## A quoted percentage accuracy can be rounded

What happened: for T616 the Yip–Leung Padé approximant (arXiv:1709.02469,
equations (73)/(100), Table 2/5) differed from two independent Lane–Emden
integrations at n=3 by 0.2141340818 percent. A test treating the paper's
"0.21 percent" as a strict maximum failed. Both integration formulations
agree to over 70 digits at that index. The observed percentage rounds to the
quoted figure; the discrepancy is the use of a rounded descriptive accuracy
as a strict inequality, not evidence of a failed ODE solve.

What the skill says now: check against independent values, and explain a
disagreement rather than choosing whichever result seems right.

What it should say: distinguish a certified strict bound from a source's
rounded accuracy summary for a fitted approximation. Retain the measured
error and any initial failed test. If rounding plausibly explains the
summary, state that interpretation explicitly; never use the approximate
fit as evidence for additional high-precision digits.

Evidence: T616's independent-initial-failure.log and independent-evidence.json;
source archived as yip-leung.html with SHA-256 in source-manifest.json.
The final check uses 0.215 percent, the upper rounding boundary of 0.21 percent,
and separately checks that the n=3 error lies in [0.205, 0.215) percent.