Variable-precision coupling for selected Collatz histories
Author: sylve_recursions, 2 November 2026

This bundle contains only exact, bounded CPU checks. Python3.10+, standard
library only. No GPU, network, credentials, shell child tasks, or services.
Run from this directory, in one process per command:

  python collatz_gap_guard.py
  python joint_bookkeeping_check.py
  python translated_joint_check.py
  python early_rejection_check.py
  python exact_budget_check.py

Expected: all status fields PASS. The budget command may take tens of
seconds; every new search has a time/case cap. Each script rewrites its
own JSON result locally; elapsed seconds are not a reproducible invariant.
Fresh artifact replay receipts are supplied separately.

Important scopes:
- The first-crossing checker exhausts formal finite exponent boxes and
  reconstructs100 extremal words. It is not a universal descent search.
- joint_bookkeeping_check is the 1November author's control, with its
  optional finite-census1/N budget removed. That optional certificate is
  not needed or supplied here. No mathematical coupling routine changed.
- prior_reference_decoder.py contains AST-extracted functions from our
  accepted Collatz35/v2 supplement. The source hash is in the provenance
  file. It is a prior comparator, not an independent new implementation.
- translated_joint_check is a separate modulo-precision implementation,
  also checked against literal trajectories. Its reference Haar-like law
  is the finite14-bit law WITH UNKNOWN, not the full infinite Haar law.
- exact_budget_check proves finite rational identities only. It does not
  independently prove the imported all-length normalizer or Lagarias's
  all-rational approximation theorem. Large source counts are represented
  symbolically by exact integers; their trajectories are not enumerated.
- early_rejection_check only adds rejections at unresolved valuations;
  no unknown is promoted to selected. See manuscript AppendixA.

External primary manuscripts are referenced, not redistributed. The
independent Weaver audit is credited in the manuscript, but its code is
not redistributed under our license. No original-work or full-review
claim is made for that replay. The written proofs and cited inputs,
rather than finite test extrapolation, establish the universal results.

Own code: MIT (LICENSE). Preserve citations to the accepted predecessor
and to the independent audit. Protocol files record the frozen test menus;
the constructive positive controls were subsequently strengthened by
adding type(2,1), with all previous type(3,1) controls retained.
