The main documentation is grouped by reader intent. Implementation history and provenance are kept separately so that API and format readers do not need to navigate chronological development records.
- Command-line tool: usage, profiles, and file/error behavior.
- Composition matrix and admission history: the current profile matrix is authoritative; publication criteria, the deferred code-generation path, and numbered admission history follow it.
- C API: public ABI, lifecycle, configuration, and errors.
- Stream format: decoder-visible byte and bit representation.
- Architecture: component boundaries, streaming contracts, limits, and composition.
-
Native 64KiB position-distance rANS: exact stream/model representation, public resources and qualification gates.
-
64-KiB LZSS Contextual compression-ratio study: frozen
mozillabaseline, short-match hypothesis, and staged full-corpus admission gates without changing existing stream bytes. -
Position-distance LZSS streaming integration: context-9 incremental state, per-frame publication, workspace, staged C/CLI integration and exact-identity admission audit.
-
64-KiB LZSS Contextual short-match candidate: reserved decoder-visible identity for 3/4-byte matches, model bounds, and staged admission gates; no public codec admission yet.
-
Contextual profile application: atomic resource-profile helpers for profile-bearing contextual codecs and the complete-default rule for codecs without such helpers.
-
LZSS match-finder strategy: exact encoder-side acceleration without changing the LZSS stream format.
-
LZSS HashChain prefix mixer experiment: private zero-workspace low-bit-distribution candidate with fixed Exact, synthetic, and Silesia admission gates.
-
LZSS HashChain bucket-scaling experiment: private larger-table candidates with checked workspace, Exact identity, and staged synthetic-to-Silesia admission gates.
-
LZSS HashChain end-to-end A/B experiment: fixed all-member codec comparison of the pre-promotion and promoted builds, including archive identity, optimized-build preflight, and resumable records.
-
LZSS ordered-tree Exact strategy evaluation: completed Red-Black, Scapegoat, and deletion-focused WAVL experiments under the existing Exact contract, including their negative admission decisions.
-
LZSS Scapegoat Tree Exact: fixed two-thirds balance, bounded iterative rebuilding, workspace, diagnostics, and private staged verification.
-
LZSS WAVL Tree Exact feasibility: private deletion-heavy hypothesis, equal-workspace contract, mutation-specification gate, and early-stop criteria.
-
LZSS WAVL Tree Exact transitions: hand-checkable insertion, deletion, successor-transplant, rank-repair, and metadata-repair rules for the private candidate.
-
LZSS HashTree Exact design: deterministic hot-bucket promotion and LCP-aware ordered search.
-
LZSS Sparse HashTree large-window experiment: fixed 4/16/64-MiB Silesia matrix for bounded hot-bucket-tree reevaluation.
-
LZSS Sparse HashTree reuse gate: bounded repeated-query admission hypothesis for private hot-bucket selection.
-
LZSS Sparse HashTree reuse-gate experiment: fixed 216-point Silesia matrix isolating repeated-query admission.
-
LZSS Sparse HashTree snapshot + delta: immutable promoted snapshots with the existing chain as a bounded delta.
-
LZSS Sparse HashTree immutable snapshot experiment: fixed comparison against HashChain and mutable Sparse at reuse sixteen.
-
LZSS Sparse HashTree snapshot delta budget: exact post-query circuit breaker that demotes an over-budget immutable snapshot bucket to the existing HashChain path.
-
LZSS Sparse HashTree snapshot delta budget synthetic experiment: fixed 126-point synthetic matrix and mechanical pre-Silesia shortlist rule.
-
LZSS BinaryTree Exact 16 MiB benchmark: bounded Silesia comparison of the global AVL and HashChain strategies.
-
Silesia external benchmark profile: non-redistributed corpus handling and large-window match-finder measurement.
-
LZSS contextual 1 MiB window: additive typed-token and context variants for extended-distance experiments.
-
LZSS contextual 4 MiB window: the next additive identity, expanded decision bounds, and per-backend memory gates.
-
LZSS contextual 16 MiB window: the reserved post-0.4.0 identity, retained
7Fboundary, and explicit one-GiB aggregate-policy gate. -
LZSS contextual 64 MiB window: the reserved high-memory identity, expanded
8Fboundary, and staged backend admission contract. -
LZSS contextual rANS 4 MiB window: the second backend's descriptor, payload, and 128-MiB workspace proof.
-
LZSS contextual rANS 16 MiB window: the staged descriptor, payload, and 512-MiB workspace proof.
-
LZSS contextual tANS 4 MiB window: the third backend's table, payload, and 128-MiB workspace proof.
-
LZSS contextual tANS 16 MiB window: the staged table, payload, and 512-MiB workspace proof.
-
LZSS contextual tANS 64 MiB window: the staged variant-5 table, payload, and four-GiB workspace proof.
-
LZSS contextual Blocked Huffman 4 MiB window: the fourth backend's descriptor, payload, and 128-MiB workspace proof.
-
LZSS contextual Blocked Huffman 16 MiB window: the staged descriptor, payload, and 512-MiB workspace proof.
-
LZSS contextual Blocked Huffman 64 MiB window: the staged descriptor, payload, and four-GiB workspace proof.
-
LZSS contextual Adaptive Huffman 4 MiB window: the fifth backend's conservative payload proof and explicit 256-MiB policy.
-
LZSS contextual Adaptive Huffman 16 MiB window: the staged model-bank, payload, and one-GiB workspace proof.
-
LZSS contextual Adaptive Huffman 64 MiB window: the staged model-bank, payload, and eight-GiB workspace proof.
-
LZSS typed-token protocol: bounded dictionary-to-context value boundary.
-
Context-model contract: invertible field separation and context selection.
-
Entropy-backend contract: bounded planning, coding, and backend substitution.
-
Experimental format 2.0: reserved decoder-visible representation and hand-checkable vector.
- Benchmarks: measurement contract and benchmark usage.
- Fuzzing: bounded fuzz targets, corpora, and sanitizer workflow.
- Interoperability: CI bundles and cross-platform verification.
- Baseline readiness: current implementation and release-evidence status, with historical admission and CI evidence.
- Release process: version namespaces, evidence gates, tagging, and publication.
Numbered design decisions, source provenance, implementation-reference records, and test-vector construction records are collected under implementation records. They document how marc was produced; they are not additional public API or stream-format specifications.
- Contributing: implementation sequence, provenance, composed-profile admission, and validation requirements.
- Changelog: user-visible contents by project release.