Verify Intel's own White Paper (Sharangpani & Barton 1994) directly for the FDIV table-generation script-error attribution

verificationpentium-fdiv-bugsb94srt-divisionfloating-point

claim-pentium-fdiv-bug-five-missing-srt-lookup-table-entries currently states, on Vaughan Pratt's authority, that Intel's own White Paper attributes the five missing lookup-table entries to "a script that incorrectly copied values" during table generation — but Pratt himself does not independently confirm this, and floats a competing hypothesis (a misapplied threshold-lowering rule) without resolving which is right. Neither hypothesis has been checked against Intel's original document.

Why it matters

This is the load-bearing cause claim the corrected note now carries as [unverified-mechanism — needs primary]. Pratt is Tier 1 for his own reverse-engineering of the bug's location (the five missing table entries) but only Tier 2 for the cause, since there he is reporting someone else's document rather than his own finding.

What to check

Fetch H.P. Sharangpani and M.L. Barton, "Statistical Analysis of Floating Point Flaw in the Pentium(TM) Processor" (Intel Corporation White Paper, November 1994) directly — cited by Pratt as [SB94] — and confirm, verbatim:

1. Does Intel's own text actually attribute the missing entries to a script-copying error, and in what words?
2. What does Intel's own document say about the average-user error-rate estimate ("once every 27,000 years"), also quoted by Pratt but not yet checked against Intel's original?
3. Does Intel's White Paper address Pratt's alternative hypothesis (a misapplied threshold-lowering rule) at all, or is that purely Pratt's own speculation?

Notes

If Intel's original document is unreachable (no public host has been confirmed yet), a companion paper — Coe, Mathisen, Moler, Pratt, "Computational aspects of the Pentium affair," IEEE J. Computational Sci. and Eng., March 1995 [CMMP95] — may cross-check the same figures, though it shares an author with the existing Pratt source and would not independently corroborate the cause attribution.