spatial/time probability distribution, converter-edge verification, and full source-to-room comparison.
Operational utilization remains derived	D2 separates intrinsic transport per primary source electron from current, workload, and schedule. With no spatial scan, there is no scan-duty correction inside the transport source.	Folding an assumed utilization factor into PHITS would obscure the approved per-primary normalization.	Any reported `mSv/h` scenario must still document beam current, pulsing, exposure schedule, and other operational factors.

Required verification and limitations

The superseded static-source verification requirements were:

prove that the source generator applies no history-dependent or time-dependent footprint translation;
verify zero sample centroid within the predefined statistical tolerance and reproduce the D3 truncated-Gaussian radial distribution;
verify the converter axis and source centre against the D4 coordinate contract in every generated Case;
reject any Case containing a non-static scan condition from the Paper 1 Study; and
compare the implemented source with the coupled D3 converter calculation before any phase-space reuse.

D5 historical acceptance record

D5 was approved for the compact static source on 2026-09-04T08:54:45Z and was superseded by the owner direction of 2026-09-10. Revised acceptance requires GUI-visible scan authoring and preview, sampler and normalization tests, finite-edge interception checks, and static-versus-scanned component comparisons.

D6 — Photonuclear comparison and neutron-reporting decision

Decision owner: project owner
Status: approved
Approval date: 2026-09-05T04:44:04Z

Approved decision

Keep photonuclear production and neutron transport enabled throughout the 5–7.5 MeV domain. Retain the neutron contribution in the primary D2 combined field; Paper 1 will not use a small-fraction rule to omit it. A separately named photon-only field may be inspected but cannot replace the combined target. Unscored neutron tallies remain unresolved, not evidence of zero production.

Use JENDL-5 photonuclear data as the production reference, conditional on isotope coverage and component verification. Compare it with ENDF/B-VIII.1 photonuclear data and the PHITS built-in photonuclear reaction models. Keep these comparison Results separate from the production Dataset; do not average libraries or silently mix their fields.

Controls and comparison contract

The installed PHITS manual distinguishes dpnmax (photonuclear library upper energy) from dmax(14) (photo-atomic library upper energy). It documents lib(14)=20u for JENDL-5 and automatic model substitution when non-neutron libraries are absent. Therefore input labels alone cannot prove library use.

Calculation	Proposed configuration	Purpose and check
JENDL-5 reference	`ipnint=1`, `lib(14)=20u`, `dpnmax=20 MeV`	Covers the Paper 1 photon-energy interval without a switch at its upper endpoint. Audit resolved isotope tables and their energy coverage.
ENDF/B-VIII.1 comparison	Same transport settings; explicitly registered ENDF/B-VIII.1 photonuclear tables and `dpnmax=20 MeV`	Change production data alone. Record actual processed-table suffixes and checksums; do not invent a PHITS library identifier before installation.
Built-in model comparison	`ipnint=1`, `dpnmax=0`	Isolate evaluated-data versus model behaviour; retain exact PHITS version and effective model settings. This is a sensitivity comparator, not a fallback production reference.
Resonance-fluorescence sensitivity	Paired reference with `ipnint=2`	`ipnint=1` excludes nuclear resonance fluorescence. Test its effect on the photon and combined fields under D7/D8 before accepting that exclusion.

Start component comparisons with pnimul=1. Any later photonuclear sampling enhancement must preserve particle weights and pass an analogue comparison under D7; it is variance reduction, not a physical yield multiplier.

Hold electromagnetic transport, material isotope fractions, neutron transport tables, thermal-scattering treatment, cutoffs, geometry, tally definitions, and normalization fixed across each production-data comparison. Record them explicitly. Neutron-library or cutoff sensitivity is a separate comparison; changing several mechanisms together cannot isolate photonuclear-data effects.

Comparisons must cover:

isotope-resolved reaction thresholds and cross sections for every constituent of the D3 converter and D4 materials, including low-abundance isotopes;
component photon-induced neutron currents, energy spectra, and angular distributions where applicable benchmark definitions are recoverable;
the coupled D3 converter at 5, 7.5 MeV, and justified intermediate and near-threshold conditions; and
neutron, photon, and combined D2 fields in interior and exterior regions of representative and boundary D4 rooms, not merely total source yield.

D8 owns exact benchmark geometries and reference values; D7 owns numerical resolution, convergence, reporting coverage, and materiality criteria. These must be fixed before accepting the comparisons. Disagreement is reported as model/data sensitivity, not interpreted automatically as a confidence interval. If the proposed reference fails these checks, amend D6 before Dataset generation.

Evidence, alternatives, and limitations

Garnaud et al. (2026) explicitly compare PHITS, MCNP6, and TRIPOLI-4 with ENDF/B-VIII.1 and JENDL-5 across 49 elemental photon targets from threshold to 30 MeV. Their abstract supports the selected library pair and observables, not numerical acceptance limits or validation of this room. The retained Sari et al. (2025) benchmark record supplies the complementary electron-accelerator comparison direction; exact cases require accessible numerical definitions under D8. Agreement between codes is verification evidence, not measurement validation.

Choosing JENDL-5 as the reference is a study-design choice supported by explicit PHITS integration, not a claim that it is universally more accurate. A model-only reference would avoid data installation but weaken the intended near-threshold comparison. Excluding neutrons below an arbitrary percentage would add an unsupported scientific threshold and could hide a locally important exterior contribution. Keeping them avoids that additional policy.

The local installation inspected for this proposal has no photonuclear u tables registered in phits/data/xsdir.jnd; its XS/jendl5 directory contains neutron, proton, and deuteron data, not the photonuclear collection. The manual explicitly requires a separate photonuclear-data installation. No libraries have been installed or transport comparisons run as part of this decision. Acquiring and processing the two evaluated collections, auditing every isotope, and recording unresolved coverage are prerequisites to reference calculations. Missing coverage must trigger an explicit D6 amendment rather than silent model substitution or removal of the affected isotope.

Activation and post-shutdown dose remain outside Paper 1. Poorly converged neutron contributions cannot be rescued by relabelling the target photon-only; their inclusion in a combined field must satisfy the D7 uncertainty and coverage contract.

D6 acceptance record

The owner approved this reference/comparison strategy and no-neutron-omission reporting rule on 2026-09-05T04:44:04Z. D6 is complete as a decision, not as evidence of installed data, completed benchmarks, or physical validation.

Pre-study decision status

ID	Decision or status	Why it blocks execution
D1	Approved `2026-09-04` — use the scope recorded above	Complete; establishes the bounded Paper 1 research domain
D2	Approved `2026-09-04` — use the field contract recorded above	Complete; defines the PHITS tally and primary Dataset target
D3	Amendment open `2026-09-10` — replace the compact circular production source with the industrial scanned-plate archetype	Plate, coolant/cover, finite-edge, and component verification remain required
D4	Approved `2026-09-04` — use the geometry and material contract recorded above	Complete; defines the bounded closed-room physical domain
D5	Amendment open `2026-09-10` — define and verify the normalized one-dimensional scan distribution	GUI scan authoring, preview, sampler, and normalization evidence are required
D6	Approved `2026-09-05` — conditional JENDL-5 reference, data/model comparisons, and neutron retention	Complete decision; data installation and verification remain required
D7	Set grid, statistical, reportable-coverage, repeated-run, completion, and 1D/2D reporting criteria from preliminary PHITS calculations	Defines which Results may enter the Dataset and how comparable profiles/maps are reported
D8	Define exact verification geometries, reference values, and comparison criteria	Defines the evidence needed before complete-room calculations
D9	Identify applicable measurement evidence and its uncertainty	Determines which physical-validation claims are possible
D10	Predefine Dataset grouping and training, model-selection, and final-test partitions	Prevents information leakage and post hoc partition changes
D11	Predefine surrogate metrics and acceptable performance by subdomain	Prevents average performance from hiding unsafe local failures

Until D7–D11 are approved, their numerical thresholds and detailed parameter ranges will remain unresolved rather than being invented by the application or this document.

Relationship to the evidence package

The Evidence Review Protocol defines the search and screening method, and the Evidence Review records the accessible-source results and the scope recommendation accepted in D1. Citation numbers follow the Literature Review, which states the support and limitations for each methodological choice. Source-level notes are maintained in the Annotated Bibliography. D2–D6 record the approved field, source, geometry, scan, and photonuclear contracts. The Research Plan defines the current research questions and scope. The project owner's paper1.md remains authoritative for research intent; differences between the broad programme and the narrower evidence-supported Paper 1 scope are reported rather than hidden.