JOB 131 — CODEC VERIFICATION SCOPE

REPRODUCE
From the USS workspace root, using Node.js 22 or later:
  node 131/tools/test-score-codec.cjs
The retained runner was executed successfully with Node.js 24.18.0. It performs
29 checks and writes only analysis/codec-report.json. Mutated packets, malformed
scores, synthetic PNG rasters and the BOM fixture remain in memory. It reads no
author or viewer interpretation. Hash comparisons confirm that the 26 source,
image, layout and decoder-resource files used by the run remain unchanged.

WHAT PASSED
The original score recovers with 61 actual USL1 concept tokens, 58 distinct
checked concepts, two deliberate blanks and 38 links. There are no lexical
substitutions. Every selected cell is bound to one complete tag-2 token carrying
its actual checked ID, registry, tier, snapshot head and canonical spoken RGB24
name. Full pronounceable text recovers rows, accents, links and dependency pins.
Repeated concepts retain their separate cell coordinates and roles.

The tests reject literal stand-ins for concept tokens, incorrect concept IDs or
heads, altered grammar pins, duplicate expanded JSON keys, misplaced concept
tokens, raw packet corruption, lexical substitutions, malformed score shapes,
non-string coordinates, unknown operators, altered spoken-name case or suffix,
wrong words-file pins, invalid blank accents, and a source-file UTF-8 BOM.
An all-blank reply and a reply containing checked ID 6256 recover correctly;
the legacy 12-bit encConcept helper is not used.

Both a synthetic raster and the finished sender/score.png pass composite
recovery. The CVP2 inset yields the same exact packet as a fresh source encode.
All 4,032 samples pass: 64 pixels for each of 63 cells, offsets -3 through +4 on
both axes around the centers supplied by receiver/layout.json. Selected-cell
samples equal the record anchor; blank samples equal #07121B. Changed selected
or blank samples, duplicate sample coordinates, altered sample radius and a
shifted carrier crop are rejected.

LIMITS
This is a supplied-codec, original-digital-file test. It is not an independent
codec implementation, optical recognition trial, print/scan or screenshot
experiment. The PNG reader requires the supported exact RGB8 PNG profile.
Layout coordinates are supplied context. The checks verify those sampled cores,
not every artwork pixel or independently authenticated grid geometry. USL1 and
resource hashes establish byte consistency, not authenticated authorship, truth
or agreement about meaning. No interpretation score or AI performance result is
claimed. Source definitions retain their provenance and applicable terms.

SHARED DEPENDENCIES AND PER-SCORE BYTES
Exact uncompressed filesystem byte counts for the frozen resources used here:
  receiver/vendor/ (12 files, including lexical forms/notices)     4,552,833
  receiver/state/ (five frozen state files)                       12,101,667
  receiver/dictionary.json (full records and indexes)             12,662,119
  receiver/grammar.json                                               3,994
  Four receiver resource/score/CLI code files                         36,183
  Shared resources total (23 files)                              29,356,796
This is approximately 27.997 MiB. The full dictionary projection deliberately
duplicates checked-record information already present in the source chain;
both shipped representations are charged in that total. They are reusable by
replies against the same frozen snapshot rather than re-created per score.

The original score's alternative transport/presentation files are:
  score.usl1                                                        12,499
  score.cvp2.png (56 x 56 pixels)                                     3,057
  score.words.txt                                                     1,886
  score.png (finished 2048 x 2048 artwork)                           269,448
  receiver/layout.json (artwork-specific decoding context)            6,824
These are alternative or companion representations of one score, not separate
messages whose sizes should automatically be added. The 3,057-byte CVP2 PNG is
not a self-contained vocabulary/grammar distribution. Shared resources plus
the finished artwork and its layout total 29,633,068 bytes before archive
compression. This explicit accounting excludes packaging manifests, handoff
instructions, redundant copied score files, decoded outputs, test evidence and
interpretations; it is not a claim about the complete downloadable archive size.
