Companion to the PathQueryRun skipped-field issue, for the same goal: platform routing all ranked / having-range queries through the unified PathQuery surface and retiring the direct indexed_* call sites.
What
A golden test (or family of tests) asserting that for the same state and arguments:
prove_query(PathQuery::new_axis_top_k(path, axis, k, offset, descending)) emits byte-identical proof bytes to prove_indexed_axis_top_k_paginated(path, axis, k, offset, descending, ...), andprove_query(PathQuery::new_axis_bounded(path, axis, lo, hi, limit, descending)) emits byte-identical bytes to the direct prove_indexed_{count,sum,avg}_query(...) for the equivalent bounds,
across all three axes, empty and populated secondaries, offset 0 / mid / past-the-end, and both directions — plus the cross-check that verify_path_query and the direct verify_indexed_axis_* entry points accept each other's bytes.
Why
The unified dispatch was built over the same indexed-axis proof family, so equality is expected — but platform's switch from the direct provers/verifiers to prove_query/verify_path_query should be a pure refactor, proven by a pinned test rather than assumed. If the bytes genuinely differ somewhere, that is not a blocker (the surface is unreleased and platform owns both prover and verifier, so both sides can switch together) — but the difference should be a documented, deliberate fact rather than a surprise in a platform CI run.
Acceptance
- Byte-equality goldens in place for the top-k-paginated and bounded shapes, or — where equality does not hold — a documented enumeration of the divergence and mutual-acceptance tests between the two verifier entry points.
🤖 Generated with Claude Code
Companion to the
PathQueryRunskipped-field issue, for the same goal: platform routing all ranked / having-range queries through the unifiedPathQuerysurface and retiring the directindexed_*call sites.What
A golden test (or family of tests) asserting that for the same state and arguments:
prove_query(PathQuery::new_axis_top_k(path, axis, k, offset, descending))emits byte-identical proof bytes toprove_indexed_axis_top_k_paginated(path, axis, k, offset, descending, ...), andprove_query(PathQuery::new_axis_bounded(path, axis, lo, hi, limit, descending))emits byte-identical bytes to the directprove_indexed_{count,sum,avg}_query(...)for the equivalent bounds,across all three axes, empty and populated secondaries, offset 0 / mid / past-the-end, and both directions — plus the cross-check that
verify_path_queryand the directverify_indexed_axis_*entry points accept each other's bytes.Why
The unified dispatch was built over the same indexed-axis proof family, so equality is expected — but platform's switch from the direct provers/verifiers to
prove_query/verify_path_queryshould be a pure refactor, proven by a pinned test rather than assumed. If the bytes genuinely differ somewhere, that is not a blocker (the surface is unreleased and platform owns both prover and verifier, so both sides can switch together) — but the difference should be a documented, deliberate fact rather than a surprise in a platform CI run.Acceptance
🤖 Generated with Claude Code