Native R reference and retained comparisons

The current mapped comparison explicitly selects the separate native count-domain source packet described below. The prior capture and its governed raw files remain historical evidence.

Historical native capture

The historical authenticated native macOS R 4.6.0 capture passed 43 core cases, 23 source-domain cases and 48 comparisons against the original R reference. The capture identifies the executed Python source revision 064d1237fd743d957538fff2cbae74db06e17338, the locked DISQOVER 0.9.14 source, the R runtime and the exact dependency versions. All 97 numeric CSV files matched the previously retained values byte for byte. The capture is a bounded numerical comparison, with no ecological accuracy or Linux release certification claim.

The native source packet at data/method-certification/native-r/capture/capture.json independently binds 138 raw members, the actual execution receipt, five locked public dependency archives and their acquisition metadata. Nine additional raw scripts and process logs remain in that packet. The core run report retains its actual elapsed time. These immutable records are preserved as source evidence; rebuilding a comparison does not execute R again or reproduce process timing.

The current comparison report and source account are derived from the separate count-domain capture described below and the original authenticated R bundle. Their native_capture_source_revision identifies the actual native execution. The governed comparison identifies its consumer through deterministic code and data digests; the execution Git revision remains in the preserved run evidence. Both explicitly state that R and the captured 43 Python expectation vectors were not newly executed for this comparison. A changed protected scientific input refuses reuse of the capture and requires a new native execution decision.

The 18 prospective optimized LOVE cases have native R results in this capture. The original reference contains 25 core cases and 23 source-domain cases; it contains no R results for those 18 later cases. The separate current Python account compares current Python with the historical reference and must retain those missing historical results explicitly. Neither comparison establishes stochastic optimizer trajectory parity, a global optimum or uncertainty calibration.

Historical annulus source capture

The preceding annulus comparison used a separate authenticated packet at data/method-certification/native-r-annulus/capture/capture.json. Its genuine native R execution used base revision ed7005d0845f4755a906d7f2642c68100f7ea3d7 with exactly two reviewed, unstaged annulus source and test edits. The manifest records both base and physical source identities, all 67 protected science hashes, the separately generated version file, and the actual execution and root-review evidence identities. The execution was not a clean checkout of that base revision.

This packet preserves all 138 actual raw members and their original execution receipt, plus the same five previously acquired public archives and original acquisition metadata. It does not replace the historical native packet described above or the 129 historical governed raw files. The existing historical retained adapter continues to authenticate those original bytes. Original 125-file R and 143-file Linux references also remain separate.

That preceding mapped comparison used the annulus packet's 97 CSV files and identified 96 declared consumer code inputs and 280 data inputs deterministically. It did not read either mapped JSON as a producer input. Rebuilding them performs no new R or Python expectation generation. The 43 core and 23 domain cases do not exercise the point-receiver annulus fallback; its separate 47-case Decimal mathematical regression evidence must not be inferred from these comparisons.

Count-domain source capture and current comparison

The current comparison and reader explicitly select data/method-certification/native-r-count-domain/capture/capture.json. Its actual native R execution used base revision 60061f9d60c519673e479fc7bf8e478cd653c4e1 with exactly the reviewed count-domain source and test overlays. The immutable manifest separately identifies the physical scientific inputs, execution receipt and two distinct local retention reviews. Those local reviews do not establish external scientific approval.

All 43 core cases, 23 source-domain cases and 48 original R comparisons passed; the 97 numeric CSV files match the preceding annulus capture byte for byte. The packet retains 138 raw members, five cached public dependency archives and their original acquisition metadata. These archives were reused for this native execution; their older acquisition receipt does not establish fresh downloads. The historical native and annulus packets remain unchanged.

The current producer declares 98 code inputs and 280 data inputs. Its configured owner selects count-domain; direct calls retain annulus as the default and must select count-domain explicitly for the current mapped comparison. An unknown, missing or unauthenticated selected packet refuses without fallback. Rebuilding these comparisons executes neither R nor new Python expectations. The native case inventory does not exercise fixed-site covariance. Its separate Python count-domain regression and mathematical producer checks provide that function's evidence and establish no covariance R parity or ecological accuracy.

Recompute the retained comparison

From an installed development environment at the repository root, choose an absent destination below artifacts/ whose parent is named producer-output:

mkdir -p artifacts/jobs/native-r-comparison/producer-output
python -m bijux_pollenomics_dev.source_completion.method_certification.native_r_comparison \
  --repository . \
  --capture-profile count-domain \
  --output artifacts/jobs/native-r-comparison/producer-output/source-products

The mapped output contains only report.json and source-replay.json. Its sibling native-comparison/ preserves the ordinary comparison output, including all 138 raw members and its actual execution revision. The command authenticates the native packet before and after, validates the original reference and checks the captured scientific source hashes against the physical repository. It never consumes the tracked comparison JSONs as producer input. The configured report owner runs the same producer in independent reference and replay lanes.

Reproduction

A new native execution uses the maintained bijux_pollenomics_dev.lra_validation.live_reference producer, the locked R runtime, Fortran toolchain and dependency archives. The source lock identifies the exact archive bytes. The public acquisition metadata at data/method-certification/native-r/public-acquisition.json records fresh public retrieval; the producer's existing cache-path label does not establish historical private archive provenance. Toolchain, source, environment and input custody must be qualified for each new execution before its results can be admitted as a separate source packet. Historical captures remain immutable.

A new run requires the qualified R, Rscript and Fortran executables. Set BIJUX_R, BIJUX_RSCRIPT and BIJUX_FORTRAN to their physical executable paths, then use an absent destination:

python -m bijux_pollenomics_dev.lra_validation.live_reference \
  --repository . \
  --output artifacts/jobs/native-r-reproduction \
  --r "$BIJUX_R" --rscript "$BIJUX_RSCRIPT" \
  --fortran "$BIJUX_FORTRAN" \
  --source-cache data/method-certification/native-r-count-domain/source-archives

Optimized LOVE objective and bounded search

The capture's 18 prospective optimized LOVE cases cover declared intermediates, taxon and ring axes, supplied-candidate wrapper replay and refusal behavior. They extend the original 25 core cases; they do not supply the missing 18 historical original-R results or pair stochastic optimizer search trajectories. The maintained solver and optimized-profile checks remain available separately:

python -m pytest -q \
  -o cache_dir=artifacts/jobs/love-optimizer-acceptance/cache \
  --basetemp=artifacts/jobs/love-optimizer-acceptance/products \
  packages/bijux-pollenomics-lra/tests/unit/optimization/test_solver.py \
  packages/bijux-pollenomics-lra/tests/unit/reconstruction/test_optimized_profile.py

Scientific and release boundary

The source replay verifier checks the retained numeric identities and current declared source closure. This verifier and the retained comparison provide separate evidence from fresh native execution. Neither supplies the locked Linux release certificate, provider rights, independent vegetation observations or full scientific acceptance.