## Preserve PARI precision before converting a Gen to decimal text What happened: T773's displayed Sage example called str() directly on a PARI Gen. PARI had computed the pairing at 681 and 1013 bits, but both strings were 0.0585226748448790. Agreement testing those short strings produced a long zero-padded approximation. The stored entries and attached generator already used the correct conversion and were unaffected. What the skill says now: computation.md requires honest precision and explains that agreement at different precisions is heuristic; it does not mention the separate display precision of a PARI Gen. What it should say: before passing a PARI real to a string-based agreement check, coerce the Gen to RealField(the requested bit precision), then convert that Sage value to text. A precision argument to PARI does not guarantee that str(Gen) prints those digits. Check the actual strings at both precisions. Evidence: repair_program_check.py executes both live Programs examples. repair-program-results.json records both raw strings, converted strings, the original error of about 4.82504e84 stored last-place units, and agreement of the correction with a separately evaluated Sage local-height polarization and the attributed saved independent MPFI enclosure. Actual versions are Sage 10.9 and numberdb 0.1.10. A regression should reject the original short string and check containment in the stored 100-significant-digit interval. Status: proposed, not a canonical skill change. Archive manifest SHA-256: 7629363ce07d2e1340528294fe639e6fc7a5b18da808ec52558687fc1db91534.