avro: differential codec probe — ask the codec which slot a field lands in, instead of guessing from names #100

#98 resolves .field(_.x) to a schema field by NAME when the codec names the whole case-field list onto distinct schema fields, else by declaration POSITION. Three codec shapes are still resolved to the wrong field, and no amount of name matching can reach them, because the names carry no recoverable signal at all: a field list that both renames beyond recognition AND reorders ({beta_name, alpha_name} vs {alpha, beta} — equal arity, no name hit); two same-physical-type columns renamed and reversed (Ev(occurredAt: Long, seqNo: Long) against {seq_no, event_ts}, where a timestamp-millis logical type annotates the same physical long and so does not disambiguate); and compensating arity, where one case field is dropped and one computed column added so the counts agree while position shifts. All three are pinned as executable known limitations in avro/src/test/scala/dev/constructive/eo/avro/vulcan/ResolutionResidualSpec.scala (cells 5a, b2, b6). The mechanism that reaches them is a DIFFERENTIAL PROBE of the codec itself: at prism construction, encode two sentinel values through codec.encode, diff the resulting slots, and observe which schema field a given case field actually lands in. It needs no user declaration and no name heuristics, it is ground truth rather than inference, and it would additionally turn a wrong hand-written field mapping from believed into refused. Medium effort — the hard part is sentinel synthesis across the Avro type space (records, unions, enums, fixed, logical types, defaults) — and it MUST degrade to the existing name/position rule rather than refuse when a sentinel cannot be synthesised, so it never becomes a new false-positive surface.

Residual-cell evidence and the full 66-cell corruption matrix: /tmp/claude-1000/-home-rhansen-workspace-constructive-dev-eo/f9d50161-f717-4239-9f84-d7f5c596bbe3/scratchpad/hazard/attack-bypass/evidence/02-ladder-bypass.tsv and .../hazard/matrix/evidence/00-MATRIX.tsv. Decision memo: .../scratchpad/HAZARD-MEMO.md §3.

Filed as a follow-up to #95 / #98.

🤖 Generated with Claude Code

https://claude.ai/code/session_0194EHFR4NamCpTHiqy7B74V