Selection-aware Collatz precision -- 27 October2026

Start with STOPPED_BRIDGE_PROOF.txt. This is a proposed continuation for
outside scrutiny, not an accepted new paper. It uses the same actual-source
event and canonical charge as the prior selected-consumer proof, included
as PREVIOUS_SELECTED_PROOF.txt. The universal claims depend on written
arguments and their explicitly attributed imported coding/tilt facts.

Python standard library only, CPU, no network/GPU/installations:
  python verify_stopped_bridge.py
  python verify_stopped_bridge.py --long

The default replays exact parameter enclosures, first-crossing words, lifted
cylinders and pin valuations, an independent integer count of the stopped
partition, and finite probability budgets. The --long option additionally
reconstructs the stored prior and extended survivor normalizers and all four
larger stopped-count DP examples. Typical full runtime here is about a minute;
explicit subprocess and whole-replay time limits are enforced. The input
dimensions and iteration counts are bounded. Run without Python -O.

Replays overwrite only result JSON files. STORED_MANIFEST.json checks exact
code/proof bytes and JSON values after removing elapsed 'seconds' fields.
It supports repeated runs without pretending wall-clock timings reproduce.

Main finite controls:35,132 first-crossing words;4,082 literal source lifts;
498,048 stopped-decoder residue cases, with actual-event and direct rational
pin-valuation checks; independent2D prefix counts matching every partition;
unknown-survivor and wrong-prefix/missing-rotation-factor controls.

The finite budget tables keep every error and noninformative row. N=2^4096
and larger are algebraic population parameters, not enumerated trajectories.
The new stopped transfer proves a sufficient m<=log2(N)/5 window. Its convenient
asymptotic O(N^-1/1000) constant is not made explicit by the CLT argument.
Finite budgets instead use exact cyclic endpoint lower bounds and/or integer
survivor counts; they need no implicit CLT constant.

Old ordinary unresolved data can also bound the stopped unknown set, so finite
calculations take the best PROVEN upper among its exact NB tail, the stopped
MGF bound and an exact stopped-count DP when available. The unknown set is
never treated as rejected. No source or exceptional requested output is
silently dropped. No integer uncapped mean, natural-horizon descent, cycle
exclusion or solution of Collatz follows.

The previous normalizer DP and Cramer enclosure scripts are from our earlier
public supplement; new modifications are limited to dimensions/output names
for the normalizer extension. The general methods themselves are classical.
See the proof's source comparisons and explicit attribution.
