Skip to content

feat(frackture): introduce micro-tier tiny payload path and dual-layout serialization - #25

Draft
cto-new[bot] wants to merge 1 commit into
mainfrom
feat/frackture-tiny-micro-tier-compression
Draft

feat(frackture): introduce micro-tier tiny payload path and dual-layout serialization#25
cto-new[bot] wants to merge 1 commit into
mainfrom
feat/frackture-tiny-micro-tier-compression

Conversation

@cto-new

@cto-new cto-new Bot commented Dec 15, 2025

Copy link
Copy Markdown
Contributor

Summary

Introduce a micro-tier payload path for ultra-tiny inputs (<100B) and a dual-layout
serialization format that preserves compatibility with legacy dict payloads.

Details

  • Add a MICRO compression tier and micro-encoder path that emits compact layouts
    (1-byte header, 24–32B symbolic, 16–24B entropy) for tiny inputs.
  • Extend FrackturePayload with a micro-layout flag and support two binary layouts in
    to_bytes/from_bytes and serialize_frackture_payload, while keeping to_legacy_dict
    compatibility.
  • Update reconstruction weighting to respect micro-tier (80/20) and tiny/two-tier
    weighting (70/30).
  • Adjust preprocessing and helpers so tiny inputs auto-select micro layout; larger tiers
    unchanged.
  • Expand tests to cover micro-tier paths and tighten size regression tests; add tiny-focused
    cases.
  • Introduce a benchmark guard to ensure serialized payload size stays under the target.

Impact and migration notes

  • This introduces a breaking change in the binary header (new micro-flag) and
    two binary payload layouts. Backwards-compatible dict payloads are preserved, but
    binary consumers must adapt to the new header interpretation.
  • Regular dict-based workflows continue to function as before.

…ialization

Introduce a dedicated micro payload path for inputs under 100 bytes that yields
smaller serialized payloads by skipping the full entropy channel (or downsampling
to 8 features) and applying heavier symbolic weighting.

Extend FrackturePayload to carry a micro layout flag and support two binary layouts
in to_bytes/from_bytes and serialize_frackture_payload, while preserving to_legacy_dict
compatibility. Update encoding/decoding, reconstruction weighting (micro: 80/20; tiny: 70/30)
and preprocessing to auto-select micro for very small inputs. Add regression tests,
new tiny-focused cases, and a benchmark guard enforcing payload_size_range.avg < 200.

This introduces a breaking change in header layout; legacy dict-based usage remains
supported, but binary header must be interpreted accordingly.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

0 participants