Research-Stack/6-Documentation/docs/specs/DECODER_FACING_RECONSTRUCTION_CORE.md

Decoder-Facing Reconstruction Core

Status: Draft v0.1 Date: 2026-05-08 Scope: compression-core interface boundary for logogram, OMCF, PIST, residual, and verifier pipelines Claim state: terminology and admission contract; not a benchmark result, decompressor implementation, or Hutter Prize claim

1. Purpose

This document defines the interface boundary for the compressed core.

Current architecture shift:

6-Documentation/docs/specs/LAW_GATED_RECONSTRUCTION_CORE_SHIFT.md

Math consistency review:

6-Documentation/docs/specs/RECONSTRUCTION_CORE_MATH_REVIEW_2026_05_09.md

Project memory promotion:

shared-data/data/stack_memory_promotions/reconstruction_core_ladder_memory_receipt.json

That shift makes this document the inspection boundary for all compression routes: route priors, glyphs, physics corpora, proof corpora, DNA-style filters, and quantum/transfold kernels remain HOLD until they provide deterministic replay, declared residual repair, byte accounting, receipts, and control-filter evidence.

The compressed representation is not assembly language, source code, bytecode in the ordinary human-inspectable sense, or a prose notation for the payload. It is a decoder-facing reconstruction core.

Canonical phrase:

This is not assembly. It is a lawful reconstruction core.

Expanded:

The compressed core should not be treated as assembly language, source code,
or a human-readable intermediate representation. It is a decoder-facing
reconstruction core: its validity is established by deterministic replay,
residual repair, receipts, and byte-exact output rather than by direct
readability.

This is not anti-inspection. It is a consequence of the compression process.
Inspection moves to the lawful interfaces: codec specification, replay rules,
receipts, residuals, hashes, benchmarks, and reconstructed bytes.

2. Interface Layers

The system separates human-facing handles from decoder-facing reconstruction material.

Layer	Purpose
human surface	logograms, prose labels, diagrams, control names, documentation
proposal surface	candidate laws, seeds, type witnesses, Mass Number basins
reconstruction core	compressed replay material, not human-readable by contract
receipt surface	hashes, O-AMVR/O-AVMR receipts, replay checks, residual policy
output surface	byte-exact reconstructed payload

The core is inspectable only through the declared interfaces. A human may inspect the core bytes, but direct readability is not a validity criterion.

3. Validity Rule

A reconstruction core is valid only when:

decode(core, residual, rules) == original_bytes
receipt verifies deterministic replay
residual policy is declared
all counted bytes satisfy the admission law

The core is invalid when:

direct inspection looks plausible
but replay fails

or:

semantic grouping looks elegant
but residual cost erases gain

4. Relationship To OMCF / PIST

The reconstruction core may contain OMCF and PIST material:

OMCF = structure + semantic mass + deterministic expansion + braid receipt
PIST = lawful surface transform over the carrier

But those internal packets are still judged by replay:

project -> mass -> braid -> PIST -> receipt -> residual -> byte-exact output

No internal notation is promoted because it is readable, elegant, or mnemonic. Promotion requires deterministic reconstruction and positive byte law.

5. Proposal / Verifier Split

Stochastic systems may propose reconstruction laws, but they are not the trust boundary.

model proposes
parser structures
verifier replays
residual repairs
byte law admits or rejects

The model may wander. The verifier does not.

Prethinker Prior

The Prethinker computational-epistemics prior sharpens this boundary for language-derived state:

LLM constructs semantic workspace
deterministic mapper admits operations
durable state remains behind the admission gate

For this stack:

model proposal != accepted atom
source envelope != payload truth
counterfactual lane != durable fact lane
structured absence != missing data
selector decision != corpus mutation

Local source manifest:

6-Documentation/docs/provenance/PRETHINKER_COMPUTATIONAL_EPISTEMICS_SOURCES.cff

Local packet generator:

4-Infrastructure/shim/prethinker_computational_epistemics_prior.py

Mass Equation Distill Receipt

The local mass-equation distill is admitted only as a coverage/routing prior, not as a total all-mathematics claim.

Current receipt:

3-Mathematical-Models/equations_parquet_tagged/mass_equations_unified_receipt.json

Human summary:

3-Mathematical-Models/equations_parquet_tagged/mass_equations_unified_receipt.md

Receipt generator:

4-Infrastructure/shim/mass_equation_distill_receipt.py

The receipt records row counts, schema, artifact hashes, source/domain coverage, feature counts, duplicate-equation hash statistics, and exclusions. Its claim boundary is explicit: the artifact is useful for OMCF/PIST routing and candidate-law discovery, but it does not prove equations, does not verify physics, does not assert full corpus coverage, and does not establish a compression benchmark.

RRC projection receipt:

3-Mathematical-Models/equations_parquet_tagged/mass_equations_rrc_projection_receipt.json

RRC projection table:

3-Mathematical-Models/equations_parquet_tagged/mass_equations_rrc_projection_table.csv

The RRC projection maps unique mass-equation hashes into lawful routing shapes and intentionally keeps under-receipted rows in HOLD. This projection is a route atlas and negative-control surface, not a proof atlas.

The Well Physics-Dynamics Prior

Polymathic AI's The Well is admitted only as an external replay/route prior for field-dynamics probes. It is not vendored and is not a compression benchmark claim.

Source overview:

https://polymathic-ai.org/the_well/datasets_overview/

Local route registry:

shared-data/data/nspace_bulk_routes/nspace_bulk_dataset_route_receipt.json

Local source manifest:

6-Documentation/docs/provenance/NSPACE_BULK_DATASET_ROUTE_SOURCES.cff

For this stack, The Well maps to:

uniform HDF5 grids -> scalar/vector/tensor fields -> boundary-condition lanes
-> PIST/OMCF replay candidates -> residual/rollout stress tests -> HOLD/ADMIT

Use is metadata-first and tiny-slice-first. Full-corpus ingest requires dataset-specific terms, storage budget receipts, field/time/trajectory subsetting, and byte-accounted replay checks.

Symbolic Regression / SRBench Prior

SRBench and ParFam are admitted only as external formula-reconstruction priors for candidate-law discovery, negative controls, and replay scoring. They are not vendored and do not establish a local benchmark result.

Source surfaces:

https://cavalab.org/srbench/datasets/
https://arxiv.org/html/2310.05537

Local route registry:

shared-data/data/nspace_bulk_routes/nspace_bulk_dataset_route_receipt.json

For this stack, SRBench/ParFam maps to:

ground-truth formulas -> candidate-law packets -> RRC/OMCF route classes
black-box datasets -> negative controls -> residual scoring -> HOLD/ADMIT
rational families -> deterministic replay candidates -> receipt checks

Ground-truth formula tasks may seed exact replay fixtures. Black-box regression tasks remain negative controls unless a proposed law replays deterministically, declares residual policy, and passes byte-accounted admission checks.

Strong-Source Route Shortlist

The following sources are admitted as separate HOLD-first route surfaces:

Source	Route role
The Well	large-scale uniform-grid physics-dynamics replay corpus
PDEBench	canonical PDE-family fixtures and baseline residual curves
RealPDEBench	real-measurement calibration for simulation residuals
ERA5 / NWP / FV3	weather-system conservation, assimilation, ensemble, and stability route priors
MeshGraphNets	irregular mesh / goxel-like topology substrate
LeanDojo + mathlib	formal proof and tactic-state routing corpus
DLMF + Feynman/SRBench	equation compression and symbolic law recovery
NuminaMath	broad math-reasoning proposal curriculum

Local route registry:

shared-data/data/nspace_bulk_routes/nspace_bulk_dataset_route_receipt.json

The routing contract is:

external source -> metadata packet -> tiny replay fixture -> residual receipt
-> RRC/OMCF candidate -> HOLD unless deterministic replay and byte law pass

NuminaMath and other informal reasoning corpora may propose candidates, but do not promote facts or proofs. LeanDojo/mathlib may route proof obligations, but local Lean replay remains the proof boundary. PDE and mesh corpora may route field dynamics and topology fixtures, but benchmark claims require explicit baseline, split, metric, and storage receipts.

Parallel Metaprobe And Replay Queue

The current parallel metaprobe launch is recorded as a HOLD receipt:

4-Infrastructure/shim/parallel_metaprobe_runs/20260509T053755Z/parallel_metaprobe_launcher_receipt.json

Receipt hash:

aa7a1d61131e47544fb25f9cbacfa5ee2a888fb67feca2dfa3f82b92af38373b

The run used 6 workers and passed 11 route-prior lanes. It is a launcher receipt only; it does not promote any source, proof, compression benchmark, or dataset ingest.

Status note:

ADMIT_FIXTURE != ADMIT

ADMIT_FIXTURE means a tiny local fixture passed its local replay and byte checks. It does not promote the whole receipt, source, codec, corpus, theorem, or benchmark lane. Promotion still requires the top-level receipt decision and control-filter stack to pass.

The replay fixture queue is:

shared-data/data/replay_fixture_queue/replay_fixture_queue_receipt.json

Current queue receipt hash:

914e6564fe33b56dd989016fe3c22369e55956a13a95d408a940d04cdd9bbf72

Human summary:

shared-data/data/replay_fixture_queue/replay_fixture_queue.md

Current queue head:

1. SRBench / ParFam
2. DLMF / Feynman Symbolic Regression
3. PDEBench
4. The Well

The first symbolic-law replay harness is:

shared-data/data/symbolic_law_replay/symbolic_law_replay_receipt.json

It currently records 3 tiny fixtures:

ADMIT_FIXTURE: 1
HOLD_DIAGNOSTIC: 2

This is not an SRBench result. It is the first local replay check for the formula-reconstruction route: exact candidate formulas may pass as fixtures, negative controls must remain HOLD, and small examples that fail byte law stay diagnostic even when replay is exact.

The first PDE-style replay harness is:

shared-data/data/pde_tiny_replay/pde_tiny_replay_receipt.json

Receipt hash:

3871a238ae84450ce6081b6d74b0120b4624c75edf8af1625c61d90ff02c74c9

It currently records 3 tiny local fixtures:

ADMIT_FIXTURE: 1
HOLD_DIAGNOSTIC: 2

This is not a PDEBench ingest or benchmark result. It is a no-download micro-fixture for deterministic advection replay: exact periodic replay may pass as a fixture, wrong-boundary controls must remain HOLD, and tiny exact examples that do not pay byte law stay diagnostic.

The first The Well-style schema probe is:

shared-data/data/the_well_tiny_probe/the_well_tiny_schema_probe_receipt.json

Receipt hash:

97ebac7697555f10b45b3e6e1a1547a28eaaf02d28ab16875607d8a8af14bfb3

It currently records 3 tiny metadata fixtures:

ADMIT_FIXTURE: 1
HOLD_DIAGNOSTIC: 2

This is not The Well data ingest, HDF5 vendoring, or a benchmark result. It is a metadata-only schema probe: scalar/vector field rank, axes, boundaries, dtype, shape, schema hashes, residuals, and byte-law accounting must replay before any field slice is admitted.

The weather-systems borrowed-math prior is:

shared-data/data/weather_systems_math_prior/weather_systems_math_prior_receipt.json

Human summary:

shared-data/data/weather_systems_math_prior/weather_systems_math_prior_receipt.md

Local source manifest:

6-Documentation/docs/provenance/WEATHER_SYSTEMS_MATH_PRIOR_SOURCES.cff

Receipt hash:

fe3c8d27d24bd45af855faa062077467d9dfd8079c4ec5ab10c80a1101b82eac

It records the no-download weather math filter:

weather_state -> transport/replay kernel + residual -> repaired state
Repair(Replay(K,Theta,Pi),R) == S

The tiny local replay currently records:

ADMIT_FIXTURE: 1
HOLD_DIAGNOSTIC: 2

This is not an NWP forecast, ERA5 ingest, atmospheric-model validation, or compression benchmark. It is a borrowed-math prior for conservative transport, boundary-condition residuals, CFL/stability diagnostics, data-assimilation innovation, ensemble spread, and residual-growth routing. Exact repair and positive byte law remain the admission boundary; weather terms remain diagnostics unless normalized and receipted.

The first MeshGraphNets-style topology probe is:

shared-data/data/meshgraphnets_tiny_probe/meshgraphnets_tiny_topology_probe_receipt.json

Receipt hash:

3bfa1dc795aed81c1b40d3e6d1937aa01dd807156fd687f0c7c015b5c85b3b25

It currently records 3 tiny topology fixtures:

ADMIT_FIXTURE: 1
HOLD_DIAGNOSTIC: 2

This is not MeshGraphNets data ingest, trajectory vendoring, or a benchmark result. It is a no-download topology probe: canonical edges, faces, boundary nodes, degree sequence, topology hashes, tiny message-pass replay, residuals, and byte-law accounting must replay before any irregular mesh slice is admitted.

The quantum cognitive-load transfold prior is:

shared-data/data/quantum_cogload_transfold/quantum_cogload_transfold_receipt.json

Human summary:

shared-data/data/quantum_cogload_transfold/quantum_cogload_transfold_receipt.md

Receipt hash:

f5dc5da184d2a920d317f08f39e02a37257189484ddf47a1220c053904c48c76

It records the merged equation:

L_QCog(H_class,rho,Omega) =
  R_Sigma_Q({C_k_Q R_k_Q(x_k_Q(H_Q,U_Q,rho,Omega);theta_k_Q)
  lambda_phi^D_f B_k_Q(Omega)} for k in {I,E,G,R,M}; theta_Sigma_Q)

H_Q = (Pauli o C_n o Q_hbar)(H_class) = sum_alpha c_alpha P_alpha

The tiny Pauli replay currently records:

ADMIT_FIXTURE: 1
HOLD_DIAGNOSTIC: 2

This is not a quantum-algorithm result, cognitive-science validation, proof, or compression benchmark. It is a route prior for Pauli-string cognitive-load features: coefficient entropy, support size, noncommutative coupling burden, entanglement placeholder, truncation residual, replay depth, measurement burden, and semantic/basis mismatch.

The quantum-basis compression objective receipt is:

shared-data/data/quantum_basis_compression_objective/quantum_basis_compression_objective_receipt.json

Human summary:

shared-data/data/quantum_basis_compression_objective/quantum_basis_compression_objective_receipt.md

Receipt hash:

c6d23f33eb89ab26bfb1135e7f102dbbe94fa710b25b6685db715508d5ab79be

It records the compression implication:

S -> (K_Q, Theta_Q, R, Pi) -> S_hat
Decode(K_Q, Theta_Q, R, Pi) == S
epsilon_byte = ||S - S_hat||_0 = 0

and the counted objective:

J_compress =
  |D| + |K_Q| + |Theta_Q| + |R| + |Pi|
  + lambda_T D_replay + lambda_L L_decode

G_Q = |S| - (|D| + |K_Q| + |Theta_Q| + |R_Q| + |Pi_Q|)

The tiny generator/residual replay currently records:

ADMIT_FIXTURE: 1
HOLD_DIAGNOSTIC: 2

This is not a quantum-compression result, Hutter claim, or proof. It is an admission objective: a basis/kernel is useful only when lawful reconstruction plus parameters, protocol, and residual replay byte-exactly and pay positive byte law.

The enwiki8 wiki-logogram slice probe is:

shared-data/data/enwiki8_wiki_logogram_probe/enwiki8_wiki_logogram_probe_receipt.json

Human summary:

shared-data/data/enwiki8_wiki_logogram_probe/enwiki8_wiki_logogram_probe_receipt.md

Encoded core:

shared-data/data/enwiki8_wiki_logogram_probe/enwiki8_wiki_logogram_probe_core.wlg1

Receipt hash:

d93c768e02781fc197e6134238808197f7dbb69be590ece90f7a6a1853c54fac

It records a bounded local enwik8 slice probe:

corpus: shared-data/corpora/enwik8
offset: 1406
length: 4096
Decode(wiki_logogram_core) == slice_bytes

The tiny local replay currently records:

ADMIT_FIXTURE: 1

Accounting:

raw slice: 4096 bytes
encoded core: 3453 bytes
packet estimate: 4025 bytes
zlib-9 baseline: 1365 bytes
bz2-9 baseline: 1469 bytes
lzma-9 baseline: 1420 bytes

This is not a Hutter Prize submission, full enwiki8 result, or evidence of compression competitiveness. It is a grammar-admission probe for MediaWiki/XML replay atoms such as wiki links, templates, XML tag pairs, text-attribute pairs, and exact empty tags. The ordinary compressors remain much smaller on this slice; the probe only shows byte-exact replay and a positive local packet estimate under truncated per-atom receipts.

The enwiki9 logogram targeter bundle receipt is:

shared-data/data/enwiki9_logogram_targeter/enwiki9_logogram_targeter_receipt.json

Human summary:

shared-data/data/enwiki9_logogram_targeter/enwiki9_logogram_targeter_receipt.md

Source bundle:

/home/allaun/Documents/ingest/enwiki9_logogram_target.zip

Receipt hash:

f77d7404ed76c1dd7697b5a02ddaf9ad02d073d40c09a5d747d8dc678ffbae9f

It records import and replay of the uploaded enwiki9 slice-targeting harness:

demo slices: 3
demo roundtrip: true
sample source: /home/allaun/Downloads/data/enwik9_data/1234567
sample bytes: 20532
sample slices: 4
sample roundtrip: true

Current accounting:

demo raw/core/packet: 948 / 1115 / 1491 bytes
sample raw/core/packet: 16384 / 17290 / 18678 bytes

This is not a canonical enwik9 corpus run, Hutter/LTCB submission, or compression-competitiveness claim. The uploaded bundle is a slice targeter, not the 1,000,000,000-byte corpus. The available local sample is 20,532 bytes, and the current grammar expands it; the useful result is byte-exact replay plus slice-class targeting for future dictionary promotion.

The enwiki9 fixed XML/MediaWiki dictionary probe is:

shared-data/data/enwiki9_logogram_xml_dict_probe/enwiki9_logogram_xml_dict_probe_receipt.json

Human summary:

shared-data/data/enwiki9_logogram_xml_dict_probe/enwiki9_logogram_xml_dict_probe_receipt.md

Receipt hash:

f9d2eb3bca3985bcc88c459b357f25fe0e194bb7c2e00385ed28b93a99357c4a

It records the v2 dictionary-promotion test:

fixed tag entries: 25
pair tag entries: 12
attribute tag entries: 9
motif entries: 11
dictionary bytes: 1064

Current accounting:

demo raw/core/packet/global-delta:
  948 / 767 / 1011 / -1127 bytes

local sample raw/core/packet/global-delta:
  16384 / 16177 / 17141 / -1821 bytes

Compared to v1:

demo v1 raw/core/packet: 948 / 1115 / 1491 bytes
local sample v1 raw/core/packet: 16384 / 17290 / 18678 bytes

This is still HOLD: exact replay passes and delta_core flipped positive, but packet and global deltas remain negative after receipt stubs and dictionary bytes are counted. The local sample is also a noncanonical HTML file, not enwik9. The valid conclusion is that fixed grammar promotion works as a core shrink step; it is not yet compression admission.

The enwiki9 receipt-aggregation probe is:

shared-data/data/enwiki9_logogram_receipt_aggregation_probe/enwiki9_logogram_receipt_aggregation_probe_receipt.json

Human summary:

shared-data/data/enwiki9_logogram_receipt_aggregation_probe/enwiki9_logogram_receipt_aggregation_probe_receipt.md

Receipt hash:

5b264b0fedf55f48bf1e790d2375bc4504b353e302a76677e5447ab39e8565d3

It records the v3 receipt-mode test:

per-atom receipt stubs -> slice_root_v1
slice receipt root bytes: 32
protocol ID bytes per slice: 4
dictionary bytes still counted globally: 1064

Current accounting:

demo raw/core/v3-packet/global-delta:
  948 / 767 / 875 / -991 bytes

local sample raw/core/v3-packet/global-delta:
  16384 / 16177 / 16321 / -1001 bytes

Compared to v2:

demo packet delta: -63 -> 73 bytes
local sample packet delta: -757 -> 63 bytes

This is still top-level HOLD: exact replay passes and packet deltas are now positive under slice-level receipt aggregation, but global deltas remain negative once the fixed dictionary is counted. The local sample is also still a noncanonical HTML file, not enwik9. The valid conclusion is that receipt aggregation fixes the packet-overhead failure for these fixtures; dictionary amortization over canonical slices remains unproven.

The enwiki9 dictionary-amortization probe is:

shared-data/data/enwiki9_logogram_dictionary_amortization_probe/enwiki9_logogram_dictionary_amortization_probe_receipt.json

Human summary:

shared-data/data/enwiki9_logogram_dictionary_amortization_probe/enwiki9_logogram_dictionary_amortization_probe_receipt.md

Receipt hash:

6eb32944aaba661e6b7d964b50d758b21a3b992f60587ab4a65b7a692b048907

It records the v4 accounting-scope test:

PASS -> ADD -> PAUSE -> SUBTRACT
timestamp role: metadata_only
generated_at_utc included in receipt hash: false
dictionary bytes: 1064
slice size: 4096
byte limit per source: 131072

Current noncanonical fixture accounting:

1234567 local HTML:
  raw/core/v3-packet/global-delta:
    20532 / 20282 / 20498 / -1030 bytes
  verdict: HOLD_GLOBAL

fawiki XML head:
  raw/core/v3-packet/global-delta:
    131072 / 125732 / 126884 / 3124 bytes
  verdict: ADMIT_FIXTURE

jawiki XML head:
  raw/core/v3-packet/global-delta:
    131072 / 130700 / 131852 / -1844 bytes
  verdict: HOLD_PACKET

viwiki XML head:
  raw/core/v3-packet/global-delta:
    131072 / 130152 / 131304 / -1296 bytes
  verdict: HOLD_PACKET

This is still top-level HOLD: one noncanonical MediaWiki XML fixture crosses the global-delta gate after dictionary amortization, but no canonical enwik9 slice, baseline-compressor comparison, full protocol package, or corpus-scale Hutter/LTCB accounting has passed. The valid conclusion is narrower: the same frozen v2 dictionary and v3 receipt mode can amortize on at least one local MediaWiki XML fixture while failing on others.

The enwiki9 canonical-slice baseline probe is:

shared-data/data/enwiki9_logogram_canonical_baseline_probe/enwiki9_logogram_canonical_baseline_probe_receipt.json

Human summary:

shared-data/data/enwiki9_logogram_canonical_baseline_probe/enwiki9_logogram_canonical_baseline_probe_receipt.md

Receipt hash:

8159af52db8a90cd599ab6f279ce090e08e507fb2673bde352a3bb163e1d8452

It records the v5 provenance and baseline test:

PROVENANCE -> PASS -> ADD -> PAUSE -> SUBTRACT -> BASELINE
codec_frozen_from: v4
encoder_changed: false
clock_participates_in_hash: false
canonical enwik9 size requirement: 1,000,000,000 bytes

Current local fixture accounting:

input: /home/allaun/Downloads/data/enwik9_data/1234567
input size: 20532 bytes
canonical claim: fixture
provenance decision: FIXTURE

raw/core/packet/global-delta:
  20532 / 20248 / 20284 / -816 bytes

baselines:
  zlib-9: 7008 bytes
  bz2-9: 7510 bytes
  lzma-9: 6672 bytes
  zstd-19: 6800 bytes

decision: HOLD_GLOBAL

This is not canonical enwik9 evidence. It is a v5 smoke receipt showing that the frozen codec, provenance gate, duplicate-window guard, clockless event chain, and baseline accounting run correctly on the existing local fixture. The fixture preserves the ladder result: replay passes and packet delta remains positive, but global delta is negative after dictionary accounting and ordinary compressors are much smaller.

The language surface-ambiguity negative-control receipt is:

shared-data/data/language_surface_ambiguity_negative_control/language_surface_ambiguity_negative_control_receipt.json

Human summary:

shared-data/data/language_surface_ambiguity_negative_control/language_surface_ambiguity_negative_control_receipt.md

Receipt hash:

f8dfa1611b6aa4f7dfea86f3d8d5cd89654fa1b0bb57a4d380841fba9be5005b

It records two language-law negative controls:

flown_by_cancellation -> HOLD_DERIVATION
buffalo_surface_collision -> HOLD_SURFACE_COLLISION

The first fixture marks the "grew/grown = flew/x" derivation as invalid even when it happens to land on the lexically correct output. Correct output is not proof of a lawful replay path.

The Buffalo fixture handles repeated word surfaces by typed role:

city_modifier
plural_noun_subject
plural_noun_relative_subject
transitive_verb_relative
transitive_verb_main
plural_noun_object

The surface token may be reused, but it cannot be collapsed into one untyped atom unless role, position, case, and replay order are preserved or carried as residual. This is not an English model or compression claim; it is a guardrail against analogy leakage and same-surface role collision.

The reconstruction-core ladder project-memory receipt is:

shared-data/data/stack_memory_promotions/reconstruction_core_ladder_memory_receipt.json

Human summary:

shared-data/data/stack_memory_promotions/reconstruction_core_ladder_memory.md

Receipt hash:

4364e9c1092020660f4ff3a64804ebc88db7136433f049ddaff9ac2880d29b39

It promotes only a compact memory pointer:

memory key: reconstruction_core_ladder_2026_05_09
lawful status: HOLD_TOP_LEVEL
next action: find/verify canonical enwik9, then run frozen v5 on canonical slices

This is not a private assistant memory write and not a compression promotion. It records receipt paths, hashes, statuses, guardrails, claim boundary, and next action pointer under the local memory-write rule.

The DNA-filtered codec objective receipt is:

shared-data/data/dna_codec_filter/dna_codec_filter_receipt.json

Human summary:

shared-data/data/dna_codec_filter/dna_codec_filter_receipt.md

Local source manifest:

6-Documentation/docs/provenance/DNA_CODEC_FILTER_SOURCES.cff

Receipt hash:

ae9b5fa3b54ef88a5035bf234b11745d6644dbfa2c046481077c16c57339b846

It records the analogy-bounded DNA codec filter:

S = Repair_R(Regulate_B_DeltaG(Replay(K,Theta,Pi)))
Repair(Replay(K,Theta,Pi),R) == S
epsilon_byte = 0

The tiny local replay currently records:

ADMIT_FIXTURE: 1
HOLD_DIAGNOSTIC: 2

This is not a DNA model, biological-data ingest, thermodynamic simulation, or compression benchmark. It is a codec filter prior: conserved motif gain, local binding compatibility, repair layering, mutation-budget pressure, and replay curvature may influence routing, but exact repair and positive byte law remain the admission boundary.

The logogram-DNA codec objective receipt is:

shared-data/data/logogram_dna_codec/logogram_dna_codec_receipt.json

Human summary:

shared-data/data/logogram_dna_codec/logogram_dna_codec_receipt.md

Receipt hash:

381ee58394408986d1971edf6ac388483786cc89fef7de982cefddcc8bc99f87

It records the Omindirection/GCCL-filtered version:

S = Repair_R(Regulate_B_DeltaG(Replay_Pi(Gamma)))
a_i = (p_i,h_i,d_i,chi_i,phi_i,x_i,tau_i,r_i,g_i,rho_i,delta_i)
payload != glyph != rendered layout

The tiny local replay currently records:

ADMIT_FIXTURE: 1
HOLD_DIAGNOSTIC: 1
QUARANTINE_DIAGNOSTIC: 1

This is not a renderer-correctness result, global logogram compression result, or biological claim. It is a law-gated symbolic-genome prior: glyph codons can route reconstruction only when payload, direction, chirality/phase, placement, residual, receipt, and adapter gates preserve byte-exact recovery or explicitly route to HOLD / QUARANTINE.

The LeanDojo/mathlib proof-boundary replay receipt is:

shared-data/data/lean_proof_replay/lean_proof_replay_receipt.json

Human summary:

shared-data/data/lean_proof_replay/lean_proof_replay_receipt.md

Local theorem fixture:

0-Core-Formalism/lean/Semantics/ExtensionScaffold/Compression/ProofReplay.lean

Receipt hash:

9855099f61256bb7e3ec30e298a32119212e3842949ae3ae0281676d4ef5851d

The local replay observed:

countedBytes repeatKernelFixture = 94
byteGainPositive repeatKernelFixture = true
admitCandidate repeatKernelFixture = true
admitCandidate residualHeavyFixture = false
admitCandidate nonExactFixture = false

Verification commands:

lake build ExtensionScaffold.Compression.ProofReplay
lake env lean ExtensionScaffold/Compression/ProofReplay.lean
lake build

All three completed successfully in the local Lean project. This is not a LeanDojo benchmark result, mathlib coverage claim, or external theorem proof. It is a proof-boundary fixture: external proof corpora may route obligations, but only local Lean replay promotes proof state.

Forward Foundation Equation Compiler

The forward-foundation compiler is the current trust boundary for equation atoms:

6-Documentation/docs/specs/FORWARD_FOUNDATION_EQUATION_COMPILER.md

Local generator:

4-Infrastructure/shim/foundation_forward_equation_compiler.py

Current receipt:

shared-data/data/foundation_forward_equation_compiler/foundation_forward_equation_compiler_receipt.json

Receipt hash:

b9c26ec185772576b27a9d176300ebe053edaae3b49650e9e2581969dda1b278

Canonical rule:

No backward trust chain. Only forward admissible generation.

The foundation set is:

F0 = {O4, SD, MN, gamma_star, H_dV, Omega, Lambda, A}

The shell equation is:

SD = L4(O4) + L3(Rg3) + chi0 + U4 + E_HD + U_under

Human theorem labels, citation chains, equation names, expert names, and logogram names are routing hints only. A trusted equation object must compile forward from F0, declare residuals, close under chi0, pay projection and budget costs, and carry a recomputable receipt. Anything that cannot compile from the foundation kernel remains HOLD, QUARANTINE, U_under, or NaN0.

The current decision is:

ACCEPT_CONTRACT_HOLD_RESULTS

This means the compiler contract itself is admitted as a local rule, but it does not promote external equations, theorem labels, or benchmark claims.

Origin metadata is explicitly not authority:

Origin may inspire. Only closure admits.
No vibes-to-axioms pipeline without a receipt.

Historical era, institution, biography, private intuition, altered-state suspicion, and aesthetic elegance may route a candidate, but cannot promote it.

Buoyancy Added-Mass Mobius Fixture

The buoyancy added-mass Mobius fixture is the first small physics equation atom compiled through the forward-foundation style:

4-Infrastructure/shim/buoyancy_added_mass_mobius_fixture.py

Current receipt:

shared-data/data/buoyancy_added_mass_mobius/buoyancy_added_mass_mobius_receipt.json

Human summary:

shared-data/data/buoyancy_added_mass_mobius/buoyancy_added_mass_mobius.md

Receipt hash:

ad2dada2df046f6576f8831bbdc72d86f0a7b88ca9ab25712ece84ad8abadfcc

It compresses the classical early-time added-mass expression:

a = g * (rho_o - rho_m) / (rho_o + C*rho_m)

into the Mass-Number Mobius form:

MN_rho = (rho_o - rho_m) / (rho_o + rho_m)
alpha_C = 2 / (1 + C)
kappa_C = (1 - C) / (1 + C)
lambda_BAM(MN_rho, C) =
  g * alpha_C * MN_rho / (1 + kappa_C * MN_rho)

Inverse:

MN_rho = (a/g) / (alpha_C - kappa_C*(a/g))

The fixture checks exact rational equivalence with the expanded form and exact rehydration back to MN_rho for:

sphere, rho_o/rho_m = 2, C = 1/2 -> a/g = 0.4
cylinder perpendicular to axis, rho_o/rho_m = 2, C = 1 -> a/g = 1/3
light sphere limit probe, rho_o/rho_m = 0, C = 1/2 -> a/g = -2

Decision:

ACCEPT_FIXTURE_WITH_BOUND_CORRECTION

This is not a new fluid theorem or experimental result. It is an algebraic compression fixture and logogram candidate. The bound claim is explicitly corrected: |a| <= g holds for the heavier/sinking branch or for suitable C >= 1 cases, but not for every density branch in the ideal added-mass model. Drag, vorticity, and boundary effects remain declared residual lanes.

Mass Number Transform Registry

The Mass Number transform registry is:

4-Infrastructure/shim/mass_number_transform_registry.py

Current receipt:

shared-data/data/mass_number_transform_registry/mass_number_transform_registry_receipt.json

Human summary:

shared-data/data/mass_number_transform_registry/mass_number_transform_registry.md

Receipt hash:

b215abe8cca08253dd62a2c2e84ff1f90fbd8e7eb5b2bb02d60dec39bbea2b9c

It records reusable Mass-Number-able transform families:

MN(a,b) = (a-b)/(a+b)
MN(a,b) = tanh(0.5*ln(a/b)) for positive a,b

Accepted exact-kernel opcodes currently include:

MN
MN_RATIO_INV
MN_MOBIUS_LOAD
MN_SPLIT
MN_REDUCED
MN_PAIR_PRODUCT
MN_BLEND
MN_REFLECT
MN_TRANSMIT_POWER
MN_BINARY_P
MN_ELASTIC_1D

MN_BINARY_ENTROPY remains HOLD_ANALYTIC until log base, numerical precision, and approximation/error policy are receipted.

Decision:

ACCEPT_REGISTRY_WITH_HOLD_ANALYTIC

This is a compression/logogram registry, not a theorem atlas. It says repeated ratio, pair, blend, reflection, binary-choice, and geometry-loaded Mobius families can route through one bounded contrast plus a small transform opcode when exact rational checks pass. Analytic or numerical transforms stay HOLD.

Cross-Domain Kernel Adapter Registry

The cross-domain kernel adapter registry is:

4-Infrastructure/shim/cross_domain_kernel_adapter_registry.py

Current receipt:

shared-data/data/cross_domain_kernel_adapters/cross_domain_kernel_adapter_registry_receipt.json

Human summary:

shared-data/data/cross_domain_kernel_adapters/cross_domain_kernel_adapter_registry.md

Receipt hash:

a66552526d5213a8122ce8f1efa56137f70c707d991ac7fdc90dc83d970ac081

Cross-domain compression is admitted only as adapter-gated kernel reuse:

X_d = A_d[K_j(theta)] + R_d + chi0
same shape does not imply same law

Current status counts:

ACCEPT_ADAPTER_FIXTURE: 1
ACCEPT_KERNEL_ADAPTER: 4
HOLD_ANALYTIC_ADAPTER: 1
HOLD_BOUNDARY_WITNESS: 1
HOLD_CONTACT_TOPOLOGY: 1

Decision:

HOLD_CROSS_DOMAIN_WITH_ACCEPTED_KERNEL_ADAPTERS

Accepted adapter entries admit algebraic reuse only. Domain truth, physical interpretation, geometry optimality, corpus compression, and benchmark claims still require source equations, deterministic replay, residual policy, and closure receipts.

The moving-sofa / couch route is intentionally HOLD_CONTACT_TOPOLOGY:

continuous motion -> finite contact grammar -> curve atoms -> area/closure receipt

It is useful as a COUCH stress surface because contact-state switching can be represented as bounded clearance contrasts, but no moving-sofa optimality or area claim is promoted.

The Earth-core / seismic-horizon route is intentionally HOLD_BOUNDARY_WITNESS: boundary-crossing waves may route hidden-structure hypotheses, but inaccessible interior uncertainty remains Underverse until source data, residuals, and closure checks exist.

Magnetic Derivative Kernel Probe

The magnetic derivative kernel probe is:

4-Infrastructure/shim/magnetic_derivative_kernel_probe.py

Current receipt:

shared-data/data/magnetic_derivative_kernels/magnetic_derivative_kernel_receipt.json

Human summary:

shared-data/data/magnetic_derivative_kernels/magnetic_derivative_kernel.md

Receipt hash:

b4617a8ff31586250efafd13e3ed402535fcad5d564922599aaa3cb95134c7e3

Current status counts:

ACCEPT_DERIVATIVE_FIXTURE: 2
ACCEPT_KERNEL_ADAPTER: 1
ACCEPT_VECTOR_FIXTURE: 1
HOLD_ANALYTIC_ADAPTER: 1
HOLD_FIELD_EQUATION: 2
HOLD_MATERIAL_ADAPTER: 1

Decision:

HOLD_MAGNETIC_DOMAIN_WITH_ACCEPTED_FIXTURES

Accepted fixtures include local scalar/vector algebra only:

d/dB [B^2/(2*mu)] = B/mu
F_x = m*dB/dx
F_B = q*cross(v,B)
Gamma_mu = MN(mu2,mu1)

Maxwell, MHD, gauge, boundary-condition, hysteresis, material-response, and measurement claims remain HOLD until units, sign conventions, source data, residual policies, and closure receipts exist.

Solids Physics Kernel Probe

The solids physics kernel probe is:

4-Infrastructure/shim/solids_physics_kernel_probe.py

Current receipt:

shared-data/data/solids_physics_kernels/solids_physics_kernel_receipt.json

Human summary:

shared-data/data/solids_physics_kernels/solids_physics_kernel.md

Receipt hash:

98501df5a36ddd8a103ff40e6c8973f93e5271df325dd43ebdbebe59e896defb

Current status counts:

ACCEPT_DERIVATIVE_FIXTURE: 1
ACCEPT_ISOTROPIC_FIXTURE: 1
ACCEPT_KERNEL_ADAPTER: 2
ACCEPT_LINEAR_ELASTIC_FIXTURE: 1
HOLD_ANALYTIC_ADAPTER: 1
HOLD_FRACTURE_ADAPTER: 1
HOLD_PLASTICITY_ADAPTER: 1
HOLD_TENSOR_ADAPTER: 1

Decision:

HOLD_SOLIDS_DOMAIN_WITH_ACCEPTED_FIXTURES

Accepted fixtures include local linear-elastic algebra only:

sigma = E*epsilon
U = E*epsilon^2/2 = sigma^2/(2E)
dU/depsilon = sigma
G = E/(2*(1+nu))
K = E/(3*(1-2*nu))
E_eff = S/2*(1-MN(E1,E2)^2)
Gamma_Z = MN(Z2,Z1)

Wave-speed, plasticity, fracture, anisotropic tensor, boundary-value, geometry, and material-model claims remain HOLD until units, conventions, source data, boundary conditions, and residual policies are receipted.

Cross-Domain Easy Wins Route Map

The cross-domain easy-wins route map is:

4-Infrastructure/shim/cross_domain_easy_wins_route_map.py

Current receipt:

shared-data/data/cross_domain_easy_wins/cross_domain_easy_wins_route_map_receipt.json

Human summary:

shared-data/data/cross_domain_easy_wins/cross_domain_easy_wins_route_map.md

Receipt hash:

ac1fe2ca6ee469c046cdb5fc78cef9efca6a601aee2e5270ef4b8c5854bb1e2e

Decision:

ADMIT_ROUTE_MAP_HOLD_FIRST

Ranked next probes:

1. electrical circuits and transmission lines
2. thermal conduction and diffusion
3. acoustics and scalar waves
4. probability, routing, and expert selection
5. two-body mechanics and orbital reductions
6. chemistry equilibrium and reaction routing
7. optics at normal incidence
8. statistics and signal scoring
9. biology expression/accessibility contrast
10. contact geometry and motion planning

This is a planning receipt only. It does not assert compression gain, domain truth, or benchmark performance. The rule is:

prefer exact local algebra first
HOLD nonlinear, field, geometry, and measurement claims

6. Control Filters

The reconstruction core is guarded by the control-filter stack:

Filter	Function
LoC/NES Monster	rejects fake pattern / mirage / overfit
FYC Gate	rejects impossible constrained-manifold traversal
COUCH	rejects unstable hysteresis or chaotic route dynamics
Tree Fiddy	bounds recursion, retries, and refinement depth
BHOCS	commits only bounded, replayable survivors

Canonical admission sketch:

Admit(X) =
  replay_valid(X)
  ∧ byte_gain(X) > 0
  ∧ residual_declared(X)
  ∧ LoC_NES_pass(X)
  ∧ FYC_pass(X)
  ∧ COUCH_stable(X)
  ∧ TreeFiddy_bounded(X)
  ∧ BHOCS_verified(X)

7. Claim Boundary

This document does not claim that the current stack beats any compression benchmark. It defines the inspection boundary for compressed artifacts:

not readable != not auditable

The audit target is replay, residual repair, byte-exact reconstruction, and receipted byte accounting.