ITERSSS*

⚠️ Curated Limitations & Negative Results

• The detailed procedure assumes a finite, uniform, alternating MAX/MIN tree with a MAX root, fixed Dewey-code ordering, supplied terminal scores and a finite minimax value. A remark that node coding generalizes to nonuniform trees does not supply a complete variable-branch memory threshold or certify cycles, transpositions, infinite games or learned move ordering.
• The memory parameter counts OPEN entries in the stable abstract replacement model. It does not bound total bytes, priority-queue overhead, node-code storage or transient allocations. M >= M0 is a premise; no operation below that threshold or extrapolation to depth-zero and other exceptional tree shapes is certified. Informal above-threshold prose is not used to erase the inclusive formal endpoint.
• The purge and inactivesucc helper use the word successor. Reading it as descendant scope is supported by greatest-depth selection and the parent-subtree explanation, but remains a disclosed interpretation. The source does not specify a tie rule among equally deep inactive candidates; lexicographic first-selection ties do not silently fill that separate gap.
• The first helper may return null, while the case predicates and loop-termination expression refer to the selected node and its status. Safe null evaluation and short-circuit conventions require implementation choices. The null case changes FLAG; it is not an explicit general failure return or an independently proved remedy for every possible lack of active work.
• The source omits its formal proofs. It states that SSS* termination and correct root value also hold for ITERSSS*, but the SSS* one-representative-per-solution-tree lemma does not carry over to the ACTIVE subset. Neither the local SPACE arithmetic nor the purge-selection argument replaces a complete proof of ITERSSS* correctness or liveness.
• Theorem 4.4 gives only a sufficient ITERSSS* non-examination condition. The two-sided SSS* iff characterization, its strict right-hand comparison and an exact memory-dependent pruning law are not transferred. The alpha-beta terminal-set inclusion uses the stated comparator and common order; it does not apply automatically after different move ordering, tie/cutoff rules or implementation changes.
• The full-memory SSS* endpoint does not make ITERSSS* a global alias of SSS* or alpha-beta, and the source distinguishes it from an alpha-beta/SSS* hybrid. Matching terminal counts at a smaller budget does not establish identical control flow. A general per-instance monotone terminal-count theorem as M increases is not supplied by the displayed experimental trend.
• The reported evaluation measures average terminal evaluations as a percentage of b^d leaves, using four uniform tree configurations, ten random terminal-value sets per configuration and five M settings. It does not report runtime results. Queue bookkeeping and leaf-evaluation cost can affect time, and no algorithm or benchmark was executed for this draft.
• The finite synthetic study does not report exact random seeds, a fully specified terminal-value generator distribution, variance or confidence intervals. Its source-reported averages and memory trends do not establish real-game generalization or strict improvement on every input; the paper calls for broader practical evaluation.
• The inspected defining source is the official five-page 1986 proceedings PDF. Its later digitization metadata does not create a new algorithm version. The earlier MA*/MAO* bibliography uses a different spelling of the first author and a slightly different title; that exact citation remains unchanged alongside the defining PDF metadata. Four cited originals, three retrieved follow-up groups and three additional source questions remain retained without alias, ownership, access or closure decisions.