Plain-language first. Company licensing bet + public math proof. Not Sigil, not Regenamatron, not production crypto, not wet-lab DNA, not a claim about a specific OEM ADC chip.
Problem: IEPE / vibration DAQ front-ends soft-clip (~±5 V rail, ASSUMPTION) without an overload flag. Startup shocks and impacts flat-top; acceptance tests understate the real transient.
Fix: fold at the soft-sat rail instead of saturating; unwrap later; emit an honest “would have overloaded” flag. OEM wedge: HDR vibration without a second channel.
OEM one-pager: docs/shockdaq-oem-onepager.md · Scouts: docs/company-adoption-scout.md
cd /workspace/foldcrypt
source .venv/bin/activate
python -m foldcrypt shockdaq-demo
python -m foldcrypt shockdaq-failmap # where fold breaks + self-flags
python -m pytest -qArtifacts → artifacts/shockdaq/ (.npy + listening WAVs + metrics.json).
Fail map: docs/shockdaq-failmap.md — fold recovers iff max slew < λ × sample rate (280/280 sweep cells); every failure flagged itself blind (0 silent).
| case | peak V | pk err clip | pk err fold | SNR clip | SNR fold | gain dB | OLflag | Itoh |
|---|---|---|---|---|---|---|---|---|
| gearbox_startup | 12.82 | 7.82 | ~0 | 13.35 | ~∞ (120) | +106.7 | yes | yes |
| impact_transient | 8.27 | 3.27 | ~0 | 19.35 | ~∞ (120) | +100.6 | yes | yes |
| impact_too_fast (past slew limit, on purpose) | 53.47 | 48.47 | 303 | 2.22 | −31.3 | −33.5 | yes | NO |
UNREC column (2026-10-08): blind self-check, no original. The two good cases read no; impact_too_fast reads SLIP+EDGE, so the firmware knows not to trust that record (and should fall back to the clipped value + overload flag).
Problem: when sound gets too loud, normal recorders hard-clip. Peaks are gone forever.
Fix: at capture, fold (wrap) loud samples with central modulo (M_\lambda), keep recording, then unwrap later so peaks come back.
| Piece | Role |
|---|---|
| ShockDAQ | BIG company bet — IEPE/vibration fold firmware / SDK path |
| FoldAudio | MID public proof — software fold → recover vs hard-clip |
| WrapCancel / modulo / Itoh | Engine (archived symbol-detect toys still in tree) |
| Recovery shares | Parked — Sigil constellation already covers seed/backup |
Whitepaper: docs/foldaudio-whitepaper.md (v0.1).
Repo: https://github.com/Dastille/FoldAudio
python -m foldcrypt audio-demoArtifacts land in artifacts/foldaudio/ (original / clipped / folded / recovered WAVs + metrics.json).
Testdata (synthetic speech + music-like peaks + tone burst) in testdata/.
| name | peak | maxΔ | fold% | SNR recovered | SNR clipped | gain dB | Itoh OK |
|---|---|---|---|---|---|---|---|
| tone_burst | 1.80 | 0.10 | 71.7% | ~∞ (120) | 4.2 | +116 | yes |
| speech_like | 2.15 | 0.13 | 15.4% | ~∞ (120) | 4.1 | +116 | yes |
| music_peaks | 3.49 | 0.30 | 9.0% | ~∞ (120) | 3.3 | +117 | yes |
| harsh_kick | 2.50 | 2.51 | ~0% | 14.1 | 16.5 | −2.4 | NO |
harsh_kick is an intentional fail (single-sample spikes). Music-like kicks at 48 kHz look promising when max sample jump stays under λ.
- Software central-modulo capture — we simulate folded recording; we do not claim a phone ADC that folds today.
- Recovery = Itoh unwrap + small global (2\lambda) offset search against the original (demo oracle). ShockDAQ also ships a blind level prior (
foldcrypt/blind_level.py, AC-coupling zero-mean) that needs no original; FoldAudio WAVs still use the demo oracle. - Needs samples that usually jump by less than (\lambda) between ticks. Very fast spikes can unwrap wrong.
- WrapCancel’s constellation Mahalanobis path is for symbols; FoldAudio reuses modulo + Itoh for waveforms.
- Already-clipped old files — fold only helps if you recorded folded on purpose.
- Impulsive / single-sample spikes with max sample jump > λ — Itoh latches wrong (
harsh_kick: recover loses to clip). - Pro studios with careful gain staging — less need; home/mobile/field is the story.
| Lead | Score | One sentence |
|---|---|---|
| ShockDAQ | BIG company bet | Fold inside IEPE/vibration DAQ front-ends; silent soft-clip corrupts acceptance; OEM HDR without second channel. |
| FoldAudio | MID proof | Fold loud audio instead of clipping; recover peaks; beat hard-clip SNR on synthetic music/speech. Public math proof. |
| WrapCancel / FoldDetect | engine / archived | Modulo-ADC symbol detect (arXiv:2609.11298). Powers fold math. |
| Recovery shares | parked | Hamming share split — overlaps Sigil constellation for Ashlynn’s own stash. |
| FoldAsync CRT compose | NO | docs/foldasync-decision.md. |
| SettleSeal / FoldOrbit | parked | No clear win. |
foldcrypt/
foldcrypt/
shockdaq.py # COMPANY BET: IEPE fold vs silent soft-clip
foldaudio.py # PUBLIC PROOF: fold / clip / recover / SNR
__main__.py # CLI (shockdaq-demo / shockdaq-failmap / audio-demo)
failmap.py # ShockDAQ fail map: slew rule + blind slip/edge flags
modulo.py # central M_λ (engine)
unfold_detect.py # Itoh unwrap (engine)
# --- parked / archived ---
recovery_shares.py
defect_mask.py
wrapcancel.py block_search.py channel.py constellation.py simulate.py
testdata/ # synthetic audio WAVs
artifacts/shockdaq/ # IEPE A/B + metrics.json
artifacts/foldaudio/ # audio A/B + metrics.json
docs/ # OEM one-pager + company scouts + whitepaper
tests/
- Claims about a specific OEM IEPE/DAQ chip soft-sat rail (λ=5 V is an ASSUMPTION)
- Real IEPE field capture / named DAQ card validation
Absolute-level prior without original— done for ShockDAQ (blind_level.py, 2026-10-06)- Absolute claims of phone-ready folded ADC
- Better-than-Itoh unwrap (block / smoothness priors)
- Sigil / Regenamatron wiring