[EricAndrechek]

Summary

Spec batch 2 (#655) made each binding's BatchResult.spans hold one entry per row, with an empty entry for a row the export did not carry: Go []*Span, Python Span | None, TS Span | null, Rust Option<Span>.

A downstream user found that in Go the change compiles silently. Code written for []Span that reads r.Spans[i].Off still builds, because Go dereferences the pointer, and then panics at run time on a nil entry. Python and TS fail the same way at run time; Rust's Option makes the compiler catch it.

• Each binding's CHANGELOG breaking-change bullet now says how old code fails there, and to check each entry before reading it.
• The spans row in docs/reference/bindings-v1.md says the same (dprint realigned that table).

Also: how a column_defaults value is read (docs/guides/batches.md, plus one CHANGELOG bullet per binding). The value is read in the body format's own field encoding, by the reader that reads that column's field. Measured by a downstream user on build 20261021.025722 (darwin-arm64):

• CSV "a,""b" is read, and an unquoted a,b is refused with error 27;
• TSV a\tb is read, and a raw tab is refused with error 27;
• JSON "42" and 42 both read into UInt32;
• a value that doesn't read is refused at the call, before any body byte.

spec/abi-v2/docs.md already states the rule; it isn't edited here.

Docs only, with no code change. Nothing under spec/abi-v2/ or include/v2/ changes. The ABI v2 fingerprint is sha256:d53239933966…41c2 both before and after this change (measured with gen.py --major 2 --fingerprint).

Test plan

• markdownlint-cli2@0.18.1, dprint@0.57.4 check, lint-spelling.sh, parity-surface.py --selftest and lint-public.sh.
• CI.

🤖 Generated with Claude Code