feat(report): ReportPlan fold substrate, zero-copy pivot, reports as OGAR projection sources

[AdaWorldAPI]

What

crates/lance-graph-report (workspace member). A coordinate-native report plan that lowers into lance-graph-quack and runs on lance-graph-mask-risc's single evaluator. It adds no second evaluator, no independent aggregate semantics and no row loop.

• ReportPlan has three parts:

  • a lazy Selection — a predicate never becomes a bitmap just by being written;
  • coordinate providers, each with a role: an ordinal Field, or a derived Bucket that is never materialized as a lane;
  • measures expressed as mergeable fold states (FoldState::identity / merge / finalize). Adding a measure does not touch the pivot engine.

• Zero-copy pivot. PhysicalKey leaves out roles, axis order and top-k. A pivot rotation is therefore the same Arc<CellSpace> with a new View: never transpose data to pivot, transpose meaning.

• Physical planner (explain shows its choices). It picks:

  • the fold key;
  • which coordinates become partitions;
  • a dense or sparse accumulator — the full product is never allocated;
  • the selection carrier: validity plane, execution extent, tile-local ops, or one reused mask.

  Every planner knob has an inertness test.

• String-free execution core. It uses only FieldId, MaskId, SourceId and ordinals, and a source-fence test enforces that. Text appears in three places only:

  • boundary: a KV built on the contract's ContentStore/ContentSink, plus a CAM codebook of ContentIds that stores no text;
  • render: JSON/CSV data export;
  • explain.

crates/lance-graph-report-ogar (EXCLUDED; path-deps the OGAR sibling).

• ReportSource implements ogar_doc_ir::resolve::DocObjectSource over any existing object source and answers grid_of for report views.
• One DocCompose can arrange object A, a report, and object B by reference:
  • rotating a report means naming another view;
  • Revision(generation) pins a report;
  • Snapshot falls back to the slot's fallback.
• Rendering goes through OGAR's Typst emitter. There is no report-specific rendering ontology.

Depends on / companions

• feat(ogar-doc-ir): two-axis resolved projection (SlotOutcome::Grid) + typst emit_grid OGAR#307: SlotOutcome::Grid / grid_of / emit_grid, needed by lance-graph-report-ogar.
• feat(z8run-lance): native report nodes over the lance-graph ABI substrate z8run#1: z8run-lance native nodes build this plan. FlowMessages carry only handles.

Tests / measurements

All suites were run locally and pass; clippy -D warnings is clean on every crate.

Suite	Tests
lance-graph-report agnostic suite (TEST 1–9 + knob twins)	14
zero-copy allocation falsifiers	2
string fence	2
reference workload (KV/CAM, label counts, rename, late hydration, Quack parity + NULL semantics)	5
lance-graph-report-ogar convergence (rotation through composition, mixed recombination, Live/Revision/Snapshot, label rename)	4

• Execution allocations do not grow with population: execution allocates the same bytes at N and 64·N. The reused-mask twin grows by exactly the mask.
• Rotation is metadata: it allocates < 64 B (view) and < 256 B (plan reinterpret), and the payload address does not change.
• Label resolution is result-sized: a 100k-row 8×12 pivot resolves 20 CAM labels, and the fold itself does 0 CAM and 0 KV work.
• Falsifiers are disable-verified:
  • making rotation a no-op fails its test;
  • dropping partition filters fails 8 tests;
  • bypassing grid_of fails its tests;
  • ignoring the view orientation fails its test.
• report_bench (4M rows, avx2 build):

Case	Time	Notes
range→count	91 µs	0 mask ops
32×12 pivot	177 ms	25 scans
rotation via result view	49 ns	16 B
rotation via plan reinterpret	124 ns	140 B

Board

• New entry, and the entries index regenerated.
• STATUS_BOARD D-RPT-1..5, LATEST_STATE (no contract delta), and the AGENT_LOG recon run.
• ISSUES: ISS-REPORT-NO-COMPOSITE-KEY-GROUP-FOLD. A 2-D fold costs one pass per partition member. The missing primitive is a multi-lane group fold in ndarray::simd/mask-risc, not a report-side loop.
• SUPERSESSION-INDEX regenerated, with no diff.

🤖 Generated with Claude Code

https://claude.ai/code/session_01DCEP2fZdHYdMCtcEVcTpS2

---

Generated by Claude Code

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Essentials

Run ID: 598c8d31-1ff5-49ee-9629-fa732e002dac

📥 Commits

Reviewing files that changed from the base of the PR and between bcec6b3 and d4dd251.

📒 Files selected for processing (27)

• .claude/board/AGENT_LOG.md
• .claude/board/ISSUES.md
• .claude/board/LATEST_STATE.md
• .claude/board/STATUS_BOARD.md
• .claude/board/entries/2026-09-23-report-plan-zero-copy-pivot-docir-convergence.md
• .claude/board/entries/README.md
• Cargo.toml
• crates/lance-graph-report-ogar/Cargo.toml
• crates/lance-graph-report-ogar/src/lib.rs
• crates/lance-graph-report-ogar/tests/convergence.rs
• crates/lance-graph-report/Cargo.toml
• crates/lance-graph-report/examples/report_bench.rs
• crates/lance-graph-report/src/batch.rs
• crates/lance-graph-report/src/boundary.rs
• crates/lance-graph-report/src/exec.rs
• crates/lance-graph-report/src/explain.rs
• crates/lance-graph-report/src/ids.rs
• crates/lance-graph-report/src/lib.rs
• crates/lance-graph-report/src/plan.rs
• crates/lance-graph-report/src/render.rs
• crates/lance-graph-report/src/result.rs
• crates/lance-graph-report/src/selection.rs
• crates/lance-graph-report/tests/agnostic.rs
• crates/lance-graph-report/tests/common/mod.rs
• crates/lance-graph-report/tests/reference_workload.rs
• crates/lance-graph-report/tests/string_fence.rs
• crates/lance-graph-report/tests/zero_copy.rs

 __________________________________________________________
< Sometimes, I pretend to be a compiler to feel important. >
 ----------------------------------------------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

---

_{Comment @coderabbitai help to get the list of available commands.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_222481ca-91ec-42b1-83cc-dd2b9306ec00)

[AdaWorldAPI]

member-tests was cancelled after the tests finished, during the cache save. No test failed.

Every step in the job passed (log for job 107266243280):

• cargo build --workspace finished in 4m09s.
• Every per-crate test step was green through dto-class-check, which finished at 16:24:59.

After the tests, Swatinem/rust-cache restored for 6 min (16:03 → 16:09) and then spent 8 min saving in its post-job step (16:25 → 16:33). That save hit the job's timeout-minutes: 30 and was cancelled.

Why this happens here and not on main: the PR adds a workspace member (lance-graph-report), which changes the Cargo manifest hash. That hash is part of the lance-graph-members cache key, so this run has to save a fresh cache that main never has to write. The latest main run of this workflow (bcec6b3) is green.

I'm re-running the failed job once. If the save times out again, the fix is outside this PR's code: raise timeout-minutes or skip the cache save on PRs. I'll report back rather than widen this PR.

---

Generated by Claude Code

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_5cbfc3cb-de63-404e-92c3-c2b88ba60157)