WHEN DELAYED THRESHOLD SENSING CANNOT HELP
An Exact Scalar Acquisition-Timing Addendum
Version 1.1 -- 9 November 2026

This is a compiled working addendum, prepared for Brian Theory through
AI-assisted development and independent internal AI review. It is not a
journal publication, externally peer-reviewed article, priority certificate,
physical experiment or consciousness claim.

Read manuscript.pdf for the seven-page manuscript; manuscript.tex is its complete source.
It uses ordinary LaTeX math and includes its bibliography. The exact claim,
source and proof ledger is in proof_ledger.json. Source access depths and
the overlap comparison are in sources.txt. AI review outcomes and substantive
responses are in review_summary.txt and responses.txt.

The key result is negative: with the stated scalar dynamics, disturbance
radius, domain, deadlines and original input alphabet, delayed sensing never
improves the optimal worst-case probe count. Adding only input 1/2 permits
batch infeasibility versus exactly one delayed probe. This is not a theorem
of general sequential-acquisition superiority or a finite-to-finite saving.

REPRODUCING EXACT CHECKS
Python 3.10 or later; standard library only. No package installation, network,
paid service, API key, solver, GPU or private source files are required.
Run from this directory, choosing output filenames that do not already exist:

  python checks.py --output rerun_original.json
  python independent_checks.py --output rerun_independent.json
  python verify_package.py

Do not use python -O: assertions are part of the checks. Both mathematical
checkers refuse to overwrite results. Compare mathematical fields with the
saved results; runtime_seconds in the first checker varies. The verification
script checks the published payload hashes; it does not modify files.

Expected original check: PASS; 1092 input words through H=6; 7182 robust
different-word pairs, first differences 5930/982/270 at steps 1/2/3; one-probe
online and impossible batch separation. Independent check: PASS; 21845 ordered
tree instances; 13284 closed-prefix endpoint cases; 10 integer-center
representations; four exact interval images. These are finite corroborations,
not proofs of the all-horizon and all-policy quantifiers. See the manuscript.

PDF BUILD AND VISUAL VERIFICATION
The unchanged reviewed source compiled twice with MiKTeX 24.1 and automatic
package installation disabled. Both passes exited successfully. The second
pass has no unresolved references/citations or overfull boxes. One underfull
bibliography line (badness 1194) is readable and inside the margins.
All seven pages were rendered with Poppler and visually inspected for
clipping, equations, margins, headings, references and page numbers.
See layout_qa.json. This is AI visual inspection, not accessibility validation.

To rebuild with configured LaTeX and the standard packages in the source:
  pdflatex --disable-installer -interaction=nonstopmode -halt-on-error manuscript.tex
  pdflatex --disable-installer -interaction=nonstopmode -halt-on-error manuscript.tex

A prior sandboxed attempt failed during initialization. A subsequently approved
scoped build succeeded without a source change, package installation, manual
setup/configuration change or security change. The earlier failure is retained
in local provenance; it is no longer an open production blocker.

The public payload is separate from internal review transcripts. It contains
only the manuscript, portable code/results, source/proof records and concise
review disclosures. It does not contain machine paths or internal chat IDs.
All paths in publication_manifest.json are relative to this directory.

Release recommendation: integrate as a qualified working addendum to
From Observable States to Acquirable Predictive Models. The PDF is compiled and visually checked; central site integration and
site-format checks remain. Historical
priority, human specialist assessment and physical applicability remain open.
No public upload or site edit was performed as part of this package.
