The editor had written the escapes of the sources as their characters:
the cases "round escaped as round" and "round twice, once escaped as
round" of release.json had a plain round, and the surrogate pair of
"a surrogate pair in a value" a plain emoji, so the shared vectors tested
no escaped name. TestStrictJSON had lost its escaped é, its pair and the
escape after a lone high surrogate. They are escapes again, and the texts
of words_test.go too, so that no mark or invisible character hides in the
source. release.json gained 26 cases of drand's JSON, not 25 as b570338
says; the draft and the CHANGELOG now say 26.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
JSON of RFC 8259 in UTF-8 whose value is an object; no object repeats a
name, and names are compared exactly once their escapes are decoded, so
"round" is round and ROUND another name; an escape of a lone surrogate
is malformed; round is a number without sign, fraction or exponent from 1
to 2^53 - 1; signature and randomness are strings (spec v0.16, 47.1). Go's
encoding/json kept the last of two repeated names and matched ROUND to
round. ParseDrandJSON, exported, is the one reader of drand's JSON:
provider/drand reads the answers of the relays with it too. The error
texts do not change.
release.json gains 25 cases of drand's JSON for v0.16.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
