Partially Informed NegaScout

⚠️ Curated Limitations & Negative Results

• PNS is described in prose, while Figure 1 lists NS and the appendix lists INS. No complete PNS listing, exact near-root retention depth, principal-variation update rule, bound-storage interface, eviction policy or cache lifetime is supplied. These details must not be filled in by silently copying INS.
• The prove-best explanation depends on valid successor upper bounds and distinguishes those bounds from a true value obtained during re-search. It does not establish that all failed-window returns are exact values or that every complete INS guard and cutoff is retained by PNS. The descriptive PNS policy is not a certified executable implementation or an independently proved optimality theorem.
• The displayed base procedures use integer scores, adjacent-value windows and a tree of supplied width and depth. This draft does not extend PNS to arbitrary real-valued window tolerances, cyclic graphs, transposition handling, stochastic games, quiescence search or unspecified early-terminal conventions.
• The source does not supply a precise PNS space formula or a hard total-memory cap. The partly ambiguous summation on p.1041 concerns retention of all INS information, and the later hash-table suggestion also concerns INS. Neither is a specified PNS storage implementation or a bound covering its entire stack and data structures.
• The 1985 description combines near-root prove-best information with principal-variation retention; the later 1987 journal describes PNS using first-level upper bounds. Their exact implementation, retention-depth and path-management correspondence remains a version question, rather than an established equivalence or demonstrated contradiction.
• The terminal-count comparisons depend on the generated trees, width, depth parity, ordering and twenty-run samples. The paper says further variability measurements are in progress. No curve digitization, runtime reproduction or universal node-dominance claim is made here.
• The reported preferences for PNS and execution-overhead observations describe the authors’ implementations and cost model. They do not establish portable speedups, byte budgets or that fewer terminal visits always yield lower running time; timing claims about INS or SSS* must not be reassigned to PNS.
• The defining artifact is the inspected official-hosted four-page 1985 publication reconstructed by ABBYY in 2006. Extracted text omits listings and parts of the bibliography; all four page images were inspected, but some formula and reference characters remain ambiguous. The seven cited originals were not inspected for this draft and are retained as source questions rather than imported proof evidence.