The 50 decimal places are transcribed from equations (1.18) and (5.3) of Varga and Carpenter [1]; a scan is available as [5]. They computed the minimax errors $E_{2n}$ for $1\leq n\leq52$ by the Remez algorithm at 100 decimal working digits and applied Richardson extrapolation in $1/n^2$ to $2nE_{2n}$. Their finite-degree errors were estimated accurate to at least 95 decimal places. Of 210 entries in the last 20 extrapolation columns, 182 agreed with all 50 reported digits. These are heuristic digits of the limit, not a certified 50-digit enclosure. No additional digits are asserted.
The transcription matches all 50 digits of [4], which records the same source computation. A separately implemented calculation at 160 decimal working digits recomputes all 52 minimax errors and the Richardson tableau. Its control is the exact degree-two minimax polynomial $x^2+1/8$, whose uniform error is $1/8$. For each degree it checks alternating residuals, locates all positive critical points, and requires the spread of their error magnitudes to be less than $10^{-90}$. Five finite-degree errors are also compared with the 20-place values in Table 2.1 of the paper.
The independent extrapolation differs from the transcribed decimal by about $1.3932\times10^{-51}$, within its one-unit last-place interval. The attached checks compare this independent result and the OEIS digits with the value read back from the database. Agreement of extrapolations does not supply a remainder bound, so the rigour remains heuristic.
