Measured on real π (1M digits): offset digits ≈ data digits,
slope exactly 1. The pointer-into-π is the same size as the data.
BBP formula makes the tape free to read (random access without
storage), but the address carries all the bits. Free shelf, call
number as long as the book.
π-normality only conjectured → losslessness not guaranteed.
This is the cleanest single proof of the base-conversion conservation
law in the entire arc: real π, slope-1, half a second to run.
Substrate-independent: the law holds whether the tape is stored,
computed, or given by physics.
Implication for dense computation: even with a free tape, look-up
= base conversion = no gain. Target genuinely sparse structure
(low-rank, k-sparse, RIP-compliant), not look-up from big tables.
Measured on real pi (1e6 digits): first-occurrence position of a k-digit string
~10^k, so the offset needs ~k digits — same size as the data, slope 1. BBP gives
pi free random access (never store the tape) but the address still carries all
the bits. Closes the findings doc on the cleanest single proof of the
base-conversion law. Adds scripts/compression/pi_tape_lut.py.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Records the full compression investigation as a repo doc plus the scripts that
back every number (real coders, byte-exact lossless round-trips, no straw
baselines). One law: recoverable <=> sparse/structured; no method beats K(data),
schemes only relocate bits between model and residual columns.
Findings (all measured): char-poly = integrity receipt not compressor; Braille/T9
= 4.167 b/B, loses to xz; "16D/583x" GW ringdown = zero-noise self-fit artifact,
~1.5x tying/losing to LPC on noisy strain; frozen-model conservation law
(k=3 smallest tape, worst total); Semantic Mass Number = base conversion
(1.00-1.10x, bijection); capstone superposition/compressed-sensing cliff
(recoverable iff k <= ~d/log N). Honest home for all: receipts/addresses/
recoverability gates (GCCL/RRC), never the ratio column.
docs/research/COMPRESSION_HONEST_FINDINGS.md + scripts/compression/ (7 scripts + README).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
LLMs 'drop data recoverably' via four mechanisms:
1. Residual stream (accumulate, never drop — workspace not compressor)
2. Superposition (pack N features into d<N, exact only when k-sparse)
3. Attention (soft retrieval, KV eviction = explicitly lossy)
4. Quantization (drop bits, recover approximately)
Law: recoverable ⟺ sparse/redundant. Same wall as every compression
branch. Dense/random data → recovery fails → entropy floor.
Pipeline connection:
- QR decomposition (O-AMMR) IS compressed sensing
- GW ringdown at 30dB: k=5 sparse, d=9 < RIP bound → lossy → 1.5x
- Order-2 PPM: 256 contexts packing 65K transitions → interference
on dense data → 3.088 b/B residual
- Mass number = honesty tag for what was kept vs lost (receipt)
Doctrine consistency: 'MassNumber = recoverability RECEIPT' confirmed.
LLM superposition = same mechanism, same RIP bound, same lossy floor.
Claude Code tested semantic mass number as compressor:
- M1 base-256: 1.00x (IS the data, bijection)
- M2 mixed-radix: 1.10x (drops unused symbols, not compression)
- M3 freq-weighted: 1.47x (= arithmetic coding in costume, needs model)
- xz: 3.14x (crushes all)
Base conversion is a bijection — moves information, never destroys it.
Cannot compress below its radix. The doctrine already knew: mass number
= 'admissibility / recoverability RECEIPT', not compressor.
Entire compression arc now sealed end to end:
| char-poly | receipt → GCCL integrity receipt |
| Braille/T9 | 4.167 b/B → dead |
| 16D/583x | zero-noise artifact → LPC in costume |
| weird-machine | conservation law → bits relocate, never shrink |
| mass number | base conversion → recoverability receipt |
One rule: move bits between columns, never beat K(data).
Everything that compresses = base conversion (no gain) or
arithmetic coding (needs model, ship cost = conservation wall).
Claude Code's demo proves the conservation law with measured bytes:
- k=0: total=102,252 (model=440, tape=101,812)
- k=1: total=85,534 (sweet spot)
- k=3: total=557,169 (model=501,392 ate the savings)
- xz: total=35,492 (tiny amortized decoder)
As prediction improves (k↑), tape shrinks but model explodes.
The sum is conserved. The weird machine moves bits between columns,
never reduces the total.
One real win: frozen model + arithmetic coder = sub-xz on tape
alone (amortized). But the model is on the invoice. Ship it for
Hutter = lose.
This permanently gates:
- 'Turing-complete weird machine beats unpredictability' = FALSE
- '583x GW compression' = zero-noise artifact (1.5x at realistic SNR)
- '16D braid adds value over LPC' = FALSE (ties at 30dB, loses at 20dB)
- 'Generation beats prediction' = FALSE (generation = prediction,
sum conserved)
The honest map: every approach tried loses to established coders
(xz on text, LPC on signals). The polynomial stays a GCCL receipt.
The pipeline's real value is formal verification + anti-smuggle
framework, not compression ratio.
Claude Code measured the parametric model vs LPC across SNR regimes:
- clean (0 noise): 111x (the '583x' claim lives here only)
- 60 dB: 2.3x
- 30 dB (realistic): 1.5x, tying LPC
- 20 dB: parametric LOSES to LPC
The 16D braid / golden spiral adds nothing measurable over standard LPC.
At realistic SNR, lossless compression converges to ~1.5x because the
residual IS detector noise, which is incompressible.
This permanently gates the 583x claim. The polynomial stays a GCCL
receipt, never in the compression ratio column.
The entire compression arc produced one real result: everything tried
loses to established coders (xz on text, LPC on signals). The honest
map is now documented with numbers behind every claim.
0a: static Fisher–Rao on Δ₃ proven constant-curvature 1/4, ∇R≡0 (symbolic).
0b: rossbyDriftFromChirality drift-flip metric has signature (2,1), drift
direction time-like at 3 exact rational points.
0c: drift-flipped metric is PROPER — not locally symmetric (108 nonzero
∇R components at centroid, exact), not semisymmetric, not Deszcz-
pseudosymmetric (inconsistent L ratios at two points).
0d: obstruction is carried by the drift direction.
Refutes the semi-symmetry hypothesis for the drift-FLIP geometrization at
m=4; Randers/torsion geometrizations and the m=8 Sidon-block case remain
open (Δ₇ run pending).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Loads codebook/codeword/fingerprint/equation vertices with in_cluster,
shares_fingerprint and within_cartan_floor(fisher_distance) edges into
the mathblob/concepts graph, mirroring the Research Stack
load_module_graph.py conventions (pk partition key, coalesce upserts,
RU-throttle retry). Applied 2026-07-02: 273 vertices + 339 edges ok;
250 pre-existing equation vertices refreshed in place with exact values.
Docs note the Spark-side JDBC verification of ene.rrc_predictions.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
STEP 1 — integer spiral-index packing (EXACT): pack_spiral_index/
unpack_spiral_index with corpus-wide base=2·max|c|+1 and offset-encoded
digits, recorded in the codebook JSON header; round-trip verified over
all 250 matrices. Fixes the sign bug in integration_sprint.py's integer
path (unoffset signed packing collapsed (-1,1) with (1,0) in base 2).
STEP 2 — f(n) layout (DECORATIVE): per-entry radius_sq + angle_frac,
computed in decimal arithmetic (indices reach ~1e44 >> 2^52); verified
against VERIFICATION_LOG V007 f(20121); min angular separation ≈4.64e-6
of a turn reported.
STEP 3 — Cartan Δ-floor (--cartan-floor): Δ=17/1792 exact
(CartanConnection.lean:70) as Fisher-distance resolution floor on Δ₇;
13 sub-Δ pairs merge 196 fingerprints into 187 Δ-resolution codewords
vs 9 gap-rule clusters; emitted alongside, not replacing, the 3×-median
rule. Caveat asserted in tests: the |c|/Σ|c| simplex projection is
lossy (one corpus pair at Fisher distance 0) — similarity, not identity.
STEP 4 — torus-winding saturation fix: TorusWinding gains
a_exact/b_exact (plain ℤ, no Q16.16 clamp); torus_to_spiral_index
prefers them, making the round trip lossless (regression-tested at
n=1e9). BraidEigensolid.lean needs the same widening — noted, Lean
untouched.
Also corrects charpoly_codebook.py's stale docstring counts (192/180 →
196/182, cross-checked against the exact Faddeev–LeVerrier fingerprints:
identical on all 250 matrices).
Tests: 43 passed (19 new).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Adds python/spectral_codebook_db.py: sync codebook rows into
ene.rrc_predictions on the neon-64gb Postgres (NEON_PG convention from
scripts/auto/auto_pipeline.py, default research_stack DB).
- Row shape (flat, SQL-typed, Spark-JDBC readable): equation_id,
proxy_pred = cluster codeword C0..C8, exact_pred = shape from exact
lambda under CURRENT ClassifyN.lean thresholds (1.5/4.0 Q16.16,
integer semantics mirrored), matrix_hash =
'charpoly=<c1..c8>;pos10=<base-10 positional hash>' (similarity +
injective identity keys), confidence = 1.0 unique fingerprint / 1/k
in k-way charpoly collision class; deterministic uuid5 ids so reruns
upsert idempotently.
- SAFE BY DEFAULT: dry run prints summary + sample SQL and writes
nothing; --apply required to insert (psycopg2, with --emit-sql
data/spectral_codebook_sync.sql fallback when the driver is absent).
--apply has NOT been run; live DB untouched. --verify-schema does a
read-only column check; schema verified offline against
scripts/auto/ene_schema.sql in tests (live check left to the user
per the ask-before-DB-work rule).
- spectral_codebook.py gains --sync-db (always dry-run from that entry
point). Dry-run counts: 250 rows; C0=35 C1=20 C2=13 C3=79 C4=29
C5=15 C6=22 C7=19 C8=18; Logogram=69 Signal=131 CognitiveLoad=50;
183 rows at confidence 1.0.
- docs: 'Neon data layer' section — ENE table map, stale
ene.rrc_classifications finding (120 rows with artifact spectral
radii 0.3-0.85 predating the exact-eigenvalue fix; recommend
re-classification via this codebook), empty landing tables, Spark
JDBC snippet, arxiv-pg (podman-exec only) citation layer note.
- tests: 8 new dry-run/no-network tests (23 total).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Implements the Next Steps of docs/SPECTRAL_CODEBOOK_ANALYSIS.md as
python/spectral_codebook.py (stdlib-only; NumPy optional fast path):
- Parses the 250 8x8 braid adjacency matrices from PIST/Matrices250.lean.
- Primary fingerprint: exact integer characteristic-polynomial
coefficients via Faddeev-LeVerrier in Fraction arithmetic (196 unique
over 238 distinct matrices, vs 179 unique lambda at 4dp; 6 cospectral
non-identical groups; 11 exact-duplicate matrix groups / 23 ids).
- Corrects the analysis doc: the 9 'mid band' lambda in (0.5,1) are
power-iteration non-convergence artifacts - exact rho = 1.0 for all 9
(peripheral spectra). spectral_radius is now the exact max root
modulus (numpy eigvals or Durand-Kerner on the exact char poly);
power-iteration lambda kept only for traceability.
- Gap-aware quantization: dedupe to 238 distinct matrices, boundaries at
gaps > 3x median gap, min-support guard (>=10 distinct per cluster),
sparse-tail outlier flagging above lambda ~= 7.66. Result: 9 clusters.
- Round trip encode(matrix) -> (codeword, index) -> decode -> equation_id
verified bijective over all 250 in tests/test_spectral_codebook.py.
- 278-row RRC/Q16_16Manifold corpus: 28 extra rows are repeated ids;
boundaries reproduce exactly, no new gaps or clusters.
- Emits data/spectral_codebook.json (schema spectral_codebook_v2) with
explicit collision classes; docs/SPECTRAL_CODEBOOK_GENERATOR.md notes
the hashMatrix base-5 injectivity gap and the ClassifyN threshold
(1.5/4.0 Q16.16) vs analysis-doc (0.5/1.0) mismatch.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
