Time series codecs

See also: rate-controlled quantize → predict → entropy code experiment; fluxcode timestamps: the time axis; fluxcode on swinging-door historian data; fluxprune: keeping the points that matter.

A comparison of 41 time-series codecs on synthetic 1 kHz signals. Data: 5 one-minute signals of each of the 12 kinds (3,600 blocks of 1000 samples, fixed seeds). Every codec encodes whole one-minute block groups (60 blocks) and is charged every byte needed to decode the block group from those bytes alone; anything that depends on a range (quantizer grids, % of range thresholds) works per 1-second block. Raw is 1000 × float64 = 8000 bytes per block.

Metrics: all errors are signed, per block, and a percentage of that block's range (max − min); tables give the median and the worst over the dataset's blocks. “Max err” is decoded max − true max (negative = peak clipped). Bits/sample counts whole block groups.

Regenerate: cd experimental && uv run python make_report.py (writes this page and its SVGs into report/; make_rate_report.py writes rate.html, make_time_report.py time.html, make_sdt_report.py sdt.html, make_prune_report.py prune.html).

Codec legend

One row per codec, in the order used by every table and chart: family (classic, constant width, entropy, fluxcode), then nominal bits per sample. The colored bar is the family's chart color. “Lossless?”: bit-exact on any input; fluxcode is bit-exact when the data sits on a decimal grid. “Constant width”: the same number of bytes for every block.

codecfamilylossless?constant width?descriptionmodule
pwlinear-minmax16-mergedclassicnonoPiecewise linear through the min and max of every 16 samples, boundary knots mergedclassic.piecewise
swingdoor-0.5%classicnonoSwinging-door trending, deviation 0.5% of rangeclassic.swinging_door
swingdoor-2%classicnonoSwinging-door trending, deviation 2% of rangeclassic.swinging_door
dct-top64classicnoyesDCT, 64 largest coefficients keptclassic.dct
pwlinear-minmax16classicnonoPiecewise linear through the min and max of every 16 samplesclassic.piecewise
pchip-minmax16classicnonoPCHIP cubic through the min and max of every 16 samplesclassic.piecewise
bintree-centerclassicnoyes2-bit narrowing tree, 1-bit leaves decoded to the center of their rangeclassic.bintree
bintree-edgeclassicnoyes2-bit narrowing tree, 1-bit leaves decoded to the edge of their rangeclassic.bintree
dpcm4-linear-deflateclassicnono4-bit DPCM, linear steps, deflateclassic.dpcm
dpcm4-cubic-deflateclassicnono4-bit DPCM, cubic steps 0.8u³ + 0.2u, deflateclassic.dpcm
quant6-delta1-deflateclassicnono6-bit uniform quantization, first difference mod 256, deflateclassic.quant
dpcm6-linear-deflateclassicnono6-bit DPCM, linear steps, deflateclassic.dpcm
dpcm6-cubic-deflateclassicnono6-bit DPCM, cubic steps 0.8u³ + 0.2u, deflateclassic.dpcm
dpcm6-mulaw-deflateclassicnono6-bit DPCM, μ-law steps (μ = 7), deflateclassic.dpcm
quant8classicnoyes8-bit uniform quantization, one byte per sampleclassic.quant
quant8-delta1-deflateclassicnono8-bit uniform quantization, first difference mod 256, deflateclassic.quant
gorilla-xorclassicyesnoGorilla XOR, losslessclassic.gorilla
cw-delta1-tree-2constant widthnoyesDelta tree: 1 halving bit per node, 2-bit sign-relative leavesconstwidth.delta_tree
cw4-delta1-bfp-e4constant widthnoyes4-bit closed-loop delta, 4-bit step exponent per 16 samplesconstwidth.delta_bfp
cw-delta1-tree-4constant widthnoyesDelta tree: 1 halving bit per node, 4-bit sign-relative leavesconstwidth.delta_tree
cw6-delta1-bfp-e4constant widthnoyes6-bit closed-loop delta, 4-bit step exponent per 16 samplesconstwidth.delta_bfp
cw8-delta1-linearconstant widthnoyes8-bit closed-loop delta, uniform steps of D/127constwidth.delta_linear
cw8-delta1-sqrtconstant widthnoyes8-bit closed-loop delta, resolution ∝ √|delta| (CLZ decode)constwidth.delta_sqrt
cw8-delta1-bfp-e2constant widthnoyes8-bit closed-loop delta, 2-bit step exponent per 16 samplesconstwidth.delta_bfp
cw8-delta1-bfp-e4constant widthnoyes8-bit closed-loop delta, 4-bit step exponent per 16 samplesconstwidth.delta_bfp
cw-delta1-tree-7constant widthnoyesDelta tree: 1 halving bit per node, 7-bit sign-relative leavesconstwidth.delta_tree
ratectl-delta0123-zstd-4bentropynonoRate control to 4 bits/sample, then delta0123-zstdentropy.ratectl_codecs
ratectl-flac-4bentropynonoRate control to 4 bits/sample, then FLACentropy.ratectl_codecs
delta0123-zstd-8entropynono8-bit quantization, delta order 0/1/2/3 by variance, zstdentropy.delta_zstd
delta12-zstd-8entropynono8-bit quantization, delta order 1/2 by variance, zstdentropy.delta_zstd
delta0123-zstd-10entropynono10-bit quantization, delta order 0/1/2/3 by variance, zstdentropy.delta_zstd
delta0123-zstd-12entropynono12-bit quantization, delta order 0/1/2/3 by variance, zstdentropy.delta_zstd
fluxcode-10fluxcodedecimal datanofluxcode, B = 10: power-of-two or decimal step, delta order 0/1/2/3, bit or byte planes, zstdflux.adapters
fluxcode-12fluxcodedecimal datanofluxcode, B = 12: power-of-two or decimal step, delta order 0/1/2/3, bit or byte planes, zstdflux.adapters
fluxcode-16fluxcodedecimal datanofluxcode, B = 16: power-of-two or decimal step, delta order 0/1/2/3, bit or byte planes, zstdflux.adapters
fluxcode-16-f0.01fluxcodedecimal datanofluxcode, B = 16: power-of-two or decimal step, delta order 0/1/2/3, bit or byte planes, zstd, noise floor f = 0.01flux.adapters
fluxcode-16-f0.03fluxcodedecimal datanofluxcode, B = 16: power-of-two or decimal step, delta order 0/1/2/3, bit or byte planes, zstd, noise floor f = 0.03flux.adapters
fluxcode-16-f0.1fluxcodedecimal datanofluxcode, B = 16: power-of-two or decimal step, delta order 0/1/2/3, bit or byte planes, zstd, noise floor f = 0.1flux.adapters
fluxcode-16-f0.25fluxcodedecimal datanofluxcode, B = 16: power-of-two or decimal step, delta order 0/1/2/3, bit or byte planes, zstd, noise floor f = 0.25flux.adapters
fluxcode-16-f0.5fluxcodedecimal datanofluxcode, B = 16: power-of-two or decimal step, delta order 0/1/2/3, bit or byte planes, zstd, noise floor f = 0.5flux.adapters
fluxcode-16-f1fluxcodedecimal datanofluxcode, B = 16: power-of-two or decimal step, delta order 0/1/2/3, bit or byte planes, zstd, noise floor f = 1flux.adapters

Summary matrix

Codecs × datasets: bits per sample over whole one-minute block groups (everything the decoder reads), and the median block RMSE (% of block range). Shading is by RMSE: green < 0.03%, none 0.03–0.2%, yellow 0.2–1%, orange 1–5%, red ≥ 5%. Sizes differ, so read both numbers.

codeclinearquadraticsin-4.12hzsin-9.87hzsin-50.3hzgauss-spikesimpulsessquare-2.24hzrandom-walkchirpnoisy-sinesensor-0.1
pwlinear-minmax16-merged1.17 b/s
0.00511%		1.17 b/s
0.0058%		1.30 b/s
0.489%		1.48 b/s
2.28%		2.18 b/s
7.98%		2.18 b/s
2.23%		2.31 b/s
10.1%		1.23 b/s
10.2%		2.18 b/s
2.39%		2.05 b/s
23.3%		1.95 b/s
2.94%		1.98 b/s
0.676%
swingdoor-0.5%1.17 b/s
0.00504%		1.17 b/s
0.00578%		1.26 b/s
0.423%		2.59 b/s
0.444%		13.51 b/s
0.172%		7.13 b/s
0.315%		11.28 b/s
0.243%		1.25 b/s
0%		10.80 b/s
0.245%		11.14 b/s
0.266%		12.99 b/s
0.193%		4.53 b/s
0.342%
swingdoor-2%1.17 b/s
0.00504%		1.17 b/s
0.00578%		1.17 b/s
0.55%		1.40 b/s
1.73%		6.48 b/s
1.59%		1.76 b/s
0.92%		2.48 b/s
1.25%		1.25 b/s
0%		3.97 b/s
1.38%		7.16 b/s
1.37%		6.84 b/s
1.27%		1.17 b/s
0.807%
dct-top641.73 b/s
0.134%		1.73 b/s
0.0847%		1.73 b/s
0.167%		1.73 b/s
0.431%		1.73 b/s
2.38%		1.73 b/s
0.872%		1.73 b/s
4.6%		1.73 b/s
5.63%		1.73 b/s
2.22%		1.73 b/s
22.3%		1.73 b/s
2.09%		1.73 b/s
0.528%
pwlinear-minmax162.16 b/s
0.00516%		2.16 b/s
0.00584%		2.16 b/s
0.429%		2.16 b/s
2.08%		2.18 b/s
7.98%		2.28 b/s
2.03%		2.31 b/s
10.1%		1.23 b/s
10.2%		2.20 b/s
2.38%		2.19 b/s
23.3%		2.17 b/s
2.91%		2.18 b/s
0.667%
pchip-minmax162.16 b/s
0.0143%		2.16 b/s
0.0127%		2.16 b/s
0.116%		2.16 b/s
0.454%		2.18 b/s
3.12%		2.28 b/s
1.11%		2.31 b/s
10.9%		1.23 b/s
10.8%		2.20 b/s
2.71%		2.19 b/s
22.8%		2.17 b/s
3.92%		2.18 b/s
0.83%
bintree-center3.14 b/s
0.0563%		3.14 b/s
0.0563%		3.14 b/s
0.622%		3.14 b/s
1.33%		3.14 b/s
4.84%		3.14 b/s
0.861%		3.14 b/s
2.15%		3.14 b/s
2.37%		3.14 b/s
0.865%		3.14 b/s
4.98%		3.14 b/s
1.52%		3.14 b/s
0.275%
bintree-edge3.14 b/s
0.14%		3.14 b/s
0.139%		3.14 b/s
1.56%		3.14 b/s
3.2%		3.14 b/s
10.7%		3.14 b/s
2.07%		3.14 b/s
1.22%		3.14 b/s
0%		3.14 b/s
2%		3.14 b/s
10%		3.14 b/s
3.57%		3.14 b/s
0.652%
dpcm4-linear-deflate0.04 b/s
0%		0.43 b/s
0.00423%		1.01 b/s
0.0532%		1.01 b/s
0.128%		0.78 b/s
0.648%		2.56 b/s
0.298%		0.45 b/s
1.57%		0.04 b/s
0%		3.94 b/s
0.311%		1.46 b/s
1.27%		3.87 b/s
0.5%		2.17 b/s
0.093%
dpcm4-cubic-deflate0.04 b/s
0%		0.36 b/s
0.0096%		0.77 b/s
0.0987%		0.67 b/s
0.241%		0.46 b/s
1.24%		3.67 b/s
0.223%		1.84 b/s
0.857%		0.04 b/s
0%		4.44 b/s
0.302%		1.78 b/s
1.62%		4.34 b/s
0.509%		2.56 b/s
0.0874%
quant6-delta1-deflate0.09 b/s
0.458%		0.17 b/s
0.458%		0.29 b/s
0.449%		0.31 b/s
0.448%		0.31 b/s
0.447%		2.13 b/s
0.431%		2.74 b/s
0.45%		0.05 b/s
0%		3.47 b/s
0.458%		2.37 b/s
0.449%		4.02 b/s
0.457%		1.47 b/s
0.459%
dpcm6-linear-deflate0.04 b/s
0%		0.65 b/s
0.00095%		1.57 b/s
0.0121%		1.59 b/s
0.0288%		1.37 b/s
0.146%		4.98 b/s
0.068%		1.93 b/s
0.883%		0.04 b/s
0%		5.68 b/s
0.0698%		3.37 b/s
0.284%		5.71 b/s
0.113%		2.17 b/s
0.021%
dpcm6-cubic-deflate0.04 b/s
0%		0.51 b/s
0.0024%		1.35 b/s
0.0233%		1.29 b/s
0.0562%		0.78 b/s
0.291%		5.85 b/s
0.0502%		4.06 b/s
0.245%		0.04 b/s
0%		6.03 b/s
0.0689%		3.51 b/s
0.362%		6.04 b/s
0.115%		2.61 b/s
0.024%
dpcm6-mulaw-deflate0.04 b/s
0%		0.53 b/s
0.00218%		1.41 b/s
0.0208%		1.37 b/s
0.0498%		0.81 b/s
0.259%		5.74 b/s
0.0471%		3.44 b/s
0.33%		0.04 b/s
1.9e-14%		6.04 b/s
0.0601%		3.59 b/s
0.319%		6.06 b/s
0.101%		2.52 b/s
0.0187%
quant88.13 b/s
0.113%		8.13 b/s
0.113%		8.13 b/s
0.112%		8.13 b/s
0.113%		8.13 b/s
0.113%		8.13 b/s
0.113%		8.13 b/s
0.113%		8.13 b/s
0%		8.13 b/s
0.113%		8.13 b/s
0.112%		8.13 b/s
0.113%		8.13 b/s
0.114%
quant8-delta1-deflate0.10 b/s
0.113%		0.21 b/s
0.113%		0.48 b/s
0.112%		0.62 b/s
0.113%		0.56 b/s
0.113%		4.31 b/s
0.113%		5.01 b/s
0.113%		0.05 b/s
0%		5.33 b/s
0.113%		5.17 b/s
0.112%		5.73 b/s
0.113%		2.34 b/s
0.114%
gorilla-xor12.98 b/s
0%		50.47 b/s
0%		64.78 b/s
0%		65.44 b/s
0%		65.97 b/s
0%		65.98 b/s
0%		66.04 b/s
0%		1.10 b/s
0%		54.72 b/s
0%		65.23 b/s
0%		65.44 b/s
0%		24.86 b/s
0%
cw-delta1-tree-23.23 b/s
1.93e-05%		3.23 b/s
0.0148%		3.23 b/s
0.157%		3.23 b/s
0.383%		3.23 b/s
2.09%		3.23 b/s
0.504%		3.23 b/s
1.07%		3.23 b/s
0%		3.23 b/s
1.08%		3.23 b/s
3.21%		3.23 b/s
1.84%		3.23 b/s
0.173%
cw4-delta1-bfp-e44.42 b/s
0.00751%		4.42 b/s
0.00751%		4.42 b/s
0.0541%		4.42 b/s
0.199%		4.42 b/s
0.952%		4.42 b/s
0.187%		4.42 b/s
0.42%		4.42 b/s
0%		4.42 b/s
0.29%		4.42 b/s
1.2%		4.42 b/s
0.483%		4.42 b/s
0.0796%
cw-delta1-tree-45.23 b/s
5.89e-06%		5.23 b/s
0.00423%		5.23 b/s
0.0454%		5.23 b/s
0.109%		5.23 b/s
0.589%		5.23 b/s
0.1%		5.23 b/s
0.293%		5.23 b/s
1.73e-08%		5.23 b/s
0.163%		5.23 b/s
0.819%		5.23 b/s
0.276%		5.23 b/s
0.0186%
cw6-delta1-bfp-e46.42 b/s
0.00179%		6.42 b/s
0.00178%		6.42 b/s
0.0127%		6.42 b/s
0.0428%		6.42 b/s
0.228%		6.42 b/s
0.0421%		6.42 b/s
0.108%		6.42 b/s
0%		6.42 b/s
0.0628%		6.42 b/s
0.265%		6.42 b/s
0.108%		6.42 b/s
0.019%
cw8-delta1-linear8.18 b/s
5.89e-06%		8.18 b/s
0.000233%		8.18 b/s
0.00294%		8.18 b/s
0.00703%		8.18 b/s
0.0357%		8.18 b/s
0.0167%		8.18 b/s
0.225%		8.18 b/s
0%		8.18 b/s
0.0172%		8.18 b/s
0.0693%		8.18 b/s
0.0276%		8.18 b/s
0.00512%
cw8-delta1-sqrt8.15 b/s
0.0159%		8.15 b/s
0.0158%		8.15 b/s
0.0257%		8.15 b/s
0.0479%		8.15 b/s
0.101%		8.15 b/s
0.0324%		8.15 b/s
0.0399%		8.15 b/s
0.00411%		8.15 b/s
0.0435%		8.15 b/s
0.105%		8.15 b/s
0.0573%		8.15 b/s
0.0201%
cw8-delta1-bfp-e28.28 b/s
0.0141%		8.28 b/s
0.0141%		8.28 b/s
0.0142%		8.28 b/s
0.0142%		8.28 b/s
0.0565%		8.28 b/s
0.0158%		8.28 b/s
0.0286%		8.28 b/s
0%		8.28 b/s
0.0155%		8.28 b/s
0.0642%		8.28 b/s
0.0261%		8.28 b/s
0.0141%
cw8-delta1-bfp-e48.41 b/s
0.000442%		8.41 b/s
0.000437%		8.41 b/s
0.00311%		8.41 b/s
0.00698%		8.41 b/s
0.0565%		8.41 b/s
0.0101%		8.41 b/s
0.0271%		8.41 b/s
0%		8.41 b/s
0.0153%		8.41 b/s
0.064%		8.41 b/s
0.0261%		8.41 b/s
0.0047%
cw-delta1-tree-78.23 b/s
0.000515%		8.23 b/s
0.000469%		8.23 b/s
0.00505%		8.23 b/s
0.0121%		8.23 b/s
0.065%		8.23 b/s
0.0116%		8.23 b/s
0.0486%		8.23 b/s
1.73e-08%		8.23 b/s
0.019%		8.23 b/s
0.0912%		8.23 b/s
0.0327%		8.23 b/s
0.00702%
ratectl-delta0123-zstd-4b0.15 b/s
2.69e-08%		1.18 b/s
2.69e-08%		4.00 b/s
1.5e-05%		4.00 b/s
0.000207%		4.00 b/s
0.00537%		4.00 b/s
0.131%		4.00 b/s
0.201%		0.23 b/s
1.36e-08%		4.00 b/s
0.284%		4.00 b/s
0.145%		4.00 b/s
0.425%		4.00 b/s
0.00209%
ratectl-flac-4b2.45 b/s
2.69e-08%		2.96 b/s
2.69e-08%		4.00 b/s
4.05e-06%		3.99 b/s
8.17e-05%		3.98 b/s
0.000427%		4.00 b/s
0.119%		4.00 b/s
0.256%		1.28 b/s
0.00108%		4.00 b/s
0.215%		4.00 b/s
0.0602%		4.00 b/s
0.396%		3.96 b/s
0.0631%
delta0123-zstd-80.15 b/s
0.113%		0.34 b/s
0.113%		1.22 b/s
0.112%		0.86 b/s
0.113%		0.60 b/s
0.113%		4.40 b/s
0.113%		4.59 b/s
0.114%		0.07 b/s
0%		4.97 b/s
0.113%		4.55 b/s
0.112%		5.64 b/s
0.113%		2.51 b/s
0.114%
delta12-zstd-80.15 b/s
0.113%		0.34 b/s
0.113%		1.22 b/s
0.112%		0.86 b/s
0.113%		0.60 b/s
0.113%		4.39 b/s
0.113%		5.00 b/s
0.114%		0.07 b/s
0%		4.97 b/s
0.113%		4.95 b/s
0.112%		5.64 b/s
0.113%		2.51 b/s
0.114%
delta0123-zstd-100.12 b/s
0.0282%		0.55 b/s
0.0282%		1.33 b/s
0.0281%		1.43 b/s
0.0281%		1.27 b/s
0.0281%		6.21 b/s
0.0282%		5.96 b/s
0.0285%		0.07 b/s
0%		6.88 b/s
0.0281%		6.31 b/s
0.028%		7.55 b/s
0.0282%		2.71 b/s
0.0282%
delta0123-zstd-120.11 b/s
0.00705%		0.76 b/s
0.00704%		1.87 b/s
0.00702%		2.20 b/s
0.00702%		3.45 b/s
0.00704%		8.10 b/s
0.00704%		5.96 b/s
0.0285%		0.07 b/s
0%		7.91 b/s
0.0141%		6.31 b/s
0.028%		7.55 b/s
0.0282%		2.98 b/s
0.00707%
fluxcode-100.02 b/s
0%		0.44 b/s
0.0387%		0.98 b/s
0.0359%		1.41 b/s
0.0358%		0.92 b/s
0.036%		5.93 b/s
0.0285%		5.52 b/s
0.0554%		0.04 b/s
0%		6.30 b/s
0.0406%		4.67 b/s
0.0359%		7.24 b/s
0.0324%		1.77 b/s
0%
fluxcode-120.02 b/s
0%		0.60 b/s
0.00949%		1.70 b/s
0.00903%		1.94 b/s
0.009%		2.73 b/s
0.00897%		7.99 b/s
0.00714%		7.64 b/s
0.0139%		0.04 b/s
0%		8.54 b/s
0.0102%		5.54 b/s
0.00901%		9.61 b/s
0.00812%		1.77 b/s
0%
fluxcode-160.02 b/s
0%		0.25 b/s
0.000605%		2.40 b/s
0.000563%		3.42 b/s
0.000563%		6.76 b/s
0.000564%		12.04 b/s
0.000447%		11.64 b/s
0.000866%		0.04 b/s
0%		12.69 b/s
0.000634%		7.82 b/s
0.000564%		13.71 b/s
0.000506%		1.77 b/s
0%
fluxcode-16-f0.010.02 b/s
0%		0.25 b/s
0.000605%		2.40 b/s
0.000563%		3.42 b/s
0.000563%		6.76 b/s
0.000564%		11.01 b/s
0.000885%		10.10 b/s
0.00176%		0.04 b/s
0%		12.69 b/s
0.000634%		7.82 b/s
0.000564%		10.67 b/s
0.00405%		1.77 b/s
0%
fluxcode-16-f0.030.02 b/s
0%		0.25 b/s
0.000605%		2.40 b/s
0.000563%		3.42 b/s
0.000563%		6.76 b/s
0.000564%		9.27 b/s
0.00354%		8.12 b/s
0.00702%		0.04 b/s
0%		12.69 b/s
0.000634%		7.82 b/s
0.000564%		8.60 b/s
0.0162%		1.77 b/s
0%
fluxcode-16-f0.10.02 b/s
0%		0.25 b/s
0.000605%		2.40 b/s
0.000563%		3.42 b/s
0.000563%		6.76 b/s
0.000564%		7.45 b/s
0.0141%		5.57 b/s
0.028%		0.04 b/s
0%		12.69 b/s
0.000634%		7.82 b/s
0.000564%		6.90 b/s
0.0334%		1.77 b/s
0%
fluxcode-16-f0.250.02 b/s
0%		0.25 b/s
0.000605%		2.40 b/s
0.000563%		3.42 b/s
0.000563%		6.76 b/s
0.000564%		6.55 b/s
0.0283%		4.74 b/s
0.0561%		0.04 b/s
0%		12.69 b/s
0.000634%		7.82 b/s
0.000564%		5.24 b/s
0.13%		1.77 b/s
0%
fluxcode-16-f0.50.02 b/s
0%		0.25 b/s
0.000605%		2.40 b/s
0.000563%		3.42 b/s
0.000563%		6.76 b/s
0.000564%		5.60 b/s
0.0567%		4.08 b/s
0.112%		0.04 b/s
0%		12.69 b/s
0.000634%		7.82 b/s
0.000564%		4.27 b/s
0.259%		1.77 b/s
0%
fluxcode-16-f10.02 b/s
0%		0.25 b/s
0.000605%		2.40 b/s
0.000563%		3.42 b/s
0.000563%		6.76 b/s
0.000564%		4.67 b/s
0.113%		3.02 b/s
0.225%		0.04 b/s
0%		12.69 b/s
0.000634%		7.82 b/s
0.000564%		3.37 b/s
0.518%		1.77 b/s
0%

Size vs error

Every codec except the fluxcode variants, one marker per family (shade by position in the family). Bits/sample and RMSE are medians over the 12 datasets, so this is a summary; the per-dataset panels below show the spread. gorilla-xor (lossless, about 65 bits/sample) is left out of the charts; it's in the tables. fluxcode is the purple lines: B = 4..16 with the noise floor and the bits target off, since both only steer the quantization that B sets directly; one line per effort, darker for more effort, the default effort 4 heavy. The diamond is the all-defaults codec (fluxcode-16-f0.25: B = 16, noise floor 0.25): the noise detector coarsens the step on white-noise blocks, so it sits left of the B = 16 end of the line. The three lines nearly coincide: effort trades encode time for a few percent of size. At B = 16, summed over the datasets, effort 1 is +1.6% and effort 9 -3.6% against the default effort; the encode times behind each effort are in docs/TUNING.md (Effort). The other B and noise-floor variants of fluxcode are in the matrix and the per-dataset tables.

Bits per sample vs RMSE, all codecs

Per dataset

Each panel is one dataset: bits/sample over its 5 one-minute block groups, RMSE the median over their blocks. The lines are fluxcode at B = 4..16, noise floor and bits target off, at efforts 4 (heavy, the default), 1 and 9; the diamond is all defaults. A marker below the heavy line beats fluxcode at equal size. Lossless points (RMSE ≤ 10−7%) sit on the 0* row. The dotted line is 4 bits/sample. Errors are against the input, which for noisy-sine counts the noise as signal.

Bits per sample vs RMSE per signal kind

Per-dataset results

linear

Straight line, slope 1 per sample. 5 one-minute signals; the plots show second 0 of the first (seed 0), full scale 999. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged14654.8×0.00511% / 0.00511%-0.000171% / -0.000171%+0% / +0%+0% / +0%0.0121% / 0.0121%64 knots per block
swingdoor-0.5%14654.8×0.00504% / 0.00504%+0.000171% / +0.000171%+0% / +0%+0% / +0%0.0121% / 0.0121%64 knots per block
swingdoor-2%14654.8×0.00504% / 0.00504%+0.000171% / +0.000171%+0% / +0%+0% / +0%0.0121% / 0.0121%64 knots per block
dct-top6421637.0×0.134% / 0.229%+3.64e-13% / +2.91e-12%-0.643% / -1.18%+0.643% / +1.18%0.643% / 1.18%
pwlinear-minmax1627029.6×0.00516% / 0.00516%+0% / +2.18e-12%+0% / +0%+0% / +0%0.0121% / 0.0121%126 knots per block
pchip-minmax1627029.6×0.0143% / 0.0143%-0.0004% / -0.0004%+0% / +1.82e-13%+0% / +0%0.0718% / 0.0718%126 knots per block
bintree-center39220.4×0.0563% / 0.0563%+0% / +9.1e-14%-0.0488% / -0.0488%+0.0977% / +0.0977%0.0977% / 0.0977%codes 00/01/10/11 per block: 6/479/289/227
bintree-edge39220.4×0.14% / 0.14%-0.000391% / -0.000391%+0% / +0%+0% / +0%0.195% / 0.195%
dpcm4-linear-deflate51638.2×0% / 1.68e-09%+0% / +1.45e-09%+0% / +2.91e-09%+0% / +0%0% / 2.91e-09%
dpcm4-cubic-deflate51638.2×0% / 1.68e-09%+0% / +1.45e-09%+0% / +2.91e-09%+0% / +0%0% / 2.91e-09%
quant6-delta1-deflate11702.0×0.458% / 0.458%+0% / +2.18e-12%+0% / +0%+0% / +0%0.787% / 0.787%
dpcm6-linear-deflate51638.2×0% / 1.68e-09%+0% / +1.45e-09%+0% / +2.91e-09%+0% / +0%0% / 2.91e-09%
dpcm6-cubic-deflate51638.2×0% / 1.68e-09%+0% / +1.45e-09%+0% / +2.91e-09%+0% / +0%0% / 2.91e-09%
dpcm6-mulaw-deflate51638.2×0% / 1.68e-09%+0% / +1.45e-09%+0% / +2.91e-09%+0% / +0%0% / 2.91e-09%
quant810167.9×0.113% / 0.113%+0% / +2.18e-12%+0% / +0%+0% / +0%0.195% / 0.195%
quant8-delta1-deflate12650.1×0.113% / 0.113%+0% / +2.18e-12%+0% / +0%+0% / +0%0.195% / 0.195%
gorilla-xor16224.9×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 0, reuse window 990, new window 9
cw-delta1-tree-240419.8×1.93e-05% / 1.93e-05%+1.67e-05% / +1.67e-05%+3.35e-05% / +3.35e-05%+0% / +0%3.35e-05% / 3.35e-05%halving bits set 0%, leaves at code limit 100%
cw4-delta1-bfp-e455214.5×0.00751% / 0.00751%+0% / +2.18e-12%+0% / -1.14e-14%+0% / +0%0.013% / 0.013%frame exponents: mean 8.0, max 8
cw-delta1-tree-465412.2×5.89e-06% / 5.89e-06%+5.1e-06% / +5.1e-06%+1.02e-05% / +1.02e-05%+0% / +0%1.02e-05% / 1.02e-05%halving bits set 0%, leaves at code limit 100%
cw6-delta1-bfp-e480210.0×0.00179% / 0.00179%+0% / +2.18e-12%+0% / +0%+0% / +0%0.00307% / 0.00307%frame exponents: mean 8.0, max 8
cw8-delta1-linear10237.8×5.89e-06% / 5.89e-06%+5.1e-06% / +5.1e-06%+1.02e-05% / +1.02e-05%+0% / +0%1.02e-05% / 1.02e-05%
cw8-delta1-sqrt10197.9×0.0159% / 0.0159%-0.00979% / -0.00979%-0.0244% / -0.0244%+0% / +0%0.0244% / 0.0244%100% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0141% / 0.0141%+0% / +2.18e-12%+0% / +0%+0% / +0%0.0244% / 0.0244%frame exponents: mean 3.0, max 3
cw8-delta1-bfp-e410517.6×0.000442% / 0.000442%+0% / +2.18e-12%+0% / +0%+0% / +0%0.000764% / 0.000764%frame exponents: mean 8.0, max 8
cw-delta1-tree-710297.8×0.000515% / 0.000515%+0.000184% / +0.000184%-0.000602% / -0.000602%+0% / +0%0.000794% / 0.000794%halving bits set 0%, leaves at code limit 100%
ratectl-delta0123-zstd-4b19419.6×2.69e-08% / 2.69e-08%-7.37e-11% / -7.39e-11%-1.91e-08% / -1.91e-08%+0% / +0%4.66e-08% / 4.66e-08%
ratectl-flac-4b30626.1×2.69e-08% / 2.69e-08%-7.37e-11% / -7.39e-11%-1.91e-08% / -1.91e-08%+0% / +0%4.66e-08% / 4.66e-08%
delta0123-zstd-818437.1×0.113% / 0.113%+0% / +2.18e-12%+0% / +0%+0% / +0%0.195% / 0.195%orders 1: 300
delta12-zstd-818437.1×0.113% / 0.113%+0% / +2.18e-12%+0% / +0%+0% / +0%0.195% / 0.195%orders 1: 300
delta0123-zstd-1015546.3×0.0282% / 0.0282%+0% / +2.18e-12%+0% / +0%+0% / +0%0.0487% / 0.0487%orders 1: 300
delta0123-zstd-1214590.4×0.00705% / 0.00705%+0% / +2.18e-12%+0% / +0%+0% / +0%0.0121% / 0.0121%orders 1: 300
fluxcode-1032711.9×0% / 2.91e-12%+0% / -3e-12%+0% / -2.91e-12%+0% / -2.91e-12%0% / 2.91e-12%orders 1: 300; step 2^0
fluxcode-1233038.0×0% / 2.91e-12%+0% / -3e-12%+0% / -2.91e-12%+0% / -2.91e-12%0% / 2.91e-12%orders 1: 300; decimal grid 10^0
fluxcode-1633038.0×0% / 2.91e-12%+0% / -3e-12%+0% / -2.91e-12%+0% / -2.91e-12%0% / 2.91e-12%orders 1: 300; decimal grid 10^0
fluxcode-16-f0.0133038.0×0% / 2.91e-12%+0% / -3e-12%+0% / -2.91e-12%+0% / -2.91e-12%0% / 2.91e-12%orders 1: 300; decimal grid 10^0
fluxcode-16-f0.0333038.0×0% / 2.91e-12%+0% / -3e-12%+0% / -2.91e-12%+0% / -2.91e-12%0% / 2.91e-12%orders 1: 300; decimal grid 10^0
fluxcode-16-f0.133038.0×0% / 2.91e-12%+0% / -3e-12%+0% / -2.91e-12%+0% / -2.91e-12%0% / 2.91e-12%orders 1: 300; decimal grid 10^0
fluxcode-16-f0.2533038.0×0% / 2.91e-12%+0% / -3e-12%+0% / -2.91e-12%+0% / -2.91e-12%0% / 2.91e-12%orders 1: 300; decimal grid 10^0
fluxcode-16-f0.533038.0×0% / 2.91e-12%+0% / -3e-12%+0% / -2.91e-12%+0% / -2.91e-12%0% / 2.91e-12%orders 1: 300; decimal grid 10^0
fluxcode-16-f133038.0×0% / 2.91e-12%+0% / -3e-12%+0% / -2.91e-12%+0% / -2.91e-12%0% / 2.91e-12%orders 1: 300; decimal grid 10^0
linear: original, decoded traces and errors

quadratic

t²/1000 from the start of the minute. 5 one-minute signals; the plots show second 0 of the first (seed 1000), full scale 998. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged14654.8×0.0058% / 0.00835%+0.000182% / +0.00547%+0% / +0%+0% / +0%0.0121% / 0.0166%64 knots per block
swingdoor-0.5%14654.8×0.00578% / 0.00778%+0.000163% / +0.00464%+0% / +0%+0% / +0%0.0121% / 0.0183%64 knots per block
swingdoor-2%14654.8×0.00578% / 0.00778%+0.000163% / +0.00464%+0% / +0%+0% / +0%0.0121% / 0.0183%64 knots per block
dct-top6421637.0×0.0847% / 0.141%+1.03e-13% / +9.23e-13%-0.495% / -0.758%+0.339% / +0.667%0.495% / 0.758%
pwlinear-minmax1627029.6×0.00584% / 0.00791%+0.000203% / +0.00432%+0% / +0%+0% / +0%0.0122% / 0.0165%126 knots per block
pchip-minmax1627029.6×0.0127% / 0.0209%+0.000175% / +0.00189%+0% / +4.91e-13%+0% / +0%0.0424% / 0.0764%126 knots per block
bintree-center39220.4×0.0563% / 0.0784%+0.0004% / -0.00454%-0.0488% / -0.0977%+0.0977% / +0.0977%0.0977% / 0.195%codes 00/01/10/11 per block: 7/477/278/238
bintree-edge39220.4×0.139% / 0.197%+0.00103% / +0.0147%+0% / +0%+0% / +0%0.195% / 0.391%
dpcm4-linear-deflate53150.1×0.00423% / 0.00813%-5.84e-06% / +0.000325%+0.0014% / -0.00717%+0% / +0%0.00726% / 0.0143%
dpcm4-cubic-deflate45179.6×0.0096% / 0.0136%-0.00011% / +0.00121%-0.00173% / -0.0218%+0% / +0%0.0165% / 0.0325%
quant6-delta1-deflate22366.9×0.458% / 0.458%-0.00176% / -0.00881%+0% / +0%+0% / +0%0.793% / 0.794%
dpcm6-linear-deflate8199.2×0.00095% / 0.00189%-1.13e-06% / +8.94e-05%-0.000137% / -0.00187%+0% / +0%0.00164% / 0.00323%
dpcm6-cubic-deflate64125.6×0.0024% / 0.00315%-7.89e-06% / +0.000193%+0.000937% / -0.00556%+0% / +0%0.00414% / 0.00787%
dpcm6-mulaw-deflate66121.2×0.00218% / 0.00282%-2.39e-05% / -0.000129%+0.000431% / -0.00371%+0% / +0%0.00377% / 0.00731%
quant810167.9×0.113% / 0.114%+0.000746% / +0.00537%+0% / +0%+0% / +0%0.196% / 0.196%
quant8-delta1-deflate27298.4×0.113% / 0.114%+0.000746% / +0.00537%+0% / +0%+0% / +0%0.196% / 0.196%
gorilla-xor63091.3×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 0, reuse window 923, new window 76
cw-delta1-tree-240419.8×0.0148% / 0.0212%+0.000104% / +0.00249%+0.00799% / -0.03%+0% / +0%0.0254% / 0.0499%halving bits set 0%, leaves at code limit 93%
cw4-delta1-bfp-e455214.5×0.00751% / 0.00776%-2.31e-05% / +0.000497%+0% / +0.013%+0% / +0%0.013% / 0.0257%frame exponents: mean 8.0, max 13
cw-delta1-tree-465412.2×0.00423% / 0.00618%-8.93e-06% / +0.000347%+0.00141% / +0.00731%+0% / +0%0.00726% / 0.0143%halving bits set 0%, leaves at code limit 80%
cw6-delta1-bfp-e480210.0×0.00178% / 0.00186%-1.96e-06% / +0.00015%+0% / -0.00465%+0% / +0%0.0031% / 0.00574%frame exponents: mean 8.0, max 13
cw8-delta1-linear10237.8×0.000233% / 0.000461%-1.5e-06% / +2.25e-05%+4.84e-05% / -0.000451%+0% / +0%0.0004% / 0.000787%
cw8-delta1-sqrt10197.9×0.0158% / 0.0159%-0.00967% / -0.00985%-0.0244% / -0.0427%+0% / +0%0.0244% / 0.0427%100% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0141% / 0.015%-0.000133% / +0.00234%+0% / +0%+0% / +0%0.0245% / 0.0245%frame exponents: mean 3.0, max 3
cw8-delta1-bfp-e410517.6×0.000437% / 0.000454%-3.26e-07% / +3.75e-05%+0% / +0.000862%+0% / +0%0.000766% / 0.00145%frame exponents: mean 8.1, max 13
cw-delta1-tree-710297.8×0.000469% / 0.000687%-2.38e-06% / +5.48e-05%-0.000185% / -0.00126%+0% / +0%0.000806% / 0.00158%halving bits set 0%, leaves at code limit 24%
ratectl-delta0123-zstd-4b14854.1×2.69e-08% / 2.75e-08%-1.9e-10% / +2.12e-09%-1.91e-08% / -1.91e-08%+0% / +0%4.65e-08% / 4.66e-08%
ratectl-flac-4b37021.6×2.69e-08% / 2.75e-08%-1.9e-10% / +2.12e-09%-1.91e-08% / -1.91e-08%+0% / +0%4.65e-08% / 4.66e-08%
delta0123-zstd-843185.6×0.113% / 0.114%+0.000746% / +0.00537%+0% / +0%+0% / +0%0.196% / 0.196%orders 1: 300
delta12-zstd-843185.6×0.113% / 0.114%+0.000746% / +0.00537%+0% / +0%+0% / +0%0.196% / 0.196%orders 1: 300
delta0123-zstd-1068117.2×0.0282% / 0.0305%-0.000337% / +0.00533%+0% / +0%+0% / +0%0.0488% / 0.0489%orders 1: 295, 2: 5
delta0123-zstd-129583.9×0.00704% / 0.00733%-2.27e-05% / +0.000574%+0% / +0%+0% / +0%0.0122% / 0.0122%orders 1: 290, 2: 10
fluxcode-1055145.9×0.0387% / 0.0565%+0.000275% / +0.00398%-2.38e-05% / -0.079%+0% / -0.0866%0.0671% / 0.0971%orders 1: 300; step 2^0..2^7
fluxcode-1275107.3×0.00949% / 0.0138%+0.000132% / +0.000596%+0.000952% / +0.0222%+0% / -0.0229%0.0168% / 0.0243%orders 1: 300; step 2^-2..2^5
fluxcode-1632254.0×0.000605% / 0.000862%+0% / -8.54e-05%-1.67e-06% / -0.0001%+0% / -3.59e-13%0.00104% / 0.00152%orders 1: 280, 2: 20; step 2^-6..2^1
fluxcode-16-f0.0132254.0×0.000605% / 0.000862%+0% / -8.54e-05%-1.67e-06% / -0.0001%+0% / -3.59e-13%0.00104% / 0.00152%orders 1: 280, 2: 20; step 2^-6..2^1
fluxcode-16-f0.0332254.0×0.000605% / 0.000862%+0% / -8.54e-05%-1.67e-06% / -0.0001%+0% / -3.59e-13%0.00104% / 0.00152%orders 1: 280, 2: 20; step 2^-6..2^1
fluxcode-16-f0.132254.0×0.000605% / 0.000862%+0% / -8.54e-05%-1.67e-06% / -0.0001%+0% / -3.59e-13%0.00104% / 0.00152%orders 1: 280, 2: 20; step 2^-6..2^1
fluxcode-16-f0.2532254.0×0.000605% / 0.000862%+0% / -8.54e-05%-1.67e-06% / -0.0001%+0% / -3.59e-13%0.00104% / 0.00152%orders 1: 280, 2: 20; step 2^-6..2^1
fluxcode-16-f0.532254.0×0.000605% / 0.000862%+0% / -8.54e-05%-1.67e-06% / -0.0001%+0% / -3.59e-13%0.00104% / 0.00152%orders 1: 280, 2: 20; step 2^-6..2^1
fluxcode-16-f132254.0×0.000605% / 0.000862%+0% / -8.54e-05%-1.67e-06% / -0.0001%+0% / -3.59e-13%0.00104% / 0.00152%orders 1: 280, 2: 20; step 2^-6..2^1
quadratic: original, decoded traces and errors

sin-4.12hz

100·sin, √17 ≈ 4.123 Hz, random phase. 5 one-minute signals; the plots show second 0 of the first (seed 2000), full scale 200. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged16249.2×0.489% / 0.5%-0.000241% / -0.0262%+0% / +7.11e-15%+0% / +0%1.04% / 1.05%72 knots per block
swingdoor-0.5%15850.7×0.423% / 0.428%-0.0013% / -0.0164%-0.0244% / -0.0733%+0.0244% / +0.0733%0.727% / 0.743%70 knots per block
swingdoor-2%14654.8×0.55% / 0.558%-7.78e-05% / -0.0187%-0.0244% / -0.293%+0.0244% / +0.293%1.07% / 1.08%64 knots per block
dct-top6421637.0×0.167% / 0.19%+2.07e-05% / +0.00992%+0.0527% / +0.42%-0.0544% / -0.419%2.74% / 3.12%
pwlinear-minmax1627029.6×0.429% / 0.439%-0.000419% / +0.0244%+0% / +7.11e-15%+0% / +0%0.92% / 0.926%126 knots per block
pchip-minmax1627029.6×0.116% / 0.121%-0.000297% / +0.00969%+0% / +1.42e-14%+0% / +0%0.283% / 0.309%126 knots per block
bintree-center39220.4×0.622% / 0.649%+0.000665% / +0.0292%-0.391% / -0.391%+0.391% / +0.391%1.56% / 1.56%codes 00/01/10/11 per block: 57/356/294/294
bintree-edge39220.4×1.56% / 1.63%+0.000201% / -0.0764%+0% / +7.11e-15%+0% / +0%3.12% / 3.12%
dpcm4-linear-deflate12663.4×0.0532% / 0.0549%-7.52e-05% / -0.00519%-0.00501% / +0.0919%+0.00825% / +0.0924%0.0924% / 0.0925%
dpcm4-cubic-deflate9782.8×0.0987% / 0.104%-0.000416% / -0.0161%+0.00972% / +0.0199%-0.00919% / -0.0196%0.21% / 0.21%
quant6-delta1-deflate36222.5×0.449% / 0.45%-0.000144% / +0.014%+0% / -2.13e-14%+0% / +0%0.793% / 0.794%
dpcm6-linear-deflate19740.7×0.0121% / 0.0125%+1.59e-05% / -0.00117%+0.000611% / +0.0208%+0.00144% / +0.0209%0.0209% / 0.0209%
dpcm6-cubic-deflate16947.4×0.0233% / 0.0249%-3.46e-05% / +0.00227%+0.0022% / +0.00445%-0.00224% / -0.00447%0.0526% / 0.0527%
dpcm6-mulaw-deflate17745.3×0.0208% / 0.022%-9.34e-05% / +0.002%+0.00395% / +0.0068%-0.00376% / -0.00681%0.0479% / 0.048%
quant810167.9×0.112% / 0.113%+1.79e-05% / +0.00809%+0% / +7.11e-15%+0% / +0%0.196% / 0.196%
quant8-delta1-deflate60133.3×0.112% / 0.113%+1.79e-05% / +0.00809%+0% / +7.11e-15%+0% / +0%0.196% / 0.196%
gorilla-xor80971.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 0, reuse window 974, new window 25
cw-delta1-tree-240419.8×0.157% / 0.224%+0.000104% / +0.025%+0.00201% / +0.0277%-0.00154% / -0.0355%0.324% / 0.628%halving bits set 11%, leaves at code limit 52%
cw4-delta1-bfp-e455214.5×0.0541% / 0.0566%-6.57e-05% / -0.00515%+0.0196% / +0.0519%-0.0204% / -0.0514%0.104% / 0.104%frame exponents: mean 5.3, max 8
cw-delta1-tree-465412.2×0.0454% / 0.0479%+7.19e-05% / +0.00488%+0.000166% / -0.00574%+6.09e-05% / +0.00562%0.0924% / 0.0925%halving bits set 10%, leaves at code limit 26%
cw6-delta1-bfp-e480210.0×0.0127% / 0.0132%+3.02e-05% / -0.00124%+0.00375% / +0.0121%-0.0036% / -0.0124%0.0248% / 0.0248%frame exponents: mean 5.3, max 8
cw8-delta1-linear10237.8×0.00294% / 0.00306%+8.32e-06% / +0.000291%-0.000183% / -0.00504%-0.000485% / -0.00506%0.0051% / 0.0051%
cw8-delta1-sqrt10197.9×0.0257% / 0.0265%-2.42e-05% / +0.00186%-0.00531% / -0.0121%+0.00452% / +0.012%0.0488% / 0.0488%38% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0142% / 0.0146%+2.28e-05% / -0.00126%-0.000144% / -0.0243%-0.000144% / -0.0243%0.0245% / 0.0245%frame exponents: mean 3.0, max 3
cw8-delta1-bfp-e410517.6×0.00311% / 0.00322%-3.63e-06% / -0.00026%+0.000594% / +0.00303%-0.000424% / -0.00304%0.00612% / 0.00613%frame exponents: mean 5.4, max 8
cw-delta1-tree-710297.8×0.00505% / 0.00532%+2.3e-06% / -0.000537%+6.4e-05% / +0.000636%+5.2e-05% / -0.000642%0.0103% / 0.0103%halving bits set 10%, leaves at code limit 8%
ratectl-delta0123-zstd-4b49916.0×1.5e-05% / 1.55e-05%-6.88e-08% / +1.11e-06%-1.02e-05% / -2.32e-05%+0% / +0%2.59e-05% / 2.6e-05%
ratectl-flac-4b50016.0×4.05e-06% / 4.3e-06%-8.95e-09% / -3.7e-07%+2.09e-06% / +3.27e-06%+0% / +0%7.04e-06% / 7.17e-06%
delta0123-zstd-815252.6×0.112% / 0.113%+1.79e-05% / +0.00809%+0% / +0%+0% / +0%0.196% / 0.196%orders 2: 300
delta12-zstd-815252.6×0.112% / 0.113%+1.79e-05% / +0.00809%+0% / +0%+0% / +0%0.196% / 0.196%orders 2: 300
delta0123-zstd-1016748.0×0.0281% / 0.0289%-6.51e-05% / +0.00293%+0% / +0%+0% / +0%0.0488% / 0.0489%orders 2: 300
delta0123-zstd-1223434.2×0.00702% / 0.00728%+4.37e-06% / +0.000621%+0% / +0%+0% / +0%0.0122% / 0.0122%orders 2: 299, 3: 1
fluxcode-1012265.5×0.0359% / 0.0365%-5.15e-07% / -0.00338%+0.000144% / +0.000884%-0.000131% / -0.000904%0.0625% / 0.0625%orders 2: 300; step 2^-2
fluxcode-1221237.7×0.00903% / 0.00928%-1.93e-05% / -0.000717%+0.000144% / +0.000884%-0.000131% / -0.000904%0.0156% / 0.0156%orders 2: 300; step 2^-4
fluxcode-1630026.7×0.000563% / 0.000588%+9.55e-07% / +5.1e-05%+0.000144% / +0.000884%-0.000131% / -0.000904%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.0130026.7×0.000563% / 0.000588%+9.55e-07% / +5.1e-05%+0.000144% / +0.000884%-0.000131% / -0.000904%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.0330026.7×0.000563% / 0.000588%+9.55e-07% / +5.1e-05%+0.000144% / +0.000884%-0.000131% / -0.000904%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.130026.7×0.000563% / 0.000588%+9.55e-07% / +5.1e-05%+0.000144% / +0.000884%-0.000131% / -0.000904%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.2530026.7×0.000563% / 0.000588%+9.55e-07% / +5.1e-05%+0.000144% / +0.000884%-0.000131% / -0.000904%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.530026.7×0.000563% / 0.000588%+9.55e-07% / +5.1e-05%+0.000144% / +0.000884%-0.000131% / -0.000904%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f130026.7×0.000563% / 0.000588%+9.55e-07% / +5.1e-05%+0.000144% / +0.000884%-0.000131% / -0.000904%0.000976% / 0.000977%orders 3: 300; step 2^-8
sin-4.12hz: original, decoded traces and errors

sin-9.87hz

100·sin, π² ≈ 9.870 Hz, random phase. 5 one-minute signals; the plots show second 0 of the first (seed 3000), full scale 200. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged18543.1×2.28% / 2.36%-0.00274% / -0.12%+0% / +7.11e-15%+0% / +0%5.17% / 5.39%84 knots per block
swingdoor-0.5%32424.7×0.444% / 0.453%-0.000222% / +0.021%+0% / +7.11e-15%+0% / +0%0.784% / 0.84%153 knots per block
swingdoor-2%17545.6×1.73% / 1.76%-0.00331% / -0.0493%-0.0733% / -0.317%+0.0733% / +0.317%2.9% / 3.02%79 knots per block
dct-top6421637.0×0.431% / 0.487%-1.77e-05% / -0.0871%+0.21% / +1.66%-0.203% / -1.65%6.88% / 7.85%
pwlinear-minmax1627029.6×2.08% / 2.17%-0.00802% / -0.105%+0% / +7.11e-15%+0% / +0%4.74% / 4.82%126 knots per block
pchip-minmax1627029.6×0.454% / 0.5%+0.0029% / +0.0548%+0% / +7.11e-15%+0% / -7.11e-15%1.32% / 1.66%126 knots per block
bintree-center39220.4×1.33% / 1.38%+0.000362% / +0.0572%-0.781% / -0.781%+0.781% / +0.781%3.12% / 3.12%codes 00/01/10/11 per block: 107/281/307/307
bintree-edge39220.4×3.2% / 3.38%-0.00383% / +0.198%+0% / +7.11e-15%+0% / +0%6.24% / 6.25%
dpcm4-linear-deflate12663.5×0.128% / 0.131%-0.000383% / +0.00911%-0.0187% / -0.221%-0.0097% / +0.221%0.221% / 0.221%
dpcm4-cubic-deflate8396.2×0.241% / 0.255%-1.07e-05% / +0.0213%+0.0487% / +0.0688%-0.047% / -0.0684%0.503% / 0.503%
quant6-delta1-deflate39203.3×0.448% / 0.453%-0.000248% / +0.0116%+0% / +2.13e-14%+0% / +0%0.793% / 0.794%
dpcm6-linear-deflate19940.2×0.0288% / 0.0304%+8.83e-06% / -0.00361%+9.6e-05% / -0.0499%+0.000399% / -0.0494%0.05% / 0.05%
dpcm6-cubic-deflate16149.7×0.0562% / 0.058%+3.34e-05% / +0.0069%+0.00728% / +0.0138%-0.00763% / -0.0136%0.126% / 0.126%
dpcm6-mulaw-deflate17146.7×0.0498% / 0.0522%+7.71e-05% / +0.00664%+0.0122% / +0.0187%-0.0117% / -0.0181%0.115% / 0.115%
quant810167.9×0.113% / 0.114%-0.000185% / +0.00497%+0% / +2.13e-14%+0% / +0%0.196% / 0.196%
quant8-delta1-deflate78103.0×0.113% / 0.114%-0.000185% / +0.00497%+0% / +2.13e-14%+0% / +0%0.196% / 0.196%
gorilla-xor81791.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 0, reuse window 980, new window 19
cw-delta1-tree-240419.8×0.383% / 0.554%-0.00185% / +0.0376%+0.0169% / +0.159%-0.0123% / -0.312%0.775% / 1.49%halving bits set 15%, leaves at code limit 51%
cw4-delta1-bfp-e455214.5×0.199% / 0.209%+0.000869% / -0.0157%+0.0259% / +0.375%-0.0109% / -0.408%0.416% / 0.417%frame exponents: mean 3.4, max 6
cw-delta1-tree-465412.2×0.109% / 0.117%-0.000187% / +0.00919%-0.00012% / -0.0274%+0.00152% / +0.0277%0.221% / 0.221%halving bits set 14%, leaves at code limit 25%
cw6-delta1-bfp-e480210.0×0.0428% / 0.0456%-0.000107% / +0.00443%+0.0169% / +0.0496%-0.0148% / +0.0496%0.099% / 0.0992%frame exponents: mean 3.6, max 6
cw8-delta1-linear10237.8×0.00703% / 0.00723%-2.46e-05% / -0.000599%-0.00104% / +0.0122%+0.00139% / +0.0122%0.0122% / 0.0122%
cw8-delta1-sqrt10197.9×0.0479% / 0.0496%+4.46e-06% / +0.00311%-0.00443% / -0.0235%+0.00692% / +0.0241%0.0976% / 0.0977%15% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0142% / 0.0147%+3.28e-05% / +0.00138%+0.000479% / -0.0245%+0.000479% / -0.0245%0.0245% / 0.0245%frame exponents: mean 3.0, max 3
cw8-delta1-bfp-e410517.6×0.00698% / 0.00725%-4.77e-06% / +0.00076%+0.00236% / -0.0122%-0.00312% / +0.0122%0.0122% / 0.0123%frame exponents: mean 4.0, max 6
cw-delta1-tree-710297.8×0.0121% / 0.0127%+6.75e-06% / +0.00157%-0.000191% / +0.00306%+0.000317% / -0.00306%0.0246% / 0.0246%halving bits set 14%, leaves at code limit 8%
ratectl-delta0123-zstd-4b50016.0×0.000207% / 0.000219%-6.44e-07% / -1.81e-05%+2.24e-05% / +0.000282%+0% / +0%0.000359% / 0.00036%
ratectl-flac-4b49816.1×8.17e-05% / 9.24e-05%+1.11e-08% / +7.86e-06%+0.000106% / +0.00015%+0% / +0%0.000142% / 0.000157%
delta0123-zstd-810874.1×0.113% / 0.114%-0.000185% / +0.00497%+0% / +0%+0% / +0%0.196% / 0.196%orders 2: 300
delta12-zstd-810874.1×0.113% / 0.114%-0.000185% / +0.00497%+0% / +0%+0% / +0%0.196% / 0.196%orders 2: 300
delta0123-zstd-1017944.7×0.0281% / 0.0285%+5.99e-05% / +0.00322%+0% / +0%+0% / +0%0.0488% / 0.0489%orders 3: 300
delta0123-zstd-1227529.1×0.00702% / 0.00727%-1.62e-05% / -0.000605%+0% / +0%+0% / +0%0.0122% / 0.0122%orders 3: 300
fluxcode-1017745.3×0.0358% / 0.0364%-9.08e-05% / -0.0023%+0.000168% / +0.00148%-0.000175% / -0.0018%0.0625% / 0.0625%orders 2: 106, 3: 194; step 2^-2
fluxcode-1224233.0×0.009% / 0.00922%+2.65e-06% / +0.000706%+0.000168% / +0.00148%-0.000175% / -0.0018%0.0156% / 0.0156%orders 3: 300; step 2^-4
fluxcode-1642718.7×0.000563% / 0.000587%+8.5e-07% / -5.16e-05%+3.21e-05% / -0.000972%-3.23e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.0142718.7×0.000563% / 0.000587%+8.5e-07% / -5.16e-05%+3.21e-05% / -0.000972%-3.23e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.0342718.7×0.000563% / 0.000587%+8.5e-07% / -5.16e-05%+3.21e-05% / -0.000972%-3.23e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.142718.7×0.000563% / 0.000587%+8.5e-07% / -5.16e-05%+3.21e-05% / -0.000972%-3.23e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.2542718.7×0.000563% / 0.000587%+8.5e-07% / -5.16e-05%+3.21e-05% / -0.000972%-3.23e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.542718.7×0.000563% / 0.000587%+8.5e-07% / -5.16e-05%+3.21e-05% / -0.000972%-3.23e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f142718.7×0.000563% / 0.000587%+8.5e-07% / -5.16e-05%+3.21e-05% / -0.000972%-3.23e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
sin-9.87hz: original, decoded traces and errors

sin-50.3hz

100·sin, 16π ≈ 50.27 Hz (~19.9 samples/cycle), random phase. 5 one-minute signals; the plots show second 0 of the first (seed 4000), full scale 200. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged27229.4×7.98% / 8.11%-0.000909% / -0.131%+0% / +7.11e-15%+0% / +0%17.2% / 17.8%127 knots per block
swingdoor-0.5%16894.7×0.172% / 0.178%+8.39e-05% / -0.00224%+0% / +7.11e-15%+0% / +0%0.815% / 0.842%836 knots per block
swingdoor-2%8119.9×1.59% / 1.62%-0.000509% / +0.0204%+0% / +7.11e-15%+0% / +0%3.27% / 3.34%396 knots per block
dct-top6421637.0×2.38% / 2.66%+0.0072% / +0.235%+4.84% / +9.48%-4.93% / -9.27%28.2% / 33.4%
pwlinear-minmax1627229.4×7.98% / 8.11%-0.000909% / -0.131%+0% / +7.11e-15%+0% / +0%17.2% / 17.8%127 knots per block
pchip-minmax1627229.4×3.12% / 3.86%-0.00611% / -0.178%+0% / +2.13e-14%+0% / +0%13.2% / 36.2%127 knots per block
bintree-center39220.4×4.84% / 4.95%-0.00319% / -0.095%-3.13% / -3.13%+3.12% / +3.13%12.5% / 12.5%codes 00/01/10/11 per block: 349/123/264/264
bintree-edge39220.4×10.7% / 11.1%+0.00277% / +0.194%+0% / +7.11e-15%+0% / +0%25% / 25%
dpcm4-linear-deflate9881.8×0.648% / 0.667%+0.000152% / -0.0786%-0.0837% / +1.12%+0.151% / -1.12%1.12% / 1.12%
dpcm4-cubic-deflate58138.2×1.24% / 1.27%-0.00365% / -0.101%+0.713% / +1.11%-0.699% / -1.04%2.55% / 2.55%
quant6-delta1-deflate39205.7×0.447% / 0.459%-0.000149% / +0.0157%+0% / -2.13e-14%+0% / +0%0.793% / 0.794%
dpcm6-linear-deflate17146.7×0.146% / 0.154%-6.68e-05% / -0.0159%+0.00257% / -0.253%+0.0113% / +0.253%0.253% / 0.254%
dpcm6-cubic-deflate9882.0×0.291% / 0.298%-0.00041% / -0.0224%+0.128% / +0.166%-0.129% / -0.165%0.638% / 0.64%
dpcm6-mulaw-deflate10179.1×0.259% / 0.271%-0.000367% / -0.024%+0.12% / +0.164%-0.118% / -0.162%0.581% / 0.583%
quant810167.9×0.113% / 0.116%+5.97e-05% / +0.00537%+0% / +2.13e-14%+0% / +0%0.196% / 0.196%
quant8-delta1-deflate70114.0×0.113% / 0.116%+5.97e-05% / +0.00537%+0% / +2.13e-14%+0% / +0%0.196% / 0.196%
gorilla-xor82461.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 0, reuse window 992, new window 7
cw-delta1-tree-240419.8×2.09% / 3.11%-0.000673% / -0.305%+1.11% / +3.81%-1.46% / -3.92%5.3% / 6.94%halving bits set 14%, leaves at code limit 50%
cw4-delta1-bfp-e455214.5×0.952% / 1.01%+0.0024% / +0.106%+0.0491% / -1.66%+0.0491% / -1.66%1.67% / 1.67%frame exponents: mean 1.0, max 1
cw-delta1-tree-465412.2×0.589% / 0.615%-0.000685% / -0.0709%+0.427% / +0.987%-0.383% / -0.953%1.12% / 1.12%halving bits set 11%, leaves at code limit 24%
cw6-delta1-bfp-e480210.0×0.228% / 0.248%-0.0011% / -0.0391%-0.0176% / +0.396%-0.0176% / +0.396%0.397% / 0.397%frame exponents: mean 1.0, max 1
cw8-delta1-linear10237.8×0.0357% / 0.037%+6.69e-07% / -0.00358%+0.00336% / -0.0619%-0.000279% / +0.0616%0.0619% / 0.0619%
cw8-delta1-sqrt10197.9×0.101% / 0.105%-0.000149% / -0.00748%-0.0317% / -0.0744%+0.0342% / +0.086%0.195% / 0.195%3% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0565% / 0.0586%-0.000193% / -0.0061%-0.00178% / +0.0976%-0.00178% / +0.0976%0.098% / 0.098%frame exponents: mean 1.0, max 1
cw8-delta1-bfp-e410517.6×0.0565% / 0.0586%-0.000193% / -0.0061%-0.00178% / +0.0976%-0.00178% / +0.0976%0.098% / 0.098%frame exponents: mean 1.0, max 1
cw-delta1-tree-710297.8×0.065% / 0.0679%-6.63e-05% / -0.0057%+0.00299% / +0.0621%-0.00252% / +0.0622%0.125% / 0.125%halving bits set 11%, leaves at code limit 8%
ratectl-delta0123-zstd-4b50016.0×0.00537% / 0.00558%-1.11e-05% / -0.000532%+0.00111% / -0.00854%+0% / +0%0.0093% / 0.00935%
ratectl-flac-4b49816.1×0.000427% / 0.000462%+8.13e-07% / +3.88e-05%-9.61e-05% / +0.000651%+0% / +0%0.000749% / 0.000767%
delta0123-zstd-876105.9×0.113% / 0.116%+5.97e-05% / +0.00537%+0% / +0%+0% / +0%0.196% / 0.196%orders 3: 300
delta12-zstd-875107.0×0.113% / 0.116%+5.97e-05% / +0.00537%+0% / +0%+0% / +0%0.196% / 0.196%orders 2: 300
delta0123-zstd-1015850.6×0.0281% / 0.0286%-0.000239% / -0.00402%+0% / +0%+0% / +0%0.0488% / 0.0489%orders 3: 300
delta0123-zstd-1243118.6×0.00704% / 0.00724%-1.67e-06% / +0.000571%+0% / +0%+0% / +0%0.0122% / 0.0122%orders 3: 300
fluxcode-1011569.5×0.036% / 0.0366%-3.58e-05% / +0.0039%+0.000133% / +0.00111%-0.000131% / -0.00111%0.0625% / 0.0625%orders 3: 300; step 2^-2
fluxcode-1234123.4×0.00897% / 0.00927%+1.99e-05% / +0.000793%+0.000133% / +0.00111%-0.000131% / -0.00111%0.0156% / 0.0156%orders 3: 300; step 2^-4
fluxcode-168459.5×0.000564% / 0.000592%-6.14e-07% / -4.88e-05%+9.7e-05% / +0.000971%-9.81e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.018459.5×0.000564% / 0.000592%-6.14e-07% / -4.88e-05%+9.7e-05% / +0.000971%-9.81e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.038459.5×0.000564% / 0.000592%-6.14e-07% / -4.88e-05%+9.7e-05% / +0.000971%-9.81e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.18459.5×0.000564% / 0.000592%-6.14e-07% / -4.88e-05%+9.7e-05% / +0.000971%-9.81e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.258459.5×0.000564% / 0.000592%-6.14e-07% / -4.88e-05%+9.7e-05% / +0.000971%-9.81e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.58459.5×0.000564% / 0.000592%-6.14e-07% / -4.88e-05%+9.7e-05% / +0.000971%-9.81e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f18459.5×0.000564% / 0.000592%-6.14e-07% / -4.88e-05%+9.7e-05% / +0.000971%-9.81e-05% / +0.000974%0.000976% / 0.000977%orders 3: 300; step 2^-8
sin-50.3hz: original, decoded traces and errors

gauss-spikes

Uniform noise ±10 + Gaussian spikes (σ = 10 samples, peak 1000), 180 per minute at random times. 5 one-minute signals; the plots show second 32 of the first (seed 5000), full scale 1995. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		peak shift (samples, plotted second)notes
pwlinear-minmax16-merged27329.3×2.23% / 38.9%+0.235% / +3%+0% / +2.2e-14%+0% / +0%12.1% / 97.6%+0, +0, +0, +0, +0, +0127 knots per block
swingdoor-0.5%8919.0×0.315% / 0.347%+0.00206% / +0.0529%+0% / -0.659%+0% / +0.0488%0.879% / 0.965%+0, -1, +0, +0, +0, +1437 knots per block
swingdoor-2%22036.4×0.92% / 1.37%+0.106% / +0.32%-0.0855% / -2.71%+0.0244% / +0.293%3.18% / 3.84%+0, -2, +0, +2, +2, -2101 knots per block
dct-top6421637.0×0.872% / 24.2%+0% / -1.58%-1.57% / -44.6%-0.76% / -6.04%3.27% / 78.4%+2, +0, +0, +0, +0, +0
pwlinear-minmax1628528.1×2.03% / 38.9%+0.178% / +3%+0% / +2.2e-14%+0% / +0%11.2% / 97.6%+0, +0, +0, +0, +0, +0133 knots per block
pchip-minmax1628528.1×1.11% / 41.5%+0.129% / +2.97%+0% / +2.2e-14%+0% / +0%6.42% / 99.4%+0, +0, +0, +0, +0, +0133 knots per block
bintree-center39220.4×0.861% / 10.6%+0.0645% / +1.48%-1.56% / -6.25%+0.195% / +6.25%6.09% / 25%+0, -1, +0, -1, -1, -1codes 00/01/10/11 per block: 291/155/484/71
bintree-edge39220.4×2.07% / 19.4%+0.0199% / +1.18%+0% / +1.12e-14%+0% / +0%12.3% / 50%+0, -1, +0, -3, -3, -1
dpcm4-linear-deflate32125.0×0.298% / 4.14%-0.000699% / +0.183%+0.036% / -5.39%+0.0533% / +6.81%0.523% / 7.12%+0, -1, +0, +0, +0, +0
dpcm4-cubic-deflate45917.4×0.223% / 5.38%+0.000744% / -0.288%+0.00224% / +14.3%-0.329% / -15.2%1.16% / 16.1%+0, +0, +0, +0, +0, +0
quant6-delta1-deflate26630.1×0.431% / 0.521%-0.0319% / -0.219%+0% / -1.18e-14%+0% / +0%0.793% / 0.794%+0, -1, +0, +0, +0, -1
dpcm6-linear-deflate62212.9×0.068% / 0.921%+4.77e-05% / -0.0309%+0.00189% / -1.28%+0.00434% / +1.47%0.118% / 1.61%+0, +0, +0, +0, +0, +0
dpcm6-cubic-deflate73110.9×0.0502% / 1.2%-1.25e-05% / +0.0971%+0.00468% / +3.06%-0.0566% / -2.93%0.267% / 3.9%+0, +0, +0, +0, +0, +0
dpcm6-mulaw-deflate71811.1×0.0471% / 1.09%+6.99e-05% / -0.0392%+0.00124% / +2.47%-0.0527% / -3.38%0.235% / 3.69%+0, +0, +0, +0, +0, +0
quant810167.9×0.113% / 0.117%-0.0009% / -0.0192%+0% / +1.12e-14%+0% / +0%0.196% / 0.196%+0, -1, +0, +0, +0, +0
quant8-delta1-deflate53914.8×0.113% / 0.117%-0.0009% / -0.0192%+0% / +1.12e-14%+0% / +0%0.196% / 0.196%+0, -1, +0, +0, +0, +0
gorilla-xor82471.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%+0, +0, +0, +0, +0, +0per block: same 0, reuse window 994, new window 5
cw-delta1-tree-240419.8×0.504% / 22%+0.000141% / -1.07%+0.114% / +47.1%-0.854% / -45.7%1.82% / 74.8%+0, +0, +0, +0, +0, -1halving bits set 29%, leaves at code limit 43%
cw4-delta1-bfp-e455214.5×0.187% / 1.96%-8.99e-05% / +0.124%+0.227% / +3.15%-0.082% / +3.15%0.827% / 3.33%+0, -1, +0, +0, +0, -1frame exponents: mean 3.9, max 6
cw-delta1-tree-465412.2×0.1% / 2.85%+0.000273% / +0.0949%+0.00423% / +5.9%-0.0824% / -5.69%0.517% / 6.76%+0, +0, +0, +0, +0, +0halving bits set 27%, leaves at code limit 4%
cw6-delta1-bfp-e480210.0×0.0421% / 0.461%+3.26e-06% / +0.0205%+0.0164% / -0.685%-0.0159% / -0.685%0.197% / 0.794%+0, +0, +0, +0, +0, +0frame exponents: mean 4.1, max 7
cw8-delta1-linear10237.8×0.0167% / 0.229%-6.34e-05% / -0.0186%+0.000683% / -0.255%+0.000501% / +0.343%0.0288% / 0.392%+0, +0, +0, +0, +0, +0
cw8-delta1-sqrt10197.9×0.0324% / 0.184%+0.000602% / -0.0103%-0.0373% / -0.286%+0.00529% / +0.103%0.181% / 0.728%+0, +0, +0, +0, +0, +054% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0158% / 0.114%+1.51e-05% / +0.0058%+0.00115% / +0.124%+9.7e-06% / +0.124%0.0486% / 0.196%+0, +0, +0, +0, +0, +0frame exponents: mean 2.8, max 3
cw8-delta1-bfp-e410517.6×0.0101% / 0.114%+1.74e-05% / +0.0058%+0.000854% / +0.124%-0.00266% / +0.124%0.0486% / 0.196%+0, +0, +0, +0, +0, +0frame exponents: mean 4.2, max 7
cw-delta1-tree-710297.8×0.0116% / 0.351%-2.05e-05% / +0.0216%+0.00167% / +0.746%-0.00639% / -0.636%0.0573% / 0.778%+0, +0, +0, +0, +0, +0halving bits set 27%, leaves at code limit 1%
ratectl-delta0123-zstd-4b50016.0×0.131% / 0.168%-0.0063% / -0.0268%-0.00538% / -0.283%+0% / +0%0.232% / 0.283%+0, +0, +0, +0, +0, +0
ratectl-flac-4b50016.0×0.119% / 0.154%-0.00207% / -0.0206%-0.183% / -0.221%+0% / +0%0.204% / 0.256%+0, -1, +0, +0, +0, +0
delta0123-zstd-855014.5×0.113% / 0.232%-0.000998% / -0.0192%+0% / +0%+0% / +0%0.196% / 0.394%+0, -1, +0, +0, +0, +0orders 0: 9, 1: 52, 2: 239; B lowered by the cap on 8 blocks (to 7)
delta12-zstd-854914.6×0.113% / 0.232%-0.000998% / -0.0192%+0% / +0%+0% / +0%0.196% / 0.394%+0, -1, +0, +0, +0, +0orders 1: 61, 2: 239; B lowered by the cap on 8 blocks (to 7)
delta0123-zstd-1077710.3×0.0282% / 0.232%-0.000221% / +0.0144%+0% / +0%+0% / +0%0.0488% / 0.394%+0, +0, +0, +0, +0, +0orders 0: 9, 1: 49, 2: 242; B lowered by the cap on 9 blocks (to 7)
delta0123-zstd-1210137.9×0.00704% / 0.232%+4.94e-06% / +0.0144%+0% / +0%+0% / +0%0.0122% / 0.394%+0, +0, +0, +0, +0, +0orders 0: 9, 1: 49, 2: 242; B lowered by the cap on 13 blocks (to 7)
fluxcode-1074110.8×0.0285% / 0.0576%+1.3e-05% / -0.0028%-0.00228% / -0.0914%-0.00186% / +0.0801%0.0492% / 0.0975%+0, +0, +0, +0, +0, +0orders 0: 108, 1: 7, 2: 185; step 2^-5..2^2
fluxcode-129998.0×0.00714% / 0.0144%-2.87e-06% / +0.000714%+0.000449% / +0.0223%-0.00178% / +0.0194%0.0123% / 0.0244%+0, +0, +0, +0, +0, +0orders 0: 108, 1: 9, 2: 183; step 2^-7..2^0
fluxcode-1615055.3×0.000447% / 0.000895%-6.04e-07% / -5.8e-05%-5.78e-05% / -0.00144%-0.000103% / +0.00149%0.000769% / 0.00152%+0, +0, +0, +0, +0, +0orders 0: 108, 1: 9, 2: 183; step 2^-11..2^-4
fluxcode-16-f0.0113775.8×0.000885% / 0.0913%-7.63e-07% / -0.00494%-4.83e-05% / +0.103%-0.000229% / -0.145%0.00154% / 0.157%+0, +0, +0, +0, +0, +0orders 0: 108, 1: 9, 2: 183; step 2^-6..2^-4 (noise floor applied on 236 blocks)
fluxcode-16-f0.0311596.9×0.00354% / 0.185%+4.79e-06% / -0.00337%+0.000331% / +0.168%-0.00126% / -0.288%0.00614% / 0.313%+0, +0, +0, +0, +0, +0orders 0: 108, 1: 9, 2: 183; step 2^-6..2^-3 (noise floor applied on 274 blocks)
fluxcode-16-f0.19318.6×0.0141% / 0.74%+1.09e-05% / +0.043%+0.00153% / +0.168%-0.00166% / -0.288%0.0246% / 1.25%+0, +0, +0, +0, +0, +0orders 0: 108, 1: 8, 2: 184; step 2^-6..2^-1 (noise floor applied on 274 blocks)
fluxcode-16-f0.258199.8×0.0283% / 0.74%-3.57e-06% / +0.043%-0.00031% / -0.652%-0.00166% / -0.288%0.0491% / 1.25%+0, +0, +0, +0, +0, +0orders 0: 108, 1: 7, 2: 185; step 2^-6..2^0 (noise floor applied on 274 blocks)
fluxcode-16-f0.570011.4×0.0567% / 0.74%+1.2e-05% / +0.043%+0.00482% / -0.652%-0.00166% / -0.288%0.0982% / 1.25%+0, +0, +0, +0, +0, +0orders 0: 108, 1: 8, 2: 184; step 2^-6..2^1 (noise floor applied on 274 blocks)
fluxcode-16-f158313.7×0.113% / 0.74%+0.000338% / +0.043%-0.000711% / -0.652%+0.194% / +0.429%0.196% / 1.25%+0, +0, +0, +0, +0, +0orders 0: 108, 1: 7, 2: 185; step 2^-6..2^2 (noise floor applied on 274 blocks)
gauss-spikes: original, decoded traces and errors

impulses

Uniform noise ±10 + single-sample impulses of +500, 180 per minute at random samples. 5 one-minute signals; the plots show second 56 of the first (seed 6000), full scale 519.7. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		peak shift (samples, plotted second)notes
pwlinear-minmax16-merged28927.7×10.1% / 40.3%+1.54% / +6.16%+0% / -1.11e-14%+0% / +0%90% / 99.3%+0, +0135 knots per block
swingdoor-0.5%14115.7×0.243% / 0.265%+0.000904% / +0.0259%+0% / -1.11e-14%+0% / +0.0244%0.838% / 0.93%+0, +0696 knots per block
swingdoor-2%31025.8×1.25% / 1.32%-0.00181% / -0.226%+0% / -1.11e-14%+0.0244% / +0.195%3.36% / 3.73%+0, +0146 knots per block
dct-top6421637.0×4.6% / 24.9%-0.125% / +1.4%-70.4% / -83%-3.56% / -17%71.1% / 83%+0, +0
pwlinear-minmax1628927.7×10.1% / 40.3%+1.54% / +6.16%+0% / -1.11e-14%+0% / +0%90% / 99.3%+0, +0135 knots per block
pchip-minmax1628927.7×10.9% / 42.6%+1.55% / +6.23%+0% / -1.11e-14%+0% / -3.44e-16%96% / 99.5%+0, +0135 knots per block
bintree-center39220.4×2.15% / 10.8%+0.317% / -0.674%-25% / -25%+0.391% / +6.25%25% / 25%+0, +0codes 00/01/10/11 per block: 419/137/404/41
bintree-edge39220.4×1.22% / 19.8%+0.0309% / -0.924%+0% / -1.11e-14%+0% / +0%3.83% / 50%+0, +0
dpcm4-linear-deflate56142.2×1.57% / 4.21%+0.00144% / -2.01%+0.806% / -5.84%+1.9% / -6.61%3.49% / 7.07%+0, +0
dpcm4-cubic-deflate23034.7×0.857% / 5.44%+0.00804% / +0.294%+1.98% / +15.7%+0.181% / -14.4%3.44% / 15.9%+0, +0
quant6-delta1-deflate34223.4×0.45% / 0.473%-0.0621% / -0.111%+0% / -1.11e-14%+0% / +0%0.793% / 0.794%+0, +0
dpcm6-linear-deflate24233.1×0.883% / 1.06%+0.00267% / +0.249%-0.0635% / +1.59%+0.148% / +1.59%1.6% / 1.61%+0, +0
dpcm6-cubic-deflate50815.7×0.245% / 1.21%+0.00228% / +0.0562%+1.54% / +3.67%-0.304% / -3.42%3.19% / 4.04%+0, +0
dpcm6-mulaw-deflate43018.6×0.33% / 1.07%+0.0016% / +0.0925%+1.37% / +3.62%-0.474% / -3.31%2.97% / 3.69%+0, +0
quant810167.9×0.113% / 0.118%-0.000516% / +0.0104%+0% / -1.11e-14%+0% / +0%0.196% / 0.196%+0, +0
quant8-delta1-deflate62612.8×0.113% / 0.118%-0.000516% / +0.0104%+0% / -1.11e-14%+0% / +0%0.196% / 0.196%+0, +0
gorilla-xor82551.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%+0, +0per block: same 0, reuse window 996, new window 3
cw-delta1-tree-240419.8×1.07% / 21.7%+0.0182% / -1.95%+1.9% / -50.6%-2.77% / -49.5%5.07% / 74.6%+0, +0halving bits set 54%, leaves at code limit 41%
cw4-delta1-bfp-e455214.5×0.42% / 1.99%-0.00089% / -0.129%+1.8% / -3.33%-0.282% / -3.29%3.23% / 3.33%+0, +0frame exponents: mean 2.9, max 5
cw-delta1-tree-465412.2×0.293% / 2.84%-0.00231% / +0.165%+0.142% / -12.1%-0.441% / -5.78%3.41% / 12.1%+0, +0halving bits set 50%, leaves at code limit 4%
cw6-delta1-bfp-e480210.0×0.108% / 0.474%-3.97e-05% / +0.0409%+0.193% / +0.793%-0.0931% / -0.745%0.783% / 0.794%+0, +0frame exponents: mean 3.1, max 5
cw8-delta1-linear10237.8×0.225% / 0.235%-0.000258% / +0.0203%-0.00232% / +0.391%-0.0077% / -0.389%0.389% / 0.394%+0, +0
cw8-delta1-sqrt10197.9×0.0399% / 0.181%+0.000737% / +0.00824%-0.0719% / -0.295%+0.00744% / +0.224%0.161% / 0.775%+0, +032% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0286% / 0.114%+5.33e-05% / -0.00935%+0.0185% / -0.195%-0.0105% / -0.184%0.193% / 0.196%+0, +0frame exponents: mean 2.7, max 3
cw8-delta1-bfp-e410517.6×0.0271% / 0.114%+6.18e-05% / -0.00935%+0.0185% / -0.196%-0.0118% / -0.184%0.193% / 0.196%+0, +0frame exponents: mean 3.2, max 5
cw-delta1-tree-710297.8×0.0486% / 0.343%+2.67e-05% / -0.0158%+0.0843% / -1.55%-0.0295% / -0.742%0.715% / 1.55%+0, +0halving bits set 50%, leaves at code limit 1%
ratectl-delta0123-zstd-4b50016.0×0.201% / 0.212%-0.00988% / -0.0285%-0.177% / -0.351%+0% / +0%0.345% / 0.351%+0, +0
ratectl-flac-4b50016.0×0.256% / 0.282%-0.00688% / -0.0341%-0.161% / -0.474%+0% / +0%0.449% / 0.477%+0, +0
delta0123-zstd-857413.9×0.114% / 0.234%-0.000674% / -0.0106%+0% / +0%+0% / +0%0.196% / 0.394%+0, +0orders 0: 300; B lowered by the cap on 15 blocks (to 7)
delta12-zstd-862612.8×0.114% / 0.234%-0.000674% / -0.0106%+0% / +0%+0% / +0%0.196% / 0.394%+0, +0orders 1: 300; B lowered by the cap on 15 blocks (to 7)
delta0123-zstd-1074410.7×0.0285% / 0.234%-0.000211% / -0.0106%+0% / +0%+0% / +0%0.0489% / 0.394%+0, +0orders 0: 300; B lowered by the cap on 115 blocks (to 7)
delta0123-zstd-1274410.7×0.0285% / 0.234%-0.000211% / -0.0106%+0% / +0%+0% / +0%0.0489% / 0.394%+0, +0orders 0: 300; B lowered by the cap on 300 blocks (to 7)
fluxcode-1069011.6×0.0554% / 0.058%-3.62e-05% / -0.00444%+0.00392% / +0.0963%-0.00315% / -0.0784%0.0963% / 0.0976%+0, +0orders 0: 300; step 2^-5..2^0
fluxcode-129548.4×0.0139% / 0.0147%+3.48e-06% / -0.0014%+0.000321% / -0.0242%-0.0028% / +0.0233%0.0241% / 0.0244%+0, +0orders 0: 300; step 2^-7..2^-2
fluxcode-1614555.5×0.000866% / 0.000912%+2.35e-07% / +7e-05%-8.79e-05% / -0.00151%-0.000139% / -0.00149%0.00151% / 0.00152%+0, +0orders 0: 300; step 2^-11..2^-6
fluxcode-16-f0.0112636.3×0.00176% / 0.0898%-2.47e-06% / -0.00294%+0.000107% / +0.0744%-0.000614% / -0.107%0.00303% / 0.156%+0, +0orders 0: 300; step 2^-5..2^-4 (noise floor applied on 300 blocks)
fluxcode-16-f0.0310157.9×0.00702% / 0.185%+2.03e-05% / -0.0112%+0.000533% / +0.298%-0.00261% / +0.296%0.0121% / 0.314%+0, +0orders 0: 300; step 2^-3 (noise floor applied on 300 blocks)
fluxcode-16-f0.169611.5×0.028% / 0.74%-8.09e-05% / -0.0359%+0.00648% / +0.359%-0.00319% / -0.392%0.0484% / 1.25%+0, +0orders 0: 300; step 2^-1 (noise floor applied on 300 blocks)
fluxcode-16-f0.2559313.5×0.0561% / 0.74%+5.17e-05% / -0.0359%+0.00731% / +0.359%-0.00319% / -0.392%0.0969% / 1.25%+0, +0orders 0: 300; step 2^-1..2^0 (noise floor applied on 300 blocks)
fluxcode-16-f0.551015.7×0.112% / 0.74%-0.000201% / -0.0359%+0.0313% / +0.359%-0.00319% / -0.392%0.194% / 1.25%+0, +0orders 0: 300; step 2^-1..2^1 (noise floor applied on 300 blocks)
fluxcode-16-f137821.2×0.225% / 0.74%+0.00114% / -0.0359%-0.0354% / -0.388%+0.385% / +0.399%0.388% / 1.25%+0, +0orders 0: 300; step 2^-1..2^2 (noise floor applied on 300 blocks)
impulses: original, decoded traces and errors

square-2.24hz

±100 square wave, √5 ≈ 2.236 Hz (hard edges). 5 one-minute signals; the plots show second 0 of the first (seed 7000), full scale 200. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged15451.8×10.2% / 15.3%+0% / -0.8%+0% / +0%+0% / +0%90% / 93.8%68 knots per block
swingdoor-0.5%15651.4×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%69 knots per block
swingdoor-2%15651.4×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%69 knots per block
dct-top6421637.0×5.63% / 6.01%+3.04e-05% / -0.7%+14.9% / +33.6%-14.6% / -33.6%47% / 71.2%
pwlinear-minmax1615451.8×10.2% / 15.3%+0% / -0.8%+0% / +0%+0% / +0%90% / 93.8%68 knots per block
pchip-minmax1615451.8×10.8% / 16.7%+0% / -0.941%+0% / +1.42e-14%+0% / +0%97.2% / 99.5%68 knots per block
bintree-center39220.4×2.37% / 2.73%+0% / -0.216%-0.195% / -0.195%+0.195% / +0.195%25% / 25%codes 00/01/10/11 per block: 34/0/484/483
bintree-edge39220.4×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%
dpcm4-linear-deflate61429.4×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%
dpcm4-cubic-deflate61429.4×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%
quant6-delta1-deflate61291.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%
dpcm6-linear-deflate61425.2×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%
dpcm6-cubic-deflate61425.2×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%
dpcm6-mulaw-deflate61425.2×1.9e-14% / 2.11e-14%+6.22e-15% / +1.07e-14%+0% / -2.84e-14%+2.84e-14% / +2.84e-14%2.84e-14% / 2.84e-14%
quant810167.9×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%
quant8-delta1-deflate61291.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%
gorilla-xor13858.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 995, reuse window 3, new window 1
cw-delta1-tree-240419.8×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%halving bits set 96%, leaves at code limit 0%
cw4-delta1-bfp-e455214.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%frame exponents: mean 13.9, max 15
cw-delta1-tree-465412.2×1.73e-08% / 6.92e-08%-1.04e-08% / +5.14e-08%+0% / +9.31e-08%-2.33e-08% / -2.33e-08%2.33e-08% / 9.31e-08%halving bits set 96%, leaves at code limit 0%
cw6-delta1-bfp-e480210.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%frame exponents: mean 13.9, max 15
cw8-delta1-linear10237.8×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%
cw8-delta1-sqrt10197.9×0.00411% / 0.00457%+0.00273% / +0.00339%+0% / -0.0061%+0.0061% / +0.0061%0.0122% / 0.0122%100% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%frame exponents: mean 2.8, max 3
cw8-delta1-bfp-e410517.6×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%frame exponents: mean 13.9, max 15
cw-delta1-tree-710297.8×1.73e-08% / 1.04e-06%-1.04e-08% / +7.71e-07%+0% / +1.4e-06%-2.33e-08% / -2.33e-08%2.33e-08% / 1.4e-06%halving bits set 96%, leaves at code limit 0%
ratectl-delta0123-zstd-4b28282.3×1.36e-08% / 1.43e-08%-9.6e-09% / -1.06e-08%-1.91e-08% / -1.91e-08%+0% / +0%1.91e-08% / 1.91e-08%
ratectl-flac-4b16049.9×0.00108% / 0.00113%-0.000764% / -0.000844%-0.00152% / -0.00152%+0% / +0%0.00152% / 0.00152%
delta0123-zstd-89883.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300
delta12-zstd-89883.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300
delta0123-zstd-109883.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; B lowered by the cap on 300 blocks (to 9)
delta0123-zstd-129883.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; B lowered by the cap on 300 blocks (to 9)
fluxcode-1051493.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^2
fluxcode-1251493.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^2
fluxcode-1651493.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^2
fluxcode-16-f0.0151493.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^2
fluxcode-16-f0.0351493.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^2
fluxcode-16-f0.151493.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^2
fluxcode-16-f0.2551493.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^2
fluxcode-16-f0.551493.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^2
fluxcode-16-f151493.5×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^2
square-2.24hz: original, decoded traces and errors

random-walk

Gaussian random walk, σ = 1 per step. 5 one-minute signals; the plots show second 0 of the first (seed 8000), full scale 23.12. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged27329.3×2.39% / 5.05%-0.00795% / -0.654%+0% / +1.08e-14%+0% / +0%9.32% / 25.2%128 knots per block
swingdoor-0.5%13505.9×0.245% / 0.322%-0.000139% / +0.0262%+0% / -0.366%+0% / +0.513%0.832% / 0.945%666 knots per block
swingdoor-2%49616.1×1.38% / 1.46%-6.43e-05% / -0.193%-0.0733% / -2.93%+0.0244% / +3.17%3.56% / 3.83%239 knots per block
dct-top6421637.0×2.22% / 4.79%+0% / -0.0529%-2.39% / -9.26%+2.23% / +7.66%7.5% / 16.2%
pwlinear-minmax1627529.0×2.38% / 5.05%-0.00739% / -0.654%+0% / +1.08e-14%+0% / +0%9.32% / 25.2%129 knots per block
pchip-minmax1627529.0×2.71% / 5.89%-0.00766% / -0.75%+0% / +4.31e-14%+0% / +0%10% / 26.8%129 knots per block
bintree-center39220.4×0.865% / 1.9%-0.00045% / +0.195%-0.781% / -3.13%+0.781% / +3.13%3.12% / 12.3%codes 00/01/10/11 per block: 265/324/206/206
bintree-edge39220.4×2% / 4.41%-0.000519% / +0.189%+0% / +1.08e-14%+0% / +0%6.24% / 24.8%
dpcm4-linear-deflate49216.3×0.311% / 0.703%+0.00136% / +0.0461%+0% / +0.89%+0% / +0.922%0.541% / 1.21%
dpcm4-cubic-deflate55514.4×0.302% / 0.641%+0.000982% / +0.044%+0.0143% / +1.37%+0% / -1.55%1.14% / 2.43%
quant6-delta1-deflate43418.4×0.458% / 0.478%+0.00144% / +0.0554%+0% / +1.08e-14%+0% / +0%0.793% / 0.794%
dpcm6-linear-deflate71011.3×0.0698% / 0.161%-0.000105% / -0.00811%-0.00223% / -0.186%+1.77e-05% / +0.265%0.122% / 0.274%
dpcm6-cubic-deflate75310.6×0.0689% / 0.146%-0.000164% / +0.0119%-0.00291% / -0.378%+0.000368% / -0.278%0.255% / 0.593%
dpcm6-mulaw-deflate75510.6×0.0601% / 0.128%+0.000194% / -0.00693%-0.00995% / -0.278%+0% / -0.306%0.227% / 0.535%
quant810167.9×0.113% / 0.118%+0.000225% / -0.00977%+0% / +1.08e-14%+0% / +0%0.196% / 0.196%
quant8-delta1-deflate66612.0×0.113% / 0.118%+0.000225% / -0.00977%+0% / +1.08e-14%+0% / +0%0.196% / 0.196%
gorilla-xor68401.2×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 0, reuse window 986, new window 13
cw-delta1-tree-240419.8×1.08% / 2.41%-0.0001% / +0.126%+0.13% / +5%-0.133% / -5.69%3.9% / 10.2%halving bits set 35%, leaves at code limit 36%
cw4-delta1-bfp-e455214.5×0.29% / 0.641%+0.000982% / -0.0378%+0.0331% / +1.19%+0% / -0.83%0.83% / 1.67%frame exponents: mean 2.8, max 5
cw-delta1-tree-465412.2×0.163% / 0.341%-0.000139% / +0.0225%+0% / +0.818%+0.013% / -1.1%0.516% / 1.47%halving bits set 31%, leaves at code limit 4%
cw6-delta1-bfp-e480210.0×0.0628% / 0.131%+6.41e-05% / +0.00837%+3.31e-15% / -0.203%+0% / -0.195%0.197% / 0.395%frame exponents: mean 3.0, max 5
cw8-delta1-linear10237.8×0.0172% / 0.0395%+3.74e-05% / +0.00194%+0.00176% / +0.0653%-0.00149% / -0.0542%0.0298% / 0.0668%
cw8-delta1-sqrt10197.9×0.0435% / 0.065%-5.55e-05% / +0.00479%-0.0608% / -0.373%+0.0637% / +0.341%0.177% / 0.373%28% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0155% / 0.032%-4.8e-06% / -0.00277%-0.000739% / -0.0552%+0% / -0.0488%0.0485% / 0.0976%frame exponents: mean 2.8, max 3
cw8-delta1-bfp-e410517.6×0.0153% / 0.032%-1.35e-05% / -0.00277%-0.0015% / -0.0552%-0.00137% / -0.0488%0.0485% / 0.0976%frame exponents: mean 3.0, max 6
cw-delta1-tree-710297.8×0.019% / 0.0404%+4.52e-05% / +0.00304%-0.0014% / +0.0863%-0.00144% / -0.103%0.0595% / 0.132%halving bits set 31%, leaves at code limit 1%
ratectl-delta0123-zstd-4b50016.0×0.284% / 0.299%+0.00103% / +0.0269%+0.156% / -0.441%+0% / +0%0.492% / 0.503%
ratectl-flac-4b50016.0×0.215% / 0.226%-0.000173% / +0.0229%+0.0273% / +0.326%+0% / +0%0.373% / 0.377%
delta0123-zstd-862112.9×0.113% / 0.118%+0.000225% / -0.00977%+0% / +0%+0% / +0%0.196% / 0.196%orders 1: 300
delta12-zstd-862112.9×0.113% / 0.118%+0.000225% / -0.00977%+0% / +0%+0% / +0%0.196% / 0.196%orders 1: 300
delta0123-zstd-108609.3×0.0281% / 0.0291%+1.02e-07% / -0.00404%+0% / +0%+0% / +0%0.0488% / 0.0489%orders 1: 300
delta0123-zstd-129898.1×0.0141% / 0.0289%-7.75e-06% / -0.00404%+0% / +0%+0% / +0%0.0244% / 0.0489%orders 1: 300; B lowered by the cap on 261 blocks (to 10)
fluxcode-1078810.2×0.0406% / 0.0567%-5.18e-05% / -0.00437%+0.0038% / +0.0883%+0.000228% / +0.089%0.0702% / 0.0973%orders 1: 300; step 2^-5..2^-3
fluxcode-1210687.5×0.0102% / 0.0142%-4.21e-06% / +0.00155%+0.000661% / +0.0223%-7.26e-05% / +0.0221%0.0175% / 0.0243%orders 1: 300; step 2^-7..2^-5
fluxcode-1615875.0×0.000634% / 0.000905%+1.84e-06% / +9.5e-05%+1.72e-05% / -0.00137%+4.13e-05% / +0.00145%0.0011% / 0.00152%orders 1: 300; step 2^-11..2^-9
fluxcode-16-f0.0115875.0×0.000634% / 0.000905%+1.84e-06% / +9.5e-05%+1.72e-05% / -0.00137%+4.13e-05% / +0.00145%0.0011% / 0.00152%orders 1: 300; step 2^-11..2^-9
fluxcode-16-f0.0315875.0×0.000634% / 0.000905%+1.84e-06% / +9.5e-05%+1.72e-05% / -0.00137%+4.13e-05% / +0.00145%0.0011% / 0.00152%orders 1: 300; step 2^-11..2^-9
fluxcode-16-f0.115875.0×0.000634% / 0.000905%+1.84e-06% / +9.5e-05%+1.72e-05% / -0.00137%+4.13e-05% / +0.00145%0.0011% / 0.00152%orders 1: 300; step 2^-11..2^-9
fluxcode-16-f0.2515875.0×0.000634% / 0.000905%+1.84e-06% / +9.5e-05%+1.72e-05% / -0.00137%+4.13e-05% / +0.00145%0.0011% / 0.00152%orders 1: 300; step 2^-11..2^-9
fluxcode-16-f0.515875.0×0.000634% / 0.000905%+1.84e-06% / +9.5e-05%+1.72e-05% / -0.00137%+4.13e-05% / +0.00145%0.0011% / 0.00152%orders 1: 300; step 2^-11..2^-9
fluxcode-16-f115875.0×0.000634% / 0.000905%+1.84e-06% / +9.5e-05%+1.72e-05% / -0.00137%+4.13e-05% / +0.00145%0.0011% / 0.00152%orders 1: 300; step 2^-11..2^-9
random-walk: original, decoded traces and errors

chirp

Linear chirp √2 → 32π Hz (≈1.41 → 100.5) repeating every second, amplitude 100. 5 one-minute signals; the plots show second 0 of the first (seed 9000), full scale 200. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged25631.2×23.3% / 24.6%+0.00107% / +1.03%+0% / +7.11e-15%+0% / +0%99.5% / 100%119 knots per block
swingdoor-0.5%13925.7×0.266% / 0.282%-0.000106% / -0.0214%+0% / +7.11e-15%+0% / +0%0.82% / 0.878%687 knots per block
swingdoor-2%8958.9×1.37% / 1.41%-0.00345% / +0.0895%+0% / +7.11e-15%+0% / +0%3.34% / 3.6%438 knots per block
dct-top6421637.0×22.3% / 23.6%+0.0952% / +1.98%+22.6% / +57.5%-20.1% / -66.1%69.9% / 110%
pwlinear-minmax1627329.3×23.3% / 24.5%-2.86e-05% / +1.05%+0% / +7.11e-15%+0% / +0%99.5% / 100%128 knots per block
pchip-minmax1627329.3×22.8% / 24%+0.0185% / +1.45%+0% / +7.11e-15%+0% / -1.42e-14%99.6% / 100%128 knots per block
bintree-center39220.4×4.98% / 5.25%-0.00328% / -0.418%-0.781% / -1.56%+0.781% / +1.56%24.3% / 25%codes 00/01/10/11 per block: 318/148/269/267
bintree-edge39220.4×10% / 10.8%+0.049% / -0.568%+0% / +7.11e-15%+0% / +0%49.3% / 50%
dpcm4-linear-deflate18343.7×1.27% / 1.34%+0.00357% / -0.18%+0.0981% / -2.19%+0.00741% / -2.17%2.18% / 2.22%
dpcm4-cubic-deflate22335.9×1.62% / 1.74%+0.000718% / -0.15%+1.74% / +3.03%-1.71% / -3.12%4.92% / 5.03%
quant6-delta1-deflate29627.0×0.449% / 0.467%-0.000827% / -0.0388%+0% / +1.42e-14%+0% / +0%0.793% / 0.794%
dpcm6-linear-deflate42119.0×0.284% / 0.308%+0.000206% / -0.0448%-0.0454% / -0.497%+0.0631% / -0.497%0.493% / 0.5%
dpcm6-cubic-deflate43918.2×0.362% / 0.385%+0.000275% / +0.0307%+0.319% / +0.591%-0.322% / -0.542%1.19% / 1.26%
dpcm6-mulaw-deflate44917.8×0.319% / 0.34%-0.00016% / +0.0337%+0.273% / +0.482%-0.277% / -0.466%1.07% / 1.15%
quant810167.9×0.112% / 0.117%-0.000165% / +0.0118%+0% / -2.13e-14%+0% / +0%0.196% / 0.196%
quant8-delta1-deflate64612.4×0.112% / 0.117%-0.000165% / +0.0118%+0% / -2.13e-14%+0% / +0%0.196% / 0.196%
gorilla-xor81541.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 0, reuse window 974, new window 25
cw-delta1-tree-240419.8×3.21% / 3.59%+0.00359% / -0.307%+5.52% / +8.75%-5.12% / -8.63%15.6% / 19.9%halving bits set 17%, leaves at code limit 39%
cw4-delta1-bfp-e455214.5×1.2% / 1.29%-0.00451% / -0.154%+0.8% / +3.3%-0.82% / -3.3%3.33% / 3.33%frame exponents: mean 1.4, max 9
cw-delta1-tree-465412.2×0.819% / 0.863%-0.00433% / -0.0791%+1.21% / +2.19%-1.07% / -2.18%2.18% / 2.21%halving bits set 19%, leaves at code limit 6%
cw6-delta1-bfp-e480210.0×0.265% / 0.282%-0.000996% / -0.0274%+0.185% / +0.791%-0.165% / -0.754%0.791% / 0.794%frame exponents: mean 1.5, max 9
cw8-delta1-linear10237.8×0.0693% / 0.0729%+3.6e-06% / +0.00669%-0.00401% / +0.121%-0.000511% / -0.12%0.12% / 0.122%
cw8-delta1-sqrt10197.9×0.105% / 0.11%+1.58e-05% / -0.0104%-0.0104% / -0.0463%+0.0096% / +0.0486%0.386% / 0.391%7% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0642% / 0.0681%-0.000247% / -0.0061%+0.0466% / +0.195%-0.0425% / -0.193%0.195% / 0.196%frame exponents: mean 1.4, max 3
cw8-delta1-bfp-e410517.6×0.064% / 0.0678%-0.000202% / +0.00586%+0.0432% / +0.191%-0.0395% / -0.193%0.195% / 0.196%frame exponents: mean 1.5, max 9
cw-delta1-tree-710297.8×0.0912% / 0.0972%-8.1e-06% / -0.00737%+0.0642% / +0.238%-0.0818% / -0.234%0.242% / 0.246%halving bits set 19%, leaves at code limit 1%
ratectl-delta0123-zstd-4b50016.0×0.145% / 0.15%-0.00133% / -0.0155%-0.00345% / +0.248%+0% / +0%0.252% / 0.255%
ratectl-flac-4b50016.0×0.0602% / 0.0626%-0.00145% / -0.00714%-0.0907% / -0.0947%+0% / +0%0.104% / 0.105%
delta0123-zstd-856914.1×0.112% / 0.117%-0.000165% / +0.0118%+0% / +0%+0% / +0%0.196% / 0.196%orders 3: 300
delta12-zstd-861812.9×0.112% / 0.117%-0.000165% / +0.0118%+0% / +0%+0% / +0%0.196% / 0.196%orders 2: 300
delta0123-zstd-1078910.1×0.028% / 0.029%+5.25e-05% / -0.00258%+0% / +0%+0% / +0%0.0488% / 0.0489%orders 3: 300
delta0123-zstd-1278910.1×0.028% / 0.029%+5.25e-05% / -0.00258%+0% / +0%+0% / +0%0.0488% / 0.0489%orders 3: 300; B lowered by the cap on 300 blocks (to 10)
fluxcode-1058413.7×0.0359% / 0.037%+3.63e-06% / +0.00329%+0.000118% / +0.00317%-0.000184% / -0.00869%0.0625% / 0.0625%orders 3: 300; step 2^-2
fluxcode-1269311.5×0.00901% / 0.0094%-8.63e-06% / -0.000825%+0.000118% / +0.00317%-0.000184% / -0.00869%0.0156% / 0.0156%orders 3: 300; step 2^-4
fluxcode-169778.2×0.000564% / 0.000583%+9.3e-07% / +5.22e-05%+6.81e-05% / -0.000964%-8.26e-05% / -0.000973%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.019778.2×0.000564% / 0.000583%+9.3e-07% / +5.22e-05%+6.81e-05% / -0.000964%-8.26e-05% / -0.000973%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.039778.2×0.000564% / 0.000583%+9.3e-07% / +5.22e-05%+6.81e-05% / -0.000964%-8.26e-05% / -0.000973%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.19778.2×0.000564% / 0.000583%+9.3e-07% / +5.22e-05%+6.81e-05% / -0.000964%-8.26e-05% / -0.000973%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.259778.2×0.000564% / 0.000583%+9.3e-07% / +5.22e-05%+6.81e-05% / -0.000964%-8.26e-05% / -0.000973%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f0.59778.2×0.000564% / 0.000583%+9.3e-07% / +5.22e-05%+6.81e-05% / -0.000964%-8.26e-05% / -0.000973%0.000976% / 0.000977%orders 3: 300; step 2^-8
fluxcode-16-f19778.2×0.000564% / 0.000583%+9.3e-07% / +5.22e-05%+6.81e-05% / -0.000964%-8.26e-05% / -0.000973%0.000976% / 0.000977%orders 3: 300; step 2^-8
chirp: original, decoded traces and errors

noisy-sine

π² ≈ 9.87 Hz sine, amplitude 100, + Gaussian noise σ = 5. 5 one-minute signals; the plots show second 0 of the first (seed 10000), full scale 225.5. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged24432.8×2.94% / 3.24%+0.00752% / +0.51%+0% / +6.6e-15%+0% / +0%10.3% / 14.9%113 knots per block
swingdoor-0.5%16244.9×0.193% / 0.215%+0.000301% / -0.0195%+0% / +6.6e-15%+0% / +0%0.816% / 0.889%803 knots per block
swingdoor-2%8559.4×1.27% / 1.36%-0.00519% / -0.139%+0% / -1.39%+0% / +1.22%3.54% / 3.8%419 knots per block
dct-top6421637.0×2.09% / 2.26%-9.35e-07% / -0.165%-2.88% / -6.38%+2.94% / +7.48%7.92% / 13%
pwlinear-minmax1627229.5×2.91% / 3.21%+0.00365% / -0.456%+0% / +6.6e-15%+0% / +0%10.3% / 14.9%127 knots per block
pchip-minmax1627229.5×3.92% / 4.61%+0.0112% / -0.563%+0% / +6.6e-15%+0% / +0%13.7% / 22.3%127 knots per block
bintree-center39220.4×1.52% / 1.68%-0.0038% / -0.132%-1.56% / -3.13%+1.56% / +3.13%6.18% / 12.5%codes 00/01/10/11 per block: 201/324/238/238
bintree-edge39220.4×3.57% / 3.93%-0.00774% / +0.263%+0% / +6.6e-15%+0% / +0%12.4% / 25%
dpcm4-linear-deflate48416.5×0.5% / 0.713%+0.00129% / -0.0486%-0.00811% / +1.02%-0.0171% / -1.02%0.868% / 1.24%
dpcm4-cubic-deflate54314.7×0.509% / 0.58%-5.79e-05% / +0.0506%+0.176% / -1.76%-0.198% / -1.98%1.86% / 2.35%
quant6-delta1-deflate50315.9×0.457% / 0.478%-8.41e-05% / -0.0445%+0% / -1.92e-14%+0% / +0%0.793% / 0.794%
dpcm6-linear-deflate71411.2×0.113% / 0.165%-3.81e-05% / -0.0109%+0.00101% / +0.24%+0.0168% / +0.255%0.196% / 0.281%
dpcm6-cubic-deflate75510.6×0.115% / 0.129%-0.000689% / +0.0124%+0.00233% / -0.459%-0.0244% / +0.411%0.418% / 0.527%
dpcm6-mulaw-deflate75810.6×0.101% / 0.118%+0.000352% / +0.0102%+0.041% / -0.454%-0.0382% / -0.441%0.37% / 0.477%
quant810167.9×0.113% / 0.117%+0.000182% / -0.00864%+0% / +6.6e-15%+0% / +0%0.196% / 0.196%
quant8-delta1-deflate71611.2×0.113% / 0.117%+0.000182% / -0.00864%+0% / +6.6e-15%+0% / +0%0.196% / 0.196%
gorilla-xor81801.0×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 0, reuse window 984, new window 15
cw-delta1-tree-240419.8×1.84% / 2.11%-0.000151% / -0.161%+2.36% / +6.44%-2.24% / -8.53%6.07% / 8.73%halving bits set 36%, leaves at code limit 31%
cw4-delta1-bfp-e455214.5×0.483% / 0.554%+0.000378% / -0.0424%+0.111% / -1.29%-0.072% / +1.63%1.62% / 1.67%frame exponents: mean 2.1, max 4
cw-delta1-tree-465412.2×0.276% / 0.306%+0.00046% / +0.0258%+0.105% / -0.975%-0.0325% / +0.854%0.805% / 1.62%halving bits set 29%, leaves at code limit 5%
cw6-delta1-bfp-e480210.0×0.108% / 0.12%-0.000108% / -0.00959%+0.016% / -0.218%-0.00851% / +0.38%0.198% / 0.397%frame exponents: mean 2.2, max 4
cw8-delta1-linear10237.8×0.0276% / 0.0396%-6.77e-05% / -0.00336%-0.00242% / +0.0552%-0.00103% / +0.065%0.0478% / 0.0685%
cw8-delta1-sqrt10197.9×0.0573% / 0.0614%+7.26e-05% / +0.00566%-0.103% / -0.386%+0.118% / +0.381%0.195% / 0.39%15% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0261% / 0.0287%+6.18e-05% / -0.00246%-0.00101% / +0.0759%+0.000519% / -0.0488%0.049% / 0.098%frame exponents: mean 2.2, max 3
cw8-delta1-bfp-e410517.6×0.0261% / 0.0287%+6.57e-05% / -0.00246%-0.00148% / +0.0759%+0.000272% / -0.0488%0.049% / 0.098%frame exponents: mean 2.2, max 4
cw-delta1-tree-710297.8×0.0327% / 0.0353%-2.58e-05% / +0.0034%+0.00183% / -0.108%+0.000815% / -0.107%0.0951% / 0.166%halving bits set 29%, leaves at code limit 1%
ratectl-delta0123-zstd-4b50016.0×0.425% / 0.444%-2.94e-05% / +0.039%+0.0827% / +0.519%+0% / +0%0.735% / 0.739%
ratectl-flac-4b50016.0×0.396% / 0.418%-0.00129% / -0.0308%-0.133% / +0.513%+0% / +0%0.688% / 0.703%
delta0123-zstd-870411.4×0.113% / 0.117%+0.000182% / -0.00864%+0% / +0%+0% / +0%0.196% / 0.196%orders 1: 300
delta12-zstd-870411.4×0.113% / 0.117%+0.000182% / -0.00864%+0% / +0%+0% / +0%0.196% / 0.196%orders 1: 300
delta0123-zstd-109438.5×0.0282% / 0.0292%+4.77e-05% / -0.00235%+0% / +0%+0% / +0%0.0488% / 0.0489%orders 1: 300
delta0123-zstd-129438.5×0.0282% / 0.0292%+4.77e-05% / -0.00235%+0% / +0%+0% / +0%0.0488% / 0.0489%orders 1: 300; B lowered by the cap on 300 blocks (to 10)
fluxcode-109058.8×0.0324% / 0.0338%+1.6e-05% / +0.00272%+0.00436% / +0.0561%-0.00139% / +0.0574%0.0561% / 0.0582%orders 1: 300; step 2^-2
fluxcode-1212026.7×0.00812% / 0.00861%-7.57e-06% / +0.00107%+0.00101% / +0.0142%-9.46e-05% / -0.0141%0.014% / 0.0145%orders 1: 300; step 2^-4
fluxcode-1617134.7×0.000506% / 0.000531%-2.11e-07% / +4.83e-05%-8.15e-06% / +0.000867%-7.99e-05% / -0.000891%0.000877% / 0.00091%orders 1: 300; step 2^-8
fluxcode-16-f0.0113336.0×0.00405% / 0.00428%+1.73e-05% / +0.000471%+0.00019% / -0.00706%-0.000457% / -0.00709%0.00701% / 0.00728%orders 1: 300; step 2^-5 (noise floor applied on 300 blocks)
fluxcode-16-f0.0310747.4×0.0162% / 0.017%+4.83e-06% / -0.00144%+0.00094% / -0.0285%+0.00131% / +0.0281%0.028% / 0.029%orders 1: 300; step 2^-3 (noise floor applied on 300 blocks)
fluxcode-16-f0.18629.3×0.0334% / 0.0673%+2.34e-05% / -0.00511%+0.00131% / -0.113%-0.00139% / +0.111%0.0571% / 0.115%orders 1: 300; step 2^-2..2^-1 (noise floor applied on 300 blocks)
fluxcode-16-f0.2565512.2×0.13% / 0.137%-8.49e-05% / +0.0141%-0.0195% / +0.226%+0.000114% / +0.226%0.224% / 0.233%orders 1: 300; step 2^0 (noise floor applied on 300 blocks)
fluxcode-16-f0.553415.0×0.259% / 0.273%-0.00164% / -0.0232%-0.0442% / -0.451%-0.000745% / -0.444%0.449% / 0.466%orders 1: 300; step 2^1 (noise floor applied on 300 blocks)
fluxcode-16-f142119.0×0.518% / 0.547%+0.00152% / -0.0484%+0.00143% / -0.907%+0.0498% / -0.91%0.897% / 0.929%orders 1: 300; step 2^2 (noise floor applied on 300 blocks)
noisy-sine: original, decoded traces and errors

sensor-0.1

Slow drift + √2 Hz wobble, rounded to 0.1 (decimal floats). 5 one-minute signals; the plots show second 0 of the first (seed 11000), full scale 11. Rows are in legend order.

codecbytes/blockratioRMSE
median / worst		mean err
median / worst		max err
median / worst		min err
median / worst		max |err|
median / worst		notes
pwlinear-minmax16-merged24832.3×0.676% / 0.87%+0.00597% / -0.235%+0% / +0%+0% / +0%2.48% / 3.44%115 knots per block
swingdoor-0.5%56614.1×0.342% / 0.38%-0.00101% / -0.0483%+0% / -0.879%+0% / +0.904%0.887% / 0.965%274 knots per block
swingdoor-2%14654.7×0.807% / 0.978%+0.00608% / +0.245%-0.879% / -2.74%+0.904% / +2.78%2.76% / 3.42%64 knots per block
dct-top6421637.0×0.528% / 0.654%+6.67e-14% / +3.02e-13%-0.473% / -2.18%+0.482% / +2.03%1.78% / 2.8%
pwlinear-minmax1627329.3×0.667% / 0.865%+0.00603% / -0.236%+0% / +0%+0% / +0%2.46% / 3.44%127 knots per block
pchip-minmax1627329.3×0.83% / 1.02%+0.00506% / -0.245%+0% / +0%+0% / +0%3.11% / 4.75%127 knots per block
bintree-center39220.4×0.275% / 0.34%+0.00118% / +0.0527%-0.195% / -0.781%+0.195% / +0.781%1.42% / 1.54%codes 00/01/10/11 per block: 137/361/253/251
bintree-edge39220.4×0.652% / 0.823%+0.000909% / +0.0851%+0% / +0%+0% / +0%2.98% / 3.1%
dpcm4-linear-deflate27129.5×0.093% / 0.124%-0.00163% / +0.0724%+0% / -0.155%+0% / +0.157%0.131% / 0.157%
dpcm4-cubic-deflate32125.0×0.0874% / 0.125%-7.34e-05% / -0.0133%+0.0189% / +0.242%-0.0166% / -0.243%0.221% / 0.272%
quant6-delta1-deflate18443.5×0.459% / 0.561%-0.0012% / -0.27%+0% / +0%+0% / +0%0.794% / 0.794%
dpcm6-linear-deflate27129.5×0.021% / 0.0281%-0.000625% / -0.017%-0.0119% / -0.0351%+0% / +0.0354%0.0296% / 0.0354%
dpcm6-cubic-deflate32624.6×0.024% / 0.0285%-5.08e-05% / +0.00221%+0.0393% / -0.0731%-0.0274% / -0.0532%0.0467% / 0.0944%
dpcm6-mulaw-deflate31525.4×0.0187% / 0.0223%+8.84e-05% / -0.00244%+0.00494% / -0.0448%-0.00494% / -0.0433%0.0397% / 0.0518%
quant810167.9×0.114% / 0.141%+0.000537% / -0.0955%+0% / +0%+0% / +0%0.196% / 0.196%
quant8-delta1-deflate29227.4×0.114% / 0.141%+0.000537% / -0.0955%+0% / +0%+0% / +0%0.196% / 0.196%
gorilla-xor31072.6×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%per block: same 545, reuse window 429, new window 25
cw-delta1-tree-240419.8×0.173% / 0.347%+6.62e-08% / +0.0739%-4.66e-08% / +1.03%+4.66e-08% / -0.99%0.922% / 1.96%halving bits set 30%, leaves at code limit 27%
cw4-delta1-bfp-e455214.5×0.0796% / 0.0941%+0.0019% / +0.0321%+0% / +0.208%+0% / +0.208%0.208% / 0.417%frame exponents: mean 4.7, max 10
cw-delta1-tree-465412.2×0.0186% / 0.0791%-0.000131% / -0.0412%+1.56e-06% / +0.155%-1.23e-06% / +0.15%0.131% / 0.206%halving bits set 24%, leaves at code limit 18%
cw6-delta1-bfp-e480210.0×0.019% / 0.0219%+0.000212% / +0.00764%+0% / +0.0496%+0% / +0.0496%0.0496% / 0.0929%frame exponents: mean 4.7, max 15
cw8-delta1-linear10237.8×0.00512% / 0.00686%-0.00015% / +0.0042%+0% / -0.00856%+0% / +0.00865%0.00722% / 0.00865%
cw8-delta1-sqrt10197.9×0.0201% / 0.0281%-0.000273% / +0.00376%-0.00824% / -0.179%+0.00513% / +0.105%0.0896% / 0.187%55% of samples use a finer-than-7-bit delta
cw8-delta1-bfp-e210357.7×0.0141% / 0.0154%+2.07e-05% / +0.00712%+0% / +0.0245%+0% / +0.0245%0.024% / 0.0245%frame exponents: mean 3.0, max 3
cw8-delta1-bfp-e410517.6×0.0047% / 0.0054%+6.14e-05% / +0.00189%+0% / +0.0123%+0% / +0.0123%0.0123% / 0.0188%frame exponents: mean 4.8, max 13
cw-delta1-tree-710297.8×0.00702% / 0.00874%-1.42e-05% / +0.0047%+5.36e-06% / -0.0169%-6.5e-06% / +0.016%0.0146% / 0.0249%halving bits set 24%, leaves at code limit 13%
ratectl-delta0123-zstd-4b50016.0×0.00209% / 0.00249%-4.66e-05% / +0.0017%+0.00271% / +0.00314%+0% / +0%0.00359% / 0.00365%
ratectl-flac-4b49416.2×0.0631% / 0.0719%-0.00132% / -0.0586%-0.11% / -0.11%+0% / +0%0.11% / 0.11%
delta0123-zstd-831425.5×0.114% / 0.141%+0.000491% / -0.0955%+0% / +0%+0% / +0%0.196% / 0.196%orders 1: 300
delta12-zstd-831425.5×0.114% / 0.141%+0.000491% / -0.0955%+0% / +0%+0% / +0%0.196% / 0.196%orders 1: 300
delta0123-zstd-1033823.6×0.0282% / 0.0304%+0.000182% / +0.0115%+0% / +0%+0% / +0%0.0489% / 0.0489%orders 1: 300
delta0123-zstd-1237321.5×0.00707% / 0.00873%+2.06e-05% / +0.00568%+0% / +0%+0% / +0%0.0122% / 0.0122%orders 1: 300
fluxcode-1022236.1×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^-1
fluxcode-1222236.1×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^-1
fluxcode-1622236.1×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^-1
fluxcode-16-f0.0122236.1×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^-1
fluxcode-16-f0.0322236.1×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^-1
fluxcode-16-f0.122236.1×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^-1
fluxcode-16-f0.2522236.1×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^-1
fluxcode-16-f0.522236.1×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^-1
fluxcode-16-f122236.1×0% / 0%+0% / +0%+0% / +0%+0% / +0%0% / 0%orders 1: 300; decimal grid 10^-1
sensor-0.1: original, decoded traces and errors

Parameter sweeps

fluxcode B sweep: power-of-two only vs decimal detection, order 1 vs orders 0–3, B = 9..16

The fluxcode package (at the repo root, format in docs/SPEC.md): quantize on a power-of-two step (min exact, step = 2e with the range in [2B−1, 2B) steps), residuals mod 216 as int16 → zigzag → 16 bit planes XX B is max_quantize_bits; the noise floor is off here and has its own sweep below. Decimal detect: if every value is within a quarter power-of-two step of a multiple of 10p (coarsest p whose step is still coarser than the power-of-two step), quantize on that grid instead and reconstruct K / 10−p, bit-exact for decimal data; values that were float32 are rounded back to float32. Otherwise (ADC counts, other steps) it falls back to the power-of-two grid. Order 1 fixes the predictor; orders 0–3 picks the lowest-variance one per block, from its first 250 samples. Continuous data: the 7,200 blocks of 12 signal types used by the benchmarks. Discretized: 9 signals rounded to a quantum (0.001, 0.01, 0.1, 2−4, a 12-bit ADC step, decimals via float32), 600 blocks each. Sizes are whole block groups (header, per-block anchors, zstd frame). Timings: one thread, µs per 1000-sample block, including zstd and the Python API.

fluxcode B sweep
versionBcontinuous bits/samplecontinuous median RMSEcontinuous worst max errdiscretized bits/samplediscretized median RMSEdiscretized worst max errbit-exact samplesencode µs (cont / disc)decode µs
order 1, power-of-two only92.540.071%0.195%4.740.0713%0.195%13.4%3.8 / 3.62.0
order 1, power-of-two only103.000.0356%0.0976%5.540.0335%0.0972%18.4%3.7 / 3.81.9
order 1, power-of-two only113.560.0179%0.0488%6.330.0166%0.0483%19.6%4.0 / 4.31.9
order 1, power-of-two only124.150.00893%0.0244%7.070.00832%0.0242%19.9%4.3 / 5.12.0
order 1, power-of-two only134.780.00447%0.0122%7.710.00419%0.0121%20.6%4.8 / 6.02.0
order 1, power-of-two only145.510.00223%0.0061%8.490.00216%0.00604%22.0%5.0 / 6.32.0
order 1, power-of-two only156.210.00112%0.00305%9.340.00112%0.00302%24.8%5.1 / 6.52.0
order 1, power-of-two only166.960.000559%0.00153%9.870.000516%0.00151%30.4%4.7 / 6.21.9
order 1, decimal detect92.490.0708%0.195%4.660.0662%0.195%23.3%3.8 / 3.61.9
order 1, decimal detect102.950.0352%0.0976%5.390.0322%0.0972%34.9%3.7 / 3.71.9
order 1, decimal detect113.500.0176%0.0488%6.130.016%0.0483%37.4%4.0 / 4.21.9
order 1, decimal detect124.100.0088%0.0244%6.740.0079%0.0242%47.3%4.2 / 4.82.0
order 1, decimal detect134.720.0044%0.0122%7.220%0.0121%56.2%4.7 / 5.52.0
order 1, decimal detect145.410.0022%0.0061%7.730%0.00604%59.4%5.0 / 5.61.9
order 1, decimal detect156.010.0011%0.00305%8.010%0.00302%73.7%4.9 / 5.71.9
order 1, decimal detect166.780.000551%0.00153%8.220%0.00151%87.7%4.5 / 5.41.9
order 0-3, power-of-two only92.510.071%0.195%4.760.0713%0.195%13.4%3.8 / 3.61.8
order 0-3, power-of-two only103.000.0356%0.0976%5.560.0335%0.0972%18.4%3.6 / 3.71.9
order 0-3, power-of-two only113.500.0179%0.0488%6.350.0166%0.0483%19.6%3.9 / 4.31.8
order 0-3, power-of-two only124.070.00893%0.0244%7.070.00832%0.0242%19.9%4.3 / 5.01.9
order 0-3, power-of-two only134.650.00447%0.0122%7.680.00419%0.0121%20.6%4.8 / 5.91.9
order 0-3, power-of-two only145.240.00223%0.0061%8.360.00216%0.00604%22.0%5.0 / 6.31.9
order 0-3, power-of-two only155.790.00112%0.00305%9.140.00112%0.00302%24.8%5.1 / 6.61.9
order 0-3, power-of-two only166.230.000559%0.00153%9.720.000516%0.00151%30.4%4.7 / 6.11.9
order 0-3, decimal detect92.450.0708%0.195%4.680.0662%0.195%23.3%3.8 / 3.61.8
order 0-3, decimal detect102.940.0352%0.0976%5.410.0322%0.0972%34.9%3.6 / 3.71.9
order 0-3, decimal detect113.450.0176%0.0488%6.160.016%0.0483%37.4%3.9 / 4.41.9
order 0-3, decimal detect124.010.0088%0.0244%6.750.0079%0.0242%47.3%4.2 / 4.71.9
order 0-3, decimal detect134.590.0044%0.0122%7.200%0.0121%56.2%4.6 / 5.31.9
order 0-3, decimal detect145.140.0022%0.0061%7.600%0.00604%59.4%4.8 / 5.41.8
order 0-3, decimal detect155.590.0011%0.00305%7.860%0.00302%73.7%4.8 / 5.51.8
order 0-3, decimal detect166.040.000551%0.00153%8.080%0.00151%87.7%4.5 / 5.11.8

Discretized signals at B = 16 (bits/sample)

signallossless at the true quantumorder 1, power-of-two onlyorder 1, decimal detectorder 0-3, power-of-two onlyorder 0-3, decimal detect
random-walk q0.00112.6512.5912.5312.5912.53
random-walk q0.019.3212.369.2812.369.28
random-walk q0.15.829.985.539.985.53
random-walk q2^-46.496.496.496.496.49
noisy-sine q0.18.9110.978.8910.978.89
noisy-sine adc128.9313.6913.6913.6913.69
sin-4.12hz q0.012.635.033.503.652.19
sensor-0.11.794.011.784.011.78
noisy-sine q0.01 via float3212.3213.7412.3213.7412.32

“Lossless at the true quantum” is the same pipeline told the quantum in advance. Decimal detection reaches it on every decimal signal; the ADC and 2−4 steps aren't decimal and fall back to the power-of-two grid. random-walk q0.001 only partly: blocks whose range exceeds 65,535 × 0.001 can't be exact in 16 bits. Byte and nibble plane layouts, which fluxcode dropped for bit planes, are compared in REPORT.md §11.

Noise floor sweep: fluxcode-16, f = 0 … 1

On blocks whose second differences look like white noise (lag-1 autocorrelation < −0.6), the step becomes at most f·σ, with σ a robust estimate from the same differences (docs/ENCODER.md §5.2). B stays 16; on these signals the noise-floor step is coarser than the 16-bit step for every f ≥ 0.01, so f alone sets the size. fluxcode's default is f = 0.25. Noisy signals only (600 blocks each, minute block groups); clean signals are unaffected at every f. Errors are median RMS over blocks in units of the true noise σ, against the input and against the same signal generated without noise.

noise floor sweep
signalf = 0f = 0.01f = 0.03f = 0.1f = 0.25f = 0.5f = 1
gauss-spikes12.04 b/s
+0.001σ / 1.001σ		11.02 b/s
+0.002σ / 1.001σ		9.29 b/s
+0.006σ / 1.001σ		7.48 b/s
+0.025σ / 1.002σ		6.61 b/s
+0.050σ / 1.003σ		5.66 b/s
+0.100σ / 1.009σ		4.74 b/s
+0.199σ / 0.991σ
impulses11.68 b/s
+0.001σ / 1.000σ		10.10 b/s
+0.002σ / 0.999σ		8.14 b/s
+0.006σ / 0.999σ		5.62 b/s
+0.025σ / 1.000σ		4.80 b/s
+0.050σ / 1.002σ		4.12 b/s
+0.100σ / 1.008σ		3.06 b/s
+0.200σ / 0.980σ
noisy-sine13.71 b/s
+0.000σ / 0.997σ		10.66 b/s
+0.002σ / 0.997σ		8.59 b/s
+0.007σ / 0.998σ		6.89 b/s
+0.015σ / 0.998σ		5.24 b/s
+0.058σ / 0.999σ		4.28 b/s
+0.115σ / 1.004σ		3.38 b/s
+0.231σ / 1.024σ
noisy-sine q0.18.89 b/s
+0.000σ / 0.999σ		8.89 b/s
+0.000σ / 0.999σ		8.58 b/s
+0.007σ / 0.999σ		6.87 b/s
+0.014σ / 0.999σ		5.24 b/s
+0.058σ / 1.001σ		4.27 b/s
+0.116σ / 1.006σ		3.38 b/s
+0.231σ / 1.027σ
noisy-sine adc1213.69 b/s
+0.000σ / 1.001σ		10.59 b/s
+0.002σ / 1.001σ		8.60 b/s
+0.007σ / 1.001σ		6.86 b/s
+0.028σ / 1.001σ		5.24 b/s
+0.058σ / 1.002σ		4.27 b/s
+0.116σ / 1.007σ		3.39 b/s
+0.231σ / 1.026σ
noisy-sine q0.01 via float3212.32 b/s
+0.000σ / 1.000σ		10.68 b/s
+0.002σ / 1.000σ		8.59 b/s
+0.007σ / 1.000σ		6.86 b/s
+0.028σ / 1.000σ		5.24 b/s
+0.058σ / 1.002σ		4.28 b/s
+0.116σ / 1.007σ		3.38 b/s
+0.231σ / 1.026σ

Each cell: bits/sample, then RMS error vs input / vs clean signal (σ). Lossless points (noisy-sine q0.1 at f ≤ 0.01, still on its decimal grid) are left off the log axis. The step is 2^floor(log2(f·σ)), so f values between powers of two can give the same step.

DPCM sweep: code bits, packing, curve, step scale (all signal kinds, + deflate)

D = step scale × max |x[i] − x[i−1]| per block. “nibble” packs two 4-bit codes per byte before deflate. One deflate call per one-minute block group. Per-kind median RMSE in % FS.

bitspackingcurveD scalebits/samplemedian RMSEworst RMSEworst max |err|linearquadraticsin-4.12hzsin-9.87hzsin-50.3hzgauss-spikesimpulsessquare-2.24hzrandom-walkchirpnoisy-sinesensor-0.1
6bytecubic c=012.430.0484%1.06%1.61%00.000950.0120.0290.150.0680.8800.070.280.110.021
6bytecubic c=0.512.560.0452%0.975%3.14%00.00190.0180.0430.220.0470.4700.0570.290.0950.013
6bytecubic c=0.812.680.0532%1.21%4.04%00.00240.0230.0560.290.050.2500.0690.360.120.024
6bytecubic c=112.650.0638%1.47%4.65%00.00280.0270.0660.330.0620.2200.0840.440.140.025
6byteμ-law μ=712.630.0485%1.09%3.69%00.00220.0210.050.260.0470.331.9e-140.060.320.10.019
6byteμ-law μ=3112.670.0649%1.47%5.43%00.00320.030.0730.370.0570.2100.0770.440.130.0083
6byteμ-law μ=25512.630.0977%2.3%8.11%00.00490.0470.110.570.0850.181.9e-140.110.670.190.022
6bytecubic c=00.52.206.87%28.9%50.7%29282222224.32.73.20.389.50.640.15
4bytecubic c=011.480.213%4.21%7.12%00.00420.0530.130.650.31.600.311.30.50.093
4bytecubic c=0.511.610.2%4.38%12.7%00.00770.0790.190.980.211.600.251.30.420.063
4bytecubic c=0.811.750.232%5.44%16.1%00.00960.0990.241.20.220.8600.31.60.510.087
4bytecubic c=111.760.272%6.46%18.5%00.0110.110.271.40.270.5800.371.90.620.12
4byteμ-law μ=311.600.193%4.22%11.9%00.00710.0730.170.880.211.600.261.30.420.07
4byteμ-law μ=711.670.211%4.74%14.6%00.00870.0890.211.10.211.41.9e-140.271.40.450.034
4byteμ-law μ=3111.750.272%6.49%20%00.0120.120.291.50.250.6200.3420.580.1
4nibblecubic c=011.380.213%4.21%7.12%00.00420.0530.130.650.31.600.311.30.50.093
4nibblecubic c=0.811.640.232%5.44%16.1%00.00960.0990.241.20.220.8600.31.60.510.087

D below the largest step causes slope overload on smooth signals (a line rising 1/sample can't be tracked with a max step of 0.5), regardless of curve. Fewer levels means less resolved noise, hence less entropy for deflate: 4-bit codes land near quant6-delta1-deflate's size with much lower error on smooth signals. Nibble packing doesn't help because deflate matches on byte boundaries.

DPCM candidates, size vs error

Design notes: decisions made where the spec was open