Bolt: tracking issue for protocol/type-fidelity gaps surfaced by certification #4890

Part of #4882.

Motivation

Four concrete depth gaps were confirmed by code audit up front: (1) temporal types serialize as ISO strings instead of native Bolt structures, (2) Duration and spatial Point have no handling at all in BoltStructureMapper/PackStreamWriter, (3) no Neo.TransientError.* error codes are defined so driver-side transient retry can't trigger correctly, (4) ROUTE returns only this node's own address - no HA-aware multi-server routing table - and Bolt 5.x is never advertised/negotiated. Group B's suites would surface exactly how each of these fails and might reveal more; this issue exists to coordinate fixes rather than guess scope upfront.

Proposed design

• This issue stays a tracking issue - as Group B suites go red on these (and possibly other) scenarios, spawn concrete, narrowly-scoped fix sub-issues from here.
• Each spawned sub-issue follows the same template (Motivation/Proposed design/Acceptance criteria/Dependencies/Effort) and links back here and to the Epic.

Status update (2026-07-05): gate lifted, children spawned

All five Group B suites are now merged (#4885 python, #4886 csharp, #4887 go, #4888 js, #4889 java), and bolt/conformance/spec.yaml pins the red cells authoritatively (current_status: expected-fail, each tracking_issue: "#4890"). The evidence exists, so children are now scoped.

Reconciliation of the original 4 gaps against actual suite evidence:

• Gap (1) temporal ISO strings -> RESOLVED by Bolt: serialize outbound temporal values as native PackStream structs instead of ISO strings #4907 (serialize outbound Bolt temporals as native PackStream structs). Temporal TYPE scenarios pass. No child needed.
• Gap (3) Neo.TransientError.* -> NOT a gap. TX-005/ERR-004 are certified positive in the Java suite: Neo.TransientError surfaces on a write-write race via overlapping explicit transactions. Documented as certified working; no child needed.
• Gap (2) Duration/Point -> still red (TYPE-011, TYPE-012).
• Gap (4) ROUTE + 5.x -> still red (CONN-004, PROTO-002).

Plus three additional gaps the suites revealed (as anticipated): RESULT-004 (summary counters), TYPE-003 (Path never constructed), ERR-002 (semantic vs syntax error code).

Spawned sub-issues (one per remaining red cell / logical group)

• Bolt: implement native Duration and Point structure mapping #4997 - Bolt: native Duration + Point structure mapping (TYPE-011, TYPE-012) - S-M
• Bolt: emit native Path structures for query results #4998 - Bolt: emit native Path structures (TYPE-003) - S
• Bolt: distinguish semantic errors from syntax errors in the RUN handler #4999 - Bolt: distinguish semantic from syntax errors (ERR-002) - S
• Bolt: populate ResultSummary counters for write queries #5000 - Bolt: populate ResultSummary counters for writes (RESULT-004) - M-L
• Bolt: decide and implement 5.x version negotiation #5001 - Bolt: decide + implement 5.x negotiation (PROTO-002) - M
• Bolt: HA-aware ROUTE response with multi-server routing table #5002 - Bolt: HA-aware ROUTE response (CONN-004) - M

Acceptance criteria

• All confirmed gaps have a corresponding spawned sub-issue once Group B's tests are red on them, or are explicitly re-classified (gaps 1 and 3 above are resolved/certified, documented here).
• Every spawned child closes its scenario in spec.yaml (flip expected-fail -> passing) or the cell is documented as an accepted limitation in the compatibility matrix.
• No gap remains silently unhandled - the A1 conformance spec is authoritative on what "handled" means.

Dependencies

Depended on Group B (the five per-language suites) surfacing red cells - now satisfied.

Effort estimate

S for this tracking issue itself (children each carry their own estimate above).