CSF — patent prior-art & freedom-to-operate review
Date: 2026-07-22 Type: Patent prior-art search + FTO assessment for the CSF compression format. Status: Planning-grade. Not legal advice — an attorney prior-art/FTO review precedes any filing (per KEYSTONE-IP-AND-BUILDOUT.md §7). Grounding contract: every patent cited was surfaced by a live search 2026-07-22 and links to its Google Patents / USPTO record; verdicts are tagged [prior-art] (technique is covered/anticipated), [public-domain] (open standard, no patent), or [thin novelty] (a narrow composition sliver, routed to defensive publication not patent).
⚠️ On the patent corpus. The repo's worldwide patent corpus (PR #2826) is a separate lane's infrastructure and is empty (harvest gated on an EPO OPS key). This review was done via live web patent search, not that corpus — so it stands on its own and can be re-verified from the linked records. When the corpus is populated, a
patent_searchover the same queries should reproduce these hits.
TL;DR
A patent search on CSF's shipped techniques finds named prior art or open standards for every one of them — best-fit codec selection, seed/regeneration storage, byte-shuffle, BCJ filters, and framed random-access archives are all occupied or public-domain. This confirms and deepens the IP register's existing honest grade (CSF §4.7: "LOW novelty — format identity + integrity, not algorithm"). The genuine deliverable is not a patent — it is (1) clean freedom-to-operate (the blocking-looking patents are narrow, packet-header-specific, or expired) and (2) one thin composition sliver worth a defensive-publication timestamp, not a filing. No CSF compression mechanism is patentable; the moat remains systems + owned data, exactly as the register already states.
Technique-by-technique prior art
| CSF technique (shipped) | Closest prior art | Verdict |
|---|---|---|
| Multi-codec best-fit + self-describing header (omni: try the panel, keep the smallest, header names the codec) | US8111704B2 Multiple compression techniques for packetized information · US6804238 selects best compression by analyzing compression ratios · US5953503A multiple preset dictionaries; header indicates the scheme used (filed— term-expired) | [prior-art] — "try several, keep smallest, self-describe in the header" is squarely anticipated. Not novel. |
| Generative members / recompute-as-storage (store a generator spec + the sha-256 of the materialized bytes; regenerate on read) | US20040267773A1 Generation of repeatable synthetic data — a deterministic generator + seed regenerates any particular entry (2004 — term-expired) · US11967975 recursive compression using seed bits · US7184597B1 | [thin novelty] — the "regenerate from a seed/recipe instead of storing bytes" core is prior art (US20040267773 even covers regenerating one entry on demand = our observer-slice). The only unoccupied sliver is the composition: a mixed lossless container that stores generative members verified by materialize-then-sha alongside best-fit-compressed members. Defensive-publication material, not a patentable algorithm. |
| Byte-shuffle / SoA planes; bitshuffle (transform 3/4/5) | HDF5 shuffle filter + Blosc bitshuffle — documented open standards (HDF Group filters, Blosc/hdf5-blosc); no patent surfaced | [public-domain] — an open, widely-implemented filter. CSF auto-selecting the stride per file is application, not invention. |
| BCJ / executable branch filter before LZMA (codec 9) | xz/LZMA BCJ filters (x86/ARM/…), UPX — public-domain, in the xz-utils standard | [public-domain] — a standard LZMA prefilter. Not novel. |
| Framed solid + slice-addressable random access (frames cut at member boundaries; O(window) reads) | US20120109909A1 Random Access Data Compression · US9503123 random access via bitwise indices · US11934346 random access of a member in a compressed tar · the zstd seekable format (BSD) + Mark Adler's zran.c (public) |
[prior-art] — "split into independent frames + a seek table mapping compressed↔decompressed offsets" is exactly the zstd seekable format and these patents. Not novel. |
| Per-file SHA-256 + footer integrity + path-traversal guard (the format identity) | Standard archive integrity (zip CRC, tar, OCI content-addressing) | [prior-art] — integrity + content addressing is ubiquitous. The value is the coined name + the guarantee, not novelty (matches register §4.7). |
The two real gains (neither is a patent)
1. Freedom-to-operate is clean [planning-grade]. The patents that look blocking are not: US5953503A (1997) and US20040267773 (2004) are past their 20-year term; US8111704/US6804238 are packet-header compression (a different field of use); the seekable-archive art is either expired-adjacent or matched by BSD-licensed open formats (zstd seekable). Net: nothing found blocks CSF from shipping any of its techniques. This is worth recording — it's the FTO half of a real IP review, and it came back green. (Attorney confirmation still precedes reliance, per register §7.)
2. One defensible sliver → defensive publication, not a filing [thin novelty]. The only composition without a clean direct hit is: a single integrity-checked container that stores (a) best-fit-compressed members, (b) generative members verified by materialize-then-hash, and (c) framed + slice-addressable reads, with loop-grounded tier selection. Even this is an assembly of known parts. The right vehicle is a defensive-publication timestamp (this doc + the CSF spec on the public repo) that blocks others from patenting the composition and preserves freedom-to-operate — exactly the strategy the register already commits to (§1: "publish the methods, trademark the names").
What this changes in the register
Nothing structurally — it validates the existing routing. KEYSTONE-IP-AND-BUILDOUT.md §4.7 already grades CSF LOW and routes it to defensive publication + the CSF trademark. This review supplies the named prior art that §7 says must precede any filing, and adds the FTO-green finding. Recommended one-line register update: cite this doc as the CSF prior-art record.
Honest scope
- Live web search, not an exhaustive professional patent database (no Derwent/PatBase); a formal FTO adds a claims-chart review by counsel.
- "Term-expired" is inferred from filing year + the 20-year term; a real expiry check accounts for maintenance-fee lapse and adjustments.
- The "no patentable mechanism" verdict is specifically about CSF's compression; the register's genuinely-novel systems items (the accountability layer, the decode/verify controller, the dilation-as-grounding-budget primitive) are graded separately there and are unaffected by this review.