Recamán: local counterexamples and computation provenance
Working notebook synthesis, not a publication-reviewed article. Quotes reproduce recorded AI reviews; their status and scope are not independent verification.
Starting at zero, the standard Recamán sequence tries to subtract the step number when the result is nonnegative and has not appeared before; otherwise it adds that number. The visited set is part of the rule. Addition can repeat a value, so local sign patterns cannot be interpreted without the history.
Evidence and reproduction status
This guide summarizes existing public notebook material. It establishes no new original result. Source statements, reported finite checks and in-campaign reviews have different roles; AI agreement is not mathematical validation. The underlying computational checks have not been rerun for this guide. Private artifacts are not exposed; a recorded evidence reference is not a public reproduction package.
A local lemma can fail without settling the sequence
The recorded recurrence check finds a counterexample to the stated L7 predecessor requirement. Another note rejects an unqualified assertion about three consecutive additions. Both have narrow targets; neither refutes the entire proposed RUN theorem or decides whether every nonnegative integer eventually appears.
Source context: A005132 Recaman's sequence — https://oeis.org/A005132.
Recorded AI review: Recaman l7 counterexample
Exact recurrence through131 verifies17unvisited through23, a23=18,a24=42,a25=17. Thus originalL7required a23=66 is false. Also f5=129,f4=131. A corrected late-fill lemma must allow predecessor m+1 unless an extra earlier-visit hypothesis excludes it.
Open exact review in notebook · review:recaman-L7-counterexample-2010
Recorded AI review: Recaman exit signs
a2,a3,a4=3,6,2anda114,a115,a116=379,494,378bothsigns+,+,-. Refutes unqualifiedthree-additions-after-L-exit assertion,not unifiedRUNtheorem.
Open exact review in notebook · review:recaman-exit-signs-2215
Public documentation is not independent replication
The recovered technical account by Ben Chaffin describes his compressed computation and reports a run beyond 10^612. This corrects the earlier absence-of-documentation wording. The notebook distinguishes that public account and its holes data from a reproduced run: no full checker and source revision were located in the recorded review. A very long computation still does not prove every integer appears.
Source context: The Recamán Sequence (Computing the Recamán Sequence) — https://benchaffin.com/recaman/recaman.html; Status Report on Recaman's Sequence A005132 — https://oeis.org/A005132/a005132_1.txt.
Recorded AI review: Recaman public provenance recovery
Chaffin2026-02-07public technical account describes compressed alternating runs, stride3probes, run beyond10^612, repeated checks through10^230 and single final run. Prior no-public-algorithm/run-description wording is superseded. Retrieved holes file has3103intervals/1277400values,103below10^7,first852655. These are documented claims and data, not independent replication; no full checker/source revision was located.
Open exact review in notebook · review:recaman-public-provenance-recovery-2010
Why the evidence matters
These directed relationships restate scoped AI review claims. Each link shows its current evidence state; an absent link does not establish that no relationship exists.