EDPW Code Release v2_0_01 -- DEVELOPMENT CHECKPOINT
MATLAB R2016a target (not independently executed on MATLAB R2016a).

Run RUN_ALL_EDPW_v2_0_01.m from this directory, or run scripts separately.
The public package has its own algorithm helpers; no internal repository needed.

Exp I: analytic 5 GS/s, N=5,000,000, w=1000 wideband workload.
       Does NOT process a 5-million-sample recording. The FFT operation
       count is a radix-2-equivalent reference, not an exact mixed-radix count.
Exp II: conceptual N_KERNEL grid arithmetic; no generalized transform
        or increased physical resolution is claimed.
Exp III-A: deterministic spectral-vulnerability / exact-coefficient checks.
Exp III-B: cost-ELIGIBILITY and fidelity; not a fair matched-output winner ranking.
Exp III-C: Monte Carlo local-contrast diagnostics using known frequency region;
           not blind discovery probability or general detection probability.
SELFTEST: tests direct/fast, pruned FFT, Goertzel and CZT equivalence.

Caveats: Selective algorithms are given oracle-selected bins in III-A;
real discovery cost is not included. Phase-inclusive Fast-EDPW cost is an
optimized estimate; actual MATLAB fast.m calculates extra exponentials.
No wall-clock benchmark or RAM peak benchmark has been independently verified.
STFT is a different time-frequency task and not directly ranked as full FFT.
Do not publish quantitative results without independent MATLAB execution.

Exp III-D: blind all-bin, equal-output-coverage comparison of FFT,
           output-pruned FFT, Goertzel, CZT, Direct and Fast EDPW.
           All tone locations withheld from algorithms; ground truth
           used only to score coarse-band recall. Results are scenario-
           dependent, not universal detection probabilities.

Experiment III-E compares a predetermined contiguous frequency band using identical
output indices across all methods. It separates exact full-record DFT methods
from EDPW short-window approximations and documents operation-count caveats.

Experiment III-E uses edpw_v2_selected_block: the code actually executes only the selected EDPW block, matching the reported selected-block arithmetic model.

Experiment III-F: quality-conditioned full-spectrum cost frontier. Error is defined against full-record FFT magnitude, not a claim of physical ground truth. The cost model is not measured MATLAB runtime.

2026-10-08 audit update: SELFTEST expanded to w=1, w=N, N=300,
selected-block consistency and exact-bin comparisons. See accompanying
VERIFICATION_REPORT_v2_0_01.txt for independently checked equations.
This does NOT replace MATLAB R2016a execution.

2026-10-08 cost-audit correction: warm/precomputed-kernel estimates are NOT the
actual cold-call arithmetic of the provided MATLAB helpers, which calculate
chirp-kernel FFTs on invocation. edpw_v2_costs now reports both estimates.
FFT cost is radix-2 only for power-of-two N. Real recurrence operations of
Goertzel are not directly comparable with complex multiplication counts.

2026-10-08 REPRODUCIBILITY UPDATE: RUN_ALL_EDPW_v2_0_01 is now a function.
It executes every experiment in base workspace, collects console output via diary,
and exports each figure to a timestamped Results_* directory. The runner
requires MATLAB execution for verification; do not treat source audit as such.
Experiment III-F now includes FFT as an eligible baseline at every threshold,
and reports both warm and cold Fast-EDPW estimated costs.

2026-10-08 COST CONSISTENCY: Experiment III-B now displays both warm and cold
Fast-EDPW estimates and labels budget eligibility rather than equal-output
performance. Experiment III-E also reports CZT cold-call estimates.

IMPORTANT: None of these cost tables are calibrated MATLAB timings, and
complex arithmetic estimates cannot be compared directly with Goertzel real
recurrence counts. A selected-band method is not a blind discovery method.

RUNNER AUDIT UPDATE (2026-10-08)
-------------------------------
RUN_ALL_EDPW_v2_0_01 now writes run_manifest.csv (PASS/FAIL, elapsed
wall-clock seconds), a .mat workspace snapshot per completed experiment,
console transcript and each generated figure. MAT snapshots preserve the
actual computed arrays rather than only their printed summaries. Timing
in the manifest measures script execution, not controlled benchmark timing.
WARNING: experiment scripts begin with `clear`, so the runner intentionally
clears the MATLAB BASE workspace; save unsaved user variables before use.
A successful archive integrity check is NOT a MATLAB execution test.

AUDIT ADDENDUM (2026-10-08)
- SELFTEST now checks offset-start CZT bands, not just full-band CZT.
- FFT radix-2 complexity is an extrapolation for non-power-of-two N.
- Reported operation counts are algorithmic models, not elapsed time or
  exact floating-point instruction counts.
- MATLAB R2016a execution has NOT been independently verified.
- The manuscript must not be finalized from unexecuted experiment scripts.

PUBLICATION FIGURES: Run Exp_III_Publication_Results_v2_0_01.m for the deterministic Section 16 figures/table. Independent Python/NumPy calculations are reported in the manuscript; MATLAB R2016a execution is not yet verified.


V2_0_02 ADDITION: Exp_III_All_Methods_v2_0_02.m reproduces the common
33-bin comparison for FFT, pruned FFT, Goertzel, CZT, Direct/Fast EDPW;
its second independent scenario displays an STFT spectrogram.
section16_reference_v2_0_02.py independently reproduces the paper figures
and CSV with NumPy/Matplotlib; this does NOT verify MATLAB execution.
The EDPW cost column refers to full-output helpers, not a selective block.
