WB01-WB04: refactor forward DOD views

[bout3fiddy]

Summary

• add borrowed ForwardSampleRequest and ForwardInputScratch views for high-resolution forward samples
• route configuredForwardInput and computeForwardSampleAtWavelengthWithScratch through named request/scratch data instead of long broad parameter lists
• keep worker-owned scratch allocation and LABOS workspace ownership unchanged

DOD friction score

• WB01 before: 351
• WB01 after: 222
• configuredForwardInput: 52 -> 0
• computeForwardSampleAtWavelengthWithScratch: 54 -> 0
• integratedSampleFromForward: 16 -> 0

Verification

• zig fmt src/forward_model/instrument_grid/grid_calculation/forward_input.zig src/forward_model/instrument_grid/grid_calculation/spectral_forward.zig
• zig test forward_input_layout_check.zig (temporary scratch assertion; removed after pass)
• zig build check
• zig build test-fast
• uv run scripts/linting/check-zig-styleguide.py src/forward_model/instrument_grid/grid_calculation/forward_input.zig src/forward_model/instrument_grid/grid_calculation/spectral_forward.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/instrument_grid/grid_calculation/forward_input.zig src/forward_model/instrument_grid/grid_calculation/spectral_forward.zig
• uv run benchmark/run_benchmark_fast.py

Parent sync

Merged latest origin/codex/instrument-grid-comment-quality during the package and re-ran the focused gates after the final merge.

[bout3fiddy]

WB02 update: Product Simulation Workspace And Plan Boundary

What changed:

• Added narrow borrowed request/output views for radiance sampling and irradiance sampling in simulate.zig.
• Kept simulateInternal as the broad product orchestrator, but stopped passing full Storage.Buffers and broad setup pieces into the sample-fill helpers.
• Added styleguide-compatible layout(64-bit) boxes for the new view structs.
• Merged latest parent draft branch codex/instrument-grid-comment-quality before final gates.

Friction score:

• WB02 current score: 814.
• simulateInternal: 63.
• fillRadianceSamples: 18.
• fillIrradianceSamples: no longer appears in WB02 top scored items after this package.

Evidence:

• Temporary same-file layout assertions passed through zig build test-unit, then were removed to preserve the no-inline-src-tests rule.
• uv run scripts/linting/check-zig-styleguide.py src/forward_model/instrument_grid/grid_calculation/simulate.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/instrument_grid/grid_calculation/simulate.zig
• git diff --check
• zig build check
• zig build test-fast
• uv run benchmark/run_benchmark_fast.py

Commit: 9c5df26 Refactor simulation sample views

[bout3fiddy]

WB03 update: Wavelength Sampling And Forward-Miss Tables

What changed:

• Added WavelengthSamplingRequest, WavelengthSamplingOutput, and WavelengthSamplingRange as stack-only borrowed views in wavelength_sampling.zig.
• Replaced the broad/long fillWavelengthSamplingPlans, fillWavelengthSamplingPlanRange, and buildWavelengthSamplingPlan signatures with request/output/range flow.
• Reduced WavelengthSamplingWorker from 176 B to 72 B by storing a pointer to the shared request plus the output view.
• Kept table ownership and allocation policy unchanged: OwnedWavelengthSampling still owns rows and side arrays; KernelStorageBuilder still owns temporary side arrays until transfer.
• Merged latest parent draft branch codex/instrument-grid-comment-quality before final gates.

Friction score:

• WB03: 264 -> 184.
• Broad owner params: 4 -> 0.
• Parameter overage: 18 -> 0.
• Broad forwarding edges: 12 -> 2.

Evidence:

• Temporary same-file layout assertions passed through zig build test-unit, then were removed to preserve the no-inline-src-tests rule.
• uv run scripts/linting/check-zig-styleguide.py src/forward_model/instrument_grid/grid_calculation/wavelength_sampling.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/instrument_grid/grid_calculation/wavelength_sampling.zig
• git diff --check
• zig build check
• zig build test-fast
• uv run benchmark/run_benchmark_fast.py

Commit: 9b90e78 Refactor wavelength sampling plan views

[bout3fiddy]

WB04 update: refactored the layer accumulation leaf family in src/forward_model/optical_properties/state_build/layer_accumulation.zig.

What changed:

• Added LayerAccumulationRequest as the borrowed view for Context, absorber build state, spectroscopy cache, collision cache, and aerosol sublayers.
• Added SublayerTarget for the row indexes/thickness passed into populateSublayer.
• Reduced ParitySupportRowWorker layout from 272 B to 184 B by borrowing the request view instead of carrying the broad setup fields in every worker row.
• Removed the hidden profileScatteringAtMidpoint helper so profileEquivalentPhaseCoefficients visibly reads the aerosol-row fields consumed by the loop.

Audit signal:

• WB04 friction score: 1169 -> 1095.
• populateSublayer item score: 59 -> 19.
• DOD layout advisory for layer_accumulation.zig: 0 findings.

Verification:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/layer_accumulation.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/layer_accumulation.zig
• git diff --check
• zig build check
• zig build test-fast
• uv run benchmark/run_benchmark_fast.py

Parent branch sync:

• Merged latest codex/instrument-grid-comment-quality before final gates.

Commit: 7a5e4a2

[bout3fiddy]

WB04 update: refactored the support-row scalar cache-miss boundary in src/forward_model/optical_properties/state_build/carrier_eval.zig.

What changed:

• Added SupportRowScalarRequest as the borrowed row view for prepared optical state, sublayer, spectroscopy cache, strong-line state, wavelength, and support-row index.
• Added SupportRowWavelengthConstants for the wavelength-local scalar constants passed through cache-miss filling.
• Replaced the 11-argument fillSharedOpticalScalarsAtSupportRowWithScalarCache boundary with fillSharedOpticalScalarsAtSupportRow(out, row, wavelength_constants).
• Kept ownership unchanged: SupportRowScalarCache still owns epoch/scalar arrays; the new request and constants structs borrow stack/cache state only.
• The new request keeps the 256 B PreparedSublayer out-of-line via pointer.

Audit signal:

• WB04 friction score: 1095 -> 1017.
• Previous P0 filler item is now classified as no-refactor-needed: params=3, broad=0, alloc_calls=0.
• DOD layout advisory for carrier_eval.zig: 0 findings.

Verification:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/carrier_eval.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/carrier_eval.zig
• git diff --check
• zig build check
• zig build test-fast
• uv run benchmark/run_benchmark_fast.py

Parent branch sync:

• Merged latest codex/instrument-grid-comment-quality before final gates.

Commit: 70902ad

[bout3fiddy]

Profile spectroscopy cache update: refactored the profile-cache value fill boundary in src/forward_model/optical_properties/state_build/layer_spectroscopy.zig.

What changed:

• Added ProfileCacheValueRequest as the borrowed view for the line list, profile thermodynamic context, prepared strong/weak states, wavelength, and prepared wavelength window.
• Reduced ProfileCacheValueWorker from 288 B to 32 B by borrowing that request instead of copying the 208 B line list and profile inputs into every worker row.
• Narrowed fillProfileSpectroscopyCacheValueRange from the long positional input list to (cache, request, start, end).
• Kept allocation and ownership unchanged: profile-cache arrays remain inline in ProfileSpectroscopyCache, and context/prepared-state/window storage is borrowed from the cache init call.

Audit signal:

• WB04 friction score: 1017 -> 1000 after regeneration.
• fillProfileSpectroscopyCacheValueRange is now classified as no-refactor-needed: params=4, broad=0, alloc_calls=0.
• DOD layout advisory for layer_spectroscopy.zig: 0 findings.

Verification:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/layer_spectroscopy.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/layer_spectroscopy.zig
• git diff --check
• zig build check
• zig build test-fast
• uv run benchmark/run_benchmark_fast.py

Parent branch sync:

• Merged latest codex/instrument-grid-comment-quality before final gates.

Commit: a513716

[bout3fiddy]

Package update: profile line-state views

Commit: 8be11b5e (Refactor profile line-state views), followed by parent sync merge d84322fa.

What changed:

• Added ProfileLineStateRequest for the profile line-state inputs shared by workers.
• Reduced ProfileLineStateWorker from the old wide per-worker row to a narrow borrowed row: request pointer, per-worker relaxation scratch slice, queue pointer, worker index.
• Narrowed fillProfileLineStateRange from the old 8-argument hot helper to (request, strong_relaxation_weights, start, end).
• Ownership stayed unchanged: profile arrays and state arrays are borrowed from prepareProfileLineStates; line-list storage ownership remains outside the request; strong relaxation scratch remains caller-owned/per-worker.

Audit signal after regeneration:

• F0409 fillProfileLineStateRange: now no-refactor-needed, params=4, broad=0, alloc_calls=0, body_lines=24.
• ProfileLineStateRequest: no structural refactor required, fields=5, slices=4, pointers=0.
• ProfileLineStateWorker: no structural refactor required, fields=4, slices=1, pointers=2.
• Bundle totals remain broad: WB04 998, WB05 1066; this item-level cleanup is visible even though the generator groups adjacent optical-preparation findings coarsely.

Verification:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/absorbers.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/absorbers.zig
• git diff --check
• zig build check
• zig build test-fast
• uv run benchmark/run_benchmark_fast.py

Parent sync:

• Merged latest origin/codex/instrument-grid-comment-quality before pushing; latest merged parent commit was 8abffef6 (Clarify instrument reference headers).

[bout3fiddy]

Package update: carrier boundary row views

Commit: a43a482a (Refactor carrier boundary row views), followed by parent sync merge 17450e69.

What changed:

• Added SourceInterfaceBoundaryRequest for the gas/particle/phase fields used to write one source-interface row.
• Added RtmQuadratureBoundaryRequest for the gas/particle/level fields used to write one RTM quadrature level.
• Narrowed private hot helpers:
  • fillSourceInterfaceFromBoundaryParts(...) -> fillSourceInterfaceFromBoundaryParts(request)
  • fillRtmQuadratureLevelFromBoundaryParts(...) -> fillRtmQuadratureLevelFromBoundaryParts(request)
• Removed broad PreparedOpticalState from the source-interface leaf helper; it now borrows only aerosol_phase_coefficients plus the output row pointer.
• No ownership or allocation behavior changed. The new requests are stack rows with borrowed pointers and value carriers.

Audit signal after regeneration:

• WB04 score: 998 -> 980.
• F0453 fillSourceInterfaceFromBoundaryParts: now no-refactor-needed, params=1, broad=0, alloc_calls=0, body_lines=25.
• F0456 fillRtmQuadratureLevelFromBoundaryParts: now no-refactor-needed, params=1, broad=0, alloc_calls=0, body_lines=17.
• New request structs are classified as request/config rows with no structural refactor required.

Layout evidence:

• SourceInterfaceBoundaryRequest: 88 B, align 8, 2 cache lines, borrowed phase/output storage.
• RtmQuadratureBoundaryRequest: 120 B, align 8, 2 cache lines, borrowed output storage.

Verification:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/carrier_eval.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/carrier_eval.zig
• git diff --check
• zig build check
• zig build test-fast
• uv run benchmark/run_benchmark_fast.py

Parent sync:

• Merged latest origin/codex/instrument-grid-comment-quality before pushing; latest merged parent commits were 369bd5fc and 6c21c103.

[bout3fiddy]

Package update: WB03 wavelength sampling worker ownership

Scope:

• Added WavelengthSamplingWorkerLease as the named temporary owner row for parallel wavelength-plan workers and thread handles.
• Moved worker/thread heap setup into buildWavelengthSampling, so fillWavelengthSamplingPlans now borrows the lease and reads as fill-only execution.
• Merged parent draft updates through 8e288db7 Clarify carrier evaluation cache routes; resolved the carrier-eval comment/request-wrapper conflict by keeping this branch's narrow request structs and preserving the parent explanatory comments.

Audit:

• WB03 before this package: 184.
• WB03 on final merged head: 169.
• WB02 remains 353 on final merged head.
• WB04 is 611 after the latest parent merge and conflict resolution.

Verification on final head 6021f7bc:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/instrument_grid/grid_calculation/wavelength_sampling.zig src/input/o2a_reference/run.zig src/forward_model/optical_properties/state_build/carrier_eval.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/instrument_grid/grid_calculation/wavelength_sampling.zig src/input/o2a_reference/run.zig src/forward_model/optical_properties/state_build/carrier_eval.zig
• git diff --check
• python3 scratch/reports/dod_memory_audit/analyze_dod_memory.py && python3 scratch/reports/dod_memory_audit/generate_friction_scores.py && python3 scratch/reports/dod_memory_audit/generate_refactor_proposal.py && python3 scratch/reports/dod_memory_audit/generate_work_bundles.py
• zig build check
• zig build test-fast
• uv run benchmark/run_benchmark_fast.py

Benchmark run 2a825cbabf024e4f83b577e5be4006ab:

• forward/no-session: 1.434s, 0.972s, 0.974s
• session setup 0.710s; first cached run 0.266s; cached repeats 0.265s, 0.265s, 0.269s
• fast-mode reference baseline 2.019s; reference oblique aerosol 2.392s; four-scene totals 2.396s, 2.393s, 2.417s
• retrieval single session totals 1.826s, 1.824s; retrieval 1.118s, 1.117s
• retrieval single fast_mode totals 1.120s, 1.118s; retrieval 0.530s, 0.529s
• retrieval sweep case 001 session/fast_mode 6.774s/3.408s (local outlier in this run)
• retrieval sweep case 003 session/fast_mode 5.345s/1.640s

[bout3fiddy]

Package update: WB04/WB05 layer accumulation deltas

Scope:

• Changed populateSublayer, populateParitySupportRow, and populateLayer to return local LayerAccumulation deltas instead of mutating a broad *LayerAccumulation passed deep into the sublayer path.
• Replaced mergeParitySupportTotals with mergeLayerAccumulationTotals, used at the caller boundary to make the flow explicit: fill row -> return contribution -> merge.
• Merged parent draft updates through 3fdeb1ed Clarify prepared state finalization handoff.

Audit on final merged head e94c5127:

• WB04: 757 -> 728.
• WB05: 948 -> 942.
• WB10 remains 113 after the parent integration-header merge.

Verification on final head e94c5127:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/layer_accumulation.zig src/forward_model/implementations/instrument/integration.zig src/forward_model/optical_properties/state_build/finalize.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/layer_accumulation.zig src/forward_model/implementations/instrument/integration.zig src/forward_model/optical_properties/state_build/finalize.zig
• git diff --check
• python3 scratch/reports/dod_memory_audit/analyze_dod_memory.py && python3 scratch/reports/dod_memory_audit/generate_friction_scores.py && python3 scratch/reports/dod_memory_audit/generate_refactor_proposal.py && python3 scratch/reports/dod_memory_audit/generate_work_bundles.py
• zig build check
• zig build test-fast
• uv run benchmark/run_benchmark_fast.py

Benchmark run 26db98ecee5c441ab092ea7047d5abae:

• forward/no-session: 1.383s, 0.972s, 0.972s
• session setup 0.708s; first cached run 0.265s; cached repeats 0.266s, 0.265s, 0.267s
• fast-mode reference baseline 2.022s; reference oblique aerosol 2.387s; four-scene totals 2.401s, 2.414s, 2.400s
• retrieval single session totals 1.845s, 1.828s; retrieval 1.122s, 1.120s
• retrieval single fast_mode totals 1.123s, 1.123s; retrieval 0.530s, 0.530s
• retrieval sweep case 001 session/fast_mode 4.869s/1.621s
• retrieval sweep case 003 session/fast_mode 5.270s/1.622s

[bout3fiddy]

WB05 package: vertical grid storage ownership

Changed OwnedVerticalGrid from fourteen independent heap column allocations to one aligned backing allocation with typed borrowed column slices. The public column fields remain the consumer-facing shape; owned_storage is now the single release point, with the layout(64-bit) comment and compiler-backed layout test updated from 224 B to 240 B.

Audit movement after merging latest parent draft commits:

• WB05 score: 942 -> 747
• WB05 ownership sites: 211 -> 146
• F0734 vertical_grid.allocate dropped out of the highest-scoring items.

Verification on pushed head ad49cfda:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/vertical_grid.zig src/forward_model/optical_properties/state_build/rtm_quadrature.zig tests/unit/forward_model/optical_properties/state_build/root_test.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/vertical_grid.zig src/forward_model/optical_properties/state_build/rtm_quadrature.zig
• python3 scratch/reports/dod_memory_audit/analyze_dod_memory.py && python3 scratch/reports/dod_memory_audit/generate_friction_scores.py && python3 scratch/reports/dod_memory_audit/generate_refactor_proposal.py && python3 scratch/reports/dod_memory_audit/generate_work_bundles.py
• git diff --check
• zig build check
• zig build test-fast
• zig build test
• uv run benchmark/run_benchmark_fast.py
• prek run --all-files

Stable benchmark rerun used for evidence:

• forward/no-session median: 0.983s
• forward/session cached median: 0.265s
• forward/fast-mode median total: 2.413s
• OE session median retrieval: 1.120s
• OE sweep fast total: 2.051s

[bout3fiddy]

WB05/WB04 package: narrow parity layer accumulation inputs

Changed the parity-support branch in layer_accumulation.zig so helper calls borrow the specific data they need instead of forwarding the full preparation context deeper into the path:

• seedParitySupportRowLayerIndices now takes layers and sublayers.
• canParallelPopulateParitySupportRows now takes row_count.
• reduceParityLayer now takes scene, vertical_grid, and sublayers.
• usesDisamarParitySupportGrid now reads scene directly.
• populateSublayer reuses the computed parity-grid route and groups support-weight/path-length selection into one branch.

Audit movement after merging latest parent draft commits:

• WB05 score: 747 -> 737
• WB04 score: 728 -> 716
• F0536 populate: 30 -> 18

Verification on pushed head 05e6f90a:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/layer_accumulation.zig src/forward_model/instrumentation/stubs/calculation_telemetry_sink.zig src/forward_model/instrumentation/telemetry.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/layer_accumulation.zig src/forward_model/instrumentation/stubs/calculation_telemetry_sink.zig src/forward_model/instrumentation/telemetry.zig
• python3 scratch/reports/dod_memory_audit/analyze_dod_memory.py && python3 scratch/reports/dod_memory_audit/generate_friction_scores.py && python3 scratch/reports/dod_memory_audit/generate_refactor_proposal.py && python3 scratch/reports/dod_memory_audit/generate_work_bundles.py
• git diff --check
• zig build check
• zig build test-fast
• zig build test
• uv run benchmark/run_benchmark_fast.py
• prek run --all-files

Stable benchmark rerun used for evidence:

• forward/no-session median: 0.975s
• forward/session cached median: 0.265s
• forward/fast-mode median total: 2.402s
• OE session median retrieval: 1.135s
• OE sweep fast total: 2.051s

[bout3fiddy]

WB05 absorber builder package

Change:

• Narrowed src/forward_model/optical_properties/state_build/absorbers.zig so buildCrossSectionAbsorbers and buildLineAbsorbers take sublayer_count instead of the whole PreparationContext.
• Hoisted exact borrowed inputs in build: cross sections, profile T/P/z slices, borrowed profile line-state slices, midpoint wavelength, and air-mass LUT. This keeps allocation/ownership at the same setup boundary but makes the data flow visible at call sites.
• Merged latest parent draft commits through ab0ba085 before push.

Audit movement after regeneration:

• WB05 score: 737 -> 686.
• F0424 build dropped out of the current friction-score table; before this package it was 27 with 9 broad-owner edges.

Verification on final pushed head 2af498b8:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/absorbers.zig src/forward_model/optical_properties/state_build/spectroscopy.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/absorbers.zig src/forward_model/optical_properties/state_build/spectroscopy.zig
• audit regeneration chain: analyze_dod_memory.py && generate_friction_scores.py && generate_refactor_proposal.py && generate_work_bundles.py
• git diff --check
• zig build check
• zig build test-fast
• zig build test (o2a_vendor_forward_reflectance current residual exact-zero)
• uv run benchmark/run_benchmark_fast.py
• prek run --all-files

Fast benchmark canary on final head:

• forward/no-session warm repeats: 0.987s, 0.988s
• forward/session cached repeats: 0.279s, 0.269s, 0.268s
• forward/fast-mode four-scene repeats: 2.407s, 2.438s, 2.427s
• retrieval single session: 1.136s, 1.132s; fast-mode: 0.538s, 0.538s
• retrieval sweep session cases: 4.223s, 4.616s; fast-mode cases: 1.038s, 1.096s

[bout3fiddy]

Parent sync after absorber package

After the absorber evidence comment, the parent draft branch advanced again with 61857eed Clarify bundled asset selection comments. I merged it cleanly and pushed final head 5f5d2d39.

Additional final-head checks after that merge:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/absorbers.zig src/input/reference_data/bundled/assets.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/absorbers.zig src/input/reference_data/bundled/assets.zig
• git diff --check
• zig build check
• zig build test-fast
• prek run --all-files

The full test and fast benchmark evidence in the previous comment was run on 2af498b8 immediately before this parent-only asset-comment sync; the absorber code did not change in the final merge.

[bout3fiddy]

WB06 LABOS tangent request package

Change:

• Added NonIntegratedReflectanceTangentRequest in src/forward_model/radiative_transfer/labos/execute.zig as an 80 B stack-only borrowed view for the non-integrated aerosol tangent route.
• Replaced the two loose ten-argument calls into nonIntegratedReflectanceTangent with request literals plus the dynamic attenuation table.
• The request borrows layers, geometry, RT rows, controls, and PLM basis; it owns no heap storage and does not change tangent allocation behavior.
• Merged latest parent draft commits through c40c77c8 before push.

Audit movement after regeneration:

• WB06 score: 486 -> 482.
• F0773 nonIntegratedReflectanceTangent dropped out of the current friction-score table; before this package it was 20 with parameter_overage=4 and hot_path_allocator_sites=4.

Verification on final pushed head 29902c62:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/radiative_transfer/labos/execute.zig src/forward_model/instrument_grid/grid_calculation/spectral_eval.zig src/output/o2_line_contributions.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/radiative_transfer/labos/execute.zig src/forward_model/instrument_grid/grid_calculation/spectral_eval.zig src/output/o2_line_contributions.zig
• audit regeneration chain: analyze_dod_memory.py && generate_friction_scores.py && generate_refactor_proposal.py && generate_work_bundles.py
• git diff --check
• zig build check
• zig build test-fast
• zig build test (o2a_vendor_forward_reflectance current residual exact-zero)
• uv run benchmark/run_benchmark_fast.py
• prek run --all-files

Fast benchmark canary on final head:

• forward/no-session warm repeats: 0.972s, 0.972s
• forward/session cached repeats: 0.282s, 0.265s, 0.273s
• forward/fast-mode four-scene repeats: 2.410s, 2.396s, 2.398s
• retrieval single session: 1.102s, 1.103s; fast-mode: 0.528s, 0.527s
• retrieval sweep session cases: 4.149s, 4.551s; fast-mode cases: 1.021s, 1.023s

[bout3fiddy]

WB04 source-interface request package

Change:

• Added SourceInterfaceSpectroscopyRequest and SourceInterfaceCarrierRequest in src/forward_model/optical_properties/state_build/source_interfaces.zig as 56 B stack-only borrowed views.
• Public source-interface boundaries now assemble request rows and delegate to request-based private helpers.
• The request rows borrow prepared optical state, layer inputs, source-interface output rows, and profile/wavelength caches; no heap storage, backing-slice copies, or ownership changes.
• Merged latest parent draft commit a8d24c15 before push.

Audit movement after regeneration:

• Targeted source_interfaces.zig public functions now score zero:
  • F0664 fillSourceInterfacesAtWavelengthWithLayers: 0
  • F0665 fillSourceInterfacesAtWavelengthWithLayersAndSpectroscopyCache: 0
  • F0666 fillSourceInterfacesAtWavelengthWithLayersAndCarrierCache: 0
• Aggregate WB04 remained 716 after regenerated inventory, but the targeted broad-forwarding/scratch findings were removed from these public source-interface boundaries.

Verification on final pushed head 2e236ebd:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/source_interfaces.zig src/forward_model/optical_properties/state_build/rtm_quadrature.zig src/input/ObservationModel.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/source_interfaces.zig src/forward_model/optical_properties/state_build/rtm_quadrature.zig src/input/ObservationModel.zig
• audit regeneration chain: analyze_dod_memory.py && generate_friction_scores.py && generate_refactor_proposal.py && generate_work_bundles.py
• git diff --check
• zig build check
• zig build test-fast
• zig build test (o2a_vendor_forward_reflectance current residual exact-zero)
• uv run benchmark/run_benchmark_fast.py
• prek run --all-files

Fast benchmark canary on final head:

• forward/no-session warm repeats: 0.972s, 0.972s
• forward/session cached repeats: 0.262s, 0.262s, 0.262s
• forward/fast-mode four-scene repeats: 2.415s, 2.397s, 2.413s
• retrieval single session: 1.106s, 1.108s; fast-mode: 0.530s, 0.545s
• retrieval sweep session cases: 4.148s, 4.555s; fast-mode cases: 1.020s, 1.024s

[bout3fiddy]

WB04 layer-spectroscopy request package

Change:

• Added LineSpectroscopyRequest and CachedSingleLineRequest in src/forward_model/optical_properties/state_build/layer_spectroscopy.zig.
• populateSublayer now prepares the line-spectroscopy row explicitly before calling the spectroscopy helpers.
• resolveCachedSingleLineEvaluation and resolveSpectroscopyEvaluation now accept narrow borrowed request rows instead of broad context/absorbers plus loose support-row scalars.
• Request rows are stack-only borrowed views: LineSpectroscopyRequest is 208 B, CachedSingleLineRequest is 120 B. No heap ownership, backing-slice copies, or allocation behavior changed.
• Merged latest parent draft commits through aabc50f8 before push.

Audit movement after regeneration on final head 4434d22c:

• WB04 score: 716 -> 694.
• Targeted layer-spectroscopy rows now score zero:
  • F0564 resolveCachedSingleLineEvaluation: 0
  • F0565 resolveSpectroscopyEvaluation: 0
  • S0171 LineSpectroscopyRequest: 0
  • S0172 CachedSingleLineRequest: 0
• populateSublayer remains nonzero (8) because it is still the support-row assembly point, but the broad deep line-spectroscopy call route is removed.

Verification on final pushed head 4434d22c:

• uv run scripts/linting/check-zig-styleguide.py src/forward_model/optical_properties/state_build/layer_spectroscopy.zig src/forward_model/optical_properties/state_build/layer_accumulation.zig src/forward_model/optical_properties/state_build/forward_layers.zig src/forward_model/optical_properties/state_build/rtm_quadrature.zig
• uv run scripts/linting/check-zig-dod-layout.py src/forward_model/optical_properties/state_build/layer_spectroscopy.zig src/forward_model/optical_properties/state_build/layer_accumulation.zig src/forward_model/optical_properties/state_build/forward_layers.zig src/forward_model/optical_properties/state_build/rtm_quadrature.zig
• audit regeneration chain: analyze_dod_memory.py && generate_friction_scores.py && generate_refactor_proposal.py && generate_work_bundles.py
• git diff --check
• zig build check
• zig build test-fast
• zig build test (o2a_vendor_forward_reflectance current residual exact-zero)
• uv run benchmark/run_benchmark_fast.py
• prek run --all-files

Fast benchmark canary on final head:

• forward/no-session warm repeats: 0.973s, 0.974s
• forward/session cached repeats: 0.262s, 0.262s, 0.262s
• forward/fast-mode four-scene repeats: 2.395s, 2.398s, 2.395s
• retrieval single session: 1.119s, 1.105s; fast-mode: 0.529s, 0.533s
• retrieval sweep session cases: 4.149s, 4.561s; fast-mode cases: 1.024s, 1.024s