Research notebook

How this research notebook works

The goal is inspectable work: a precise question, bounded evidence, and honest uncertainty. Research activity is not the same as a mathematical result.

What is available now

One small reproduction packet has public programs, exact inputs and outputs, SHA-256 hashes, runtime records and scoped review. Topic guides summarize existing working notes. The notebook retains sources, AI reviews, graph links and article edit history.

These pages are published snapshots, not a live research feed. The deployment record identifies the released source and build. The original campaign artifacts remain private unless specific files are reviewed for public release.

Separate the kinds of claims

What was actually checked

The pilot uses exact integer arithmetic, two separately constructed programs, full trajectory comparison, first-hit minimality, transition and affine-identity checks, boundary cases and intentional corruptions. AI-assisted reviewers checked the definitions, source statement and implementation scope. This does not constitute external, human or formal verification. The live article publication gate remains unchanged.

How to spend the next research tokens

  1. Choose one concrete uncertainty: a missing hypothesis, ambiguous citation, proposed local lemma or small certificate.
  2. State the question, domain, expected deliverable and resource cap before starting.
  3. Prefer a bounded source check, counterexample search or independently constructed verifier.
  4. Stop at the cap, on missing inputs, or on a failed check. Record UNKNOWN, failure or a retraction instead of silently broadening the task.
  5. Save the evidence and seek review that can falsify the claim. Publish only the inspected files and distinguish known results from proposed novelty.

This is a process for future explicitly scoped work. It does not schedule recurring research, launch a campaign, allocate a new budget or promise external review.

What is still missing

Most archived notes have no public executable reproduction package. Historical novelty has not been exhaustively assessed. Substantial proposed proofs still need appropriate independent mathematical scrutiny; formalization or external human review may help where practical. No universal reliability score is inferred from the number of AI reviews.

Publication policy

Research is published only on our own websites. No third-party repository deposit, journal submission or preprint-server submission is authorized. Links to primary literature are citations, not submissions. Hosting a document here does not make it peer reviewed.

Three independent labels

Output type
Literature note, reproduction, exploratory analysis, technical report or proposed proof. These describe the work, not its reliability.
Verification
State the actual checks: source inspection, executable replay, separately constructed verifier, formal proof or external expert review. Record scope, failures and missing checks. AI agreement alone is insufficient.
Publication
Private draft, public working record, revised, superseded or withdrawn. Public availability is separate from scientific validation. Older archive labels are historical and do not imply completion of today's review gate.

Before releasing a result

  1. State the precise question, prior work, assumptions and contribution. Distinguish exploratory work from a prospectively recorded protocol; never describe a retrospective plan as preregistration.
  2. Preserve methods, inputs, software versions, seeds where relevant, resource limits and stopping criteria. Report negative and inconclusive results, uncertainty and deviations from the plan.
  3. Trace every claim to evidence and primary references, with exact source versions and relevant locations. Record inaccessible sources and gaps in prior-art searches.
  4. Reproduce the bounded computation and seek a genuinely separate check. A completed run, a valid manifest and a plausible explanation do not establish a theorem or novelty.
  5. Review the public file allowlist for private material and redistribution rights. Keep third-party papers as source links unless their license permits the intended redistribution. A public download does not imply an open license.
  6. Record human responsibility and actual AI assistance without inventing authors, affiliations, reviewers, funding or model details. Model-generated text and model review are disclosed as such.

Citations, versions and corrections

Released packets expose a content-addressed evidence URL, machine-readable record, file hashes and a suggested citation. Cite the exact evidence identifier and access date, not just the homepage. SHA-256 identifies bytes; it does not certify their truth. We do not assign a DOI or imply archival preservation by an outside institution.

Substantive changes require a new evidence identifier, repeated applicable checks and a dated correction explaining affected claims. Preserve earlier released versions and their correction or withdrawal notices. A withdrawal should leave a public explanation rather than quietly erase the claim; private information accidentally disclosed is removed. Existing article edit history and deployment records remain available. Long-term retention still depends on this site's backups and operation.

The initial version record documents the existing bounded reproduction only. It does not promote private research, certify the legacy archive or claim external review. New packets remain subject to their publication gate.

Standards informing this workflow

These are workflow references. No certification, endorsement or full compliance assessment is claimed.