Lossless & Bounded-Loss Compression

Compression shrinks the archive without discarding physics, using lossless coding where forensics demand it and bounded-loss only where theory permits.

Compress without lying

Petabyte-scale telemetry must be compressed, but the fabric will not silently discard information a future analysis might need. It uses lossless compression for raw forensic channels and bounded-loss compression only where the discarded content is provably below the noise floor, with the bound recorded so any consumer knows the guarantee.

Lossless where it counts

• Raw Mirnov, Thomson, and strain traces are stored losslessly for disruption and quench forensics.
• Delta and predictive coding exploit the temporal smoothness of most channels.
• Entropy coding packs the residuals near their information-theoretic limit.
• Reconstruction is bit-exact, so re-reduction yields identical features.

Bounded loss where safe

Some channels carry content well above their physics bandwidth or below their noise floor. There, the fabric may quantize to a bound tied to the sensor's own noise, discarding only what was never information. The bound is chosen from the channel's information budget and stored in metadata, so bounded-loss is auditable, not arbitrary.

def bounded_quant(x, noise_sigma, k=0.25):
    # step below the sensor noise: discards no real information
    step = k * noise_sigma
    import math
    return round(x/step)*step, step   # value, and the recorded bound

Auditable by design

Every stored series records whether it is lossless or bounded-loss and, if bounded, the exact bound. This keeps the archive trustworthy for both machines: a breeder disruption or a burner plug event is always available at the fidelity its physics requires, and lineage shows precisely what compression was applied.