Skip to content

Evidence Review

Review boundary

The approved original-source bundle contains exactly two papers: Peri and Orion (2017), six pages, and Chowdhury et al. (2024), three pages. Claims below come only from those PDFs and their page-separated local text.

UNRESOLVED: The programme calls for a systematic evidence-review protocol. Earlier derived prose reported a search and screening exercise, but derived prose is not scientific confirmation and the supplied bundle does not verify its counts or completeness. A systematic review has therefore not been completed.

APPROVED DECISION: Treat the two-paper review as a supplied-bundle baseline. Before defining D3/D5 candidates, conduct a broad cited primary-source review using scholarly databases, technically usable industrial and manufacturer sources, and backward and forward citation searches. Stop after two successive search rounds identify no new candidate family or design constraint.

Literature synthesis

EXTERNAL LITERATURE: Peri and Orion examine 5 and 7.5 MeV pencil electron beams incident on high-Z converters using MCNP 4C2. They calculate angular photon dose, bremsstrahlung spectra, and ordinary-concrete attenuation, and report dependence on energy, target, emission angle, and shielding thickness [Peri and Orion 2017, pp. 1-6]. This supports conditioning Paper 1 calculations on energy and geometry orientation and may inform later D8 comparison definitions.

Their isolated targets include 1.5 mm Ta for 5 MeV and 2.3 mm Ta for 7.5 MeV; the authors selected target dimensions in relation to electron range, not as universal industrial scanned-head optima [Peri and Orion 2017, p. 2]. Their model uses pencil beams, angular detector sectors, and spherical concrete shells, not a finite scanned head or complete rectangular industrial room [Peri and Orion 2017, pp. 2-3]. Their photon dose uses the F6:p tally and a KERMA approximation, not the approved H*(10) target [Peri and Orion 2017, p. 2].

Peri and Orion report angular dose-rate calculations using 10^6-10^7 histories and spectral calculations using 10^7-10^8 histories, generally targeting their stated relative statistical uncertainties; thick concrete required geometry-splitting variance reduction [Peri and Orion 2017, pp. 3-5]. These are attributes of their configurations, not Paper 1 history, variance-reduction, uncertainty, or acceptance criteria.

They compare selected MCNP results with earlier calculations and a measured 2.8 MeV iron-target spectrum [Peri and Orion 2017, pp. 2-4]. Those comparisons are limited checks of their systems. The paper does not verify D2-D6, PHITS, RAMAL-EBX, a complete Paper 1 room, or a Paper 1 surrogate.

EXTERNAL LITERATURE: Chowdhury et al. use PHITS-generated responses to train a one-dimensional CNN predicting downstream neutron differential flux through concrete. They calculate 250 quasi-monoenergetic neutron responses from 1-250 MeV, construct 10,000 superposed datasets, report a 70:30 training/testing split, and test a realistic Ca-48-derived spectrum distinct from their training spectra [Chowdhury et al. 2024, pp. 1-3]. They report approximately 10% effective-dose agreement in their demonstration and millisecond-scale model calculation [Chowdhury et al. 2024, pp. 1, 3]. This supports the technical plausibility of a PHITS-trained surrogate in a different shielding problem.

That study concerns high-energy neutron spectra, simplified slab shielding, and one-dimensional spectral output. The test retained the simulation geometry used for training-data generation, although its realistic input spectrum differed from the training spectra [Chowdhury et al. 2024, pp. 2-3]. It does not establish transferability to electron-to-bremsstrahlung conversion, complete three-dimensional rooms, H*(10), whole-configuration final-test governance, photonuclear sensitivity, voxel reportability, or local underprediction control.

Applicability

  • EXTERNAL LITERATURE: Peri and Orion may supply transport context and candidate D8 comparison data only after exact geometry, quantity conversion, reference values, uncertainties, and comparison criteria are approved.
  • EXTERNAL LITERATURE: Their energy-, target-, angle-, and concrete-thickness dependence supports explicitly conditioning and stratifying Paper 1 calculations, but their reported target dimensions must not be copied as approved D3 values.
  • EXTERNAL LITERATURE: Chowdhury et al. supports feasibility, not transferability or validation. Its approximately 10% example result is not a Paper 1 tolerance.
  • PROPOSED METHOD: Keep all voxels, profiles, replicas, and paired variants belonging to one physical configuration in one Dataset partition.
  • PROPOSED METHOD: Evaluate by energy, geometry boundary, barrier role, region, particle, field level, uncertainty, worst case, and underprediction rather than only by global mean.

Evidence gaps

PRELIMINARY EVIDENCE: Within the supplied bundle, neither study combines conventional industrial-LINAC converted-X-ray operation at 5-7.5 MeV, a finite scanned source head, complete PHITS room fields, particle-resolved photonuclear treatment, whole-configuration holdout, and a three-dimensional H*(10) surrogate. This is a bundle-limited gap, not a universal novelty or priority claim.

UNRESOLVED: The bundle does not establish industrial Ta-water-SS304 layer dimensions, scan weighting, finite-edge treatment, suitable D7 criteria, D8 acceptance comparisons, applicable D9 measurements, D10 partitions, or D11 performance tolerances.

Limitations

  • Two papers cannot establish comprehensive coverage or universal novelty.
  • Neither paper provides applicable complete-room measurement evidence for D9.
  • Peri and Orion's absorbed-dose/KERMA treatment cannot be relabelled as H*(10) [Peri and Orion 2017, p. 2].
  • Their isolated-target and spherical-shell calculations cannot validate the finite scanned source and complete rectangular room.
  • Chowdhury et al.'s neutron-spectrum surrogate cannot be assumed to transfer to the Paper 1 source, quantity, geometry, field representation, or partitioning procedure.
  • Literature plausibility, implementation capability, successful execution, transport completion, convergence, verification, validation, uncertainty, and acceptance remain distinct.