README.md
Signal Reconstruction
README.md
1 # Signal Reconstruction visual companion
2
3 `Signal_Reconstruction_Visual_Guide.pdf`, `.tex`, and `.md` are built from the
4 same mathematical content in `build_guide.py`. The complete Markdown files are
5 included. `guide_content.json` retains section text, equations, source anchors,
6 figure captions, and the explanatory SG crosswalk.
7
8 The guide contains 21 original matplotlib figures, each supplied as PDF, SVG,
9 and PNG. The helix in Figure 15 retains genuine three-dimensional coefficient
10 and offset geometry. No generated imagery is used.
11
12 `generate_figures.py` checks 333 finite assertions using rational and
13 Gaussian-rational controls. Continuous curves and singular-value diagnostics
14 use ordinary floating evaluation and do not certify interval bounds or analytic
15 limits. The finite tested ranges and exclusions are in
16 `validation/computation_contract.json`.
17
18 The current run is `validation/visual_run_final_20261006/manifest.json`, with
19 scientific results and exact figure hashes in the adjacent
20 `scientific_result.json`. It completed under 60 seconds wall time, 45 seconds
21 per-process CPU time, 2 GiB address-space limit, one CPU core, and cooperative
22 one-thread library caps. Its input and declared outputs passed the v2 manifest
23 validator. External dependency payload provenance and freshness are recorded in
24 `../verification/DEPENDENCY_PROVENANCE.json` and
25 `../verification/DEPENDENCY_FRESHNESS.json`; the manifest itself records versions
26 and the listed input driver hash, not the complete external dependency closure.
27
28 The earlier `validation/visual_run_20261006` is superseded. Its manifest was
29 preserved unchanged. `validation/HISTORICAL_RUN_STATUS.json` identifies the exact
30 matching historical driver and result/log validation root. Those 333 repeated
31 assertions are not counted again. The old figure payloads were replaced; their
32 recorded hashes describe the historical drawing, not final figures.
33
34 `validation/draft_checks.json` is also superseded preliminary output. It reports
35 334 checks including one vacuous empty-preparation check later removed, and
36 earlier figure hashes. It was not produced under the bounded runner and is not
37 current publication evidence. The unchanged draft is indexed with its hash in
38 `HISTORICAL_RUN_STATUS.json`. The final 333-assertion result is authoritative.
39 The computation contract remains current: its tested bounds and conventions
40 apply to the final run, whose actual driver hash is pinned by the manifest.
41
42 `validation/LAUNCH_ATTEMPT.json` records an initial pre-execution rejection:
43 the runner could not locate a relative contract path from the caller directory.
44 The corrected absolute path launched the succeeding bounded run.
45
46 Rebuild prose and PDF from the release root:
47
48 ```sh
49 python3 visual/build_guide.py
50 python3 build_release.py --only visual
51 ```
52
53 Regenerate figures under a fresh bounded manifest with the mathbox
54 computation-audit runner, using the recorded contract, driver, and caps. Select
55 a new output directory; the runner does not overwrite existing evidence.
56
57 `validation/source_anchors.json` records exact raw file hashes and selector
58 locations. `validation/claim_crosswalk.json` pins the current traditional claim
59 catalogue. The map is explanatory coverage, not a claim that every complete
60 traditional proof is reproduced in the visual guide.
61
62 Related earlier Lean sources do not certify this complete unknown-edge and
63 self-calibration paper. No fresh Lean verification is claimed here. The complete
64 release records distinguish inherited formal evidence, source inspection,
65 fresh finite numerical runs, and open conditions.