# RRLS Studio v0.12 — Exact Operator Market

## Release correction

Earlier bundles contained newer engine modules but the classic number-pattern GUI still
displayed `RRLS Studio v0.7`, and the default `launch_windows.bat` opened that legacy
window directly.

v0.12 fixes this:
- all visible GUI version labels are v0.12;
- the default launcher opens `RRLS Research Studio v0.12`, a hub for all engines;
- the Number Pattern Studio remains available as a dedicated tool;
- Universal Corpus, Common Crawl and Exact Operator Market labs are directly launchable.

## New operator-market hypothesis

For each local block RRLS now searches a reversible relation/coordinate representation:

- RAW
- DELTA-1
- XOR-1
- BITPLANE

and lets the transformed residual compete across:

- zlib
- Brotli
- LZMA

The winner must pay its operator/coder ID and must reconstruct the original block
bit-identically.

This is not a claim that the standard residual coders are RRLS inventions. The research
question is whether local relational coordinate selection creates additional compression
after ordinary coder selection has already been accounted for.

## Held-out uncontrolled real-data benchmark

Input:
- 8,377,100 held-out bytes
- 1,569 independent blocks
- actual Python/library/docs/icons/binary material from the execution runtime

Results:

| System | Compression |
|---|---:|
| zlib | 2.39609x |
| LZMA | 2.51479x |
| Brotli | 2.51871x |
| Per-block residual-coder market, RAW coordinates only | 2.61486x |
| RRLS relation + residual-coder market | 2.62587x |

Decomposition:
- coder-selection gain over best single coder:
  **3.817%**
- additional gain caused specifically by reversible RRLS coordinate transforms:
  **0.421%**
- total gain over best single coder:
  **4.254%**

The transform contribution is small but positive on this held-out set. It therefore
survives this benchmark, while many previous RRLS operator ideas were rejected.

## Relation vocabulary over relation choices

The winning market IDs themselves are not independent.

Measured on the held-out block sequence:

- H(operator) = 1.6879 bits/block
- H(operator | previous operator) =
  1.5298 bits/block

Knowing the preceding relation reduces operator uncertainty by approximately
**9.36%**.

This does not imply a fundamental physical language; it establishes that the chosen
relation sequence contains higher-order reusable structure.

## Compact relation metadata

The `.rrls12` format now recursively compresses its own description layer:
- operator IDs are packed into 4-bit symbols or zlib-compressed when cheaper;
- original and residual lengths use variable-length integers;
- residual payloads are stored contiguously.

All tested compact streams round-trip exactly.

## Exactness

Round-trip tests include:
- OS-entropy bytes
- long runs / periodic material
- structured numeric u32 data
- repetitive text-like data

Every test reconstructed the exact original SHA-256 / byte stream.

## Important limits

1. A portfolio of existing compressors is not itself a novel universal solver.
2. The measured +0.421% transform gain is dataset-dependent and needs replication.
3. The current transform vocabulary is tiny.
4. Operator selection is currently block-local rather than globally optimized.
5. Common Crawl binary WET access remains blocked in the packaging runtime, so no live
   Common Crawl compression number is claimed here.
6. The strongest next experiment is frozen transform/operator transfer across separate
   real corpora and separate Common Crawl months.

The research target remains:

    representation search
    -> exact local relation
    -> measured cost reduction
    -> relation token
    -> relation tokens become the next coordinate alphabet
    -> repeat only when the next layer pays for itself