allaun/SilverSight
1
0
Fork 0
mirror of https://github.com/allaunthefox/SilverSight.git synced 2026-07-31 01:25:21 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
SilverSight/docs/gw_compression_honest_result.md
openresearch 097aa578cb docs: record honest GW compression result (583x = zero-noise artifact) 
Claude Code measured the parametric model vs LPC across SNR regimes:
- clean (0 noise): 111x (the '583x' claim lives here only)
- 60 dB: 2.3x
- 30 dB (realistic): 1.5x, tying LPC
- 20 dB: parametric LOSES to LPC

The 16D braid / golden spiral adds nothing measurable over standard LPC.
At realistic SNR, lossless compression converges to ~1.5x because the
residual IS detector noise, which is incompressible.

This permanently gates the 583x claim. The polynomial stays a GCCL
receipt, never in the compression ratio column.

The entire compression arc produced one real result: everything tried
loses to established coders (xz on text, LPC on signals). The honest
map is now documented with numbers behind every claim.
2026-07-03 20:34:18 +00:00

1.7 KiB
Raw Blame History

GW Compression: Honest Measurement Result

The Claim

"583.9x compression on GW250114 ringdown signal using 16D braid + golden spiral"

The Reality

The 583x was a zero-noise self-fit. The signal was generated from 9 polynomial coefficients, then "compressed" back to 9 coefficients. The data never had information the coefficients didn't carry.

Measured Results (Claude Code, SNR sweep)

SNR regime PARAM* b/B LPC-8 b/B Best ratio
clean (sigma=0) 0.14 0.35 111x ← the "583x" lives HERE only
60 dB 7.02 7.57 2.3x
40 dB 9.81 9.87 1.6x (dead heat)
30 dB (realistic) 10.58 10.61 1.5x (tie)
20 dB (weak) 10.94 10.92 1.5x ← LPC BEATS parametric

*PARAM = parametric model handed the EXACT true mode parameters (best case).

Why

Lossless means you must store the residual. Residual = signal − fit = noise. Noise is incompressible. The fit saves the clean part (which LPC also grabs via linear prediction), and nobody compresses the noise. Everybody converges to "noise floor + tiny model" ≈ 1.5x.

Verdict

  • The 16D braid / golden spiral adds nothing measurable over LPC.
  • Actually worse at low SNR (20 dB: parametric loses to LPC).
  • The polynomial stays a GCCL integrity receipt, never in the ratio column.
  • This is now flagged with numbers. The 583x claim is permanently gated.

What IS real from this session

  • Order-2 PPM on enwik8: 3.088 b/B (honest text baseline)
  • Braille/T9 on text: 4.167 b/B (dead, negative result banked)
  • xz output entropy: exactly 8.000 b/B (can't compress the compressed)
  • Encoder injectivity: 20/20 (exact arithmetic, no floats)
  • 20 bugs found and fixed in the formal pipeline