adityamukho · Nov 4, 2026 · Nov 4, 2026
@@ -22,6 +22,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Documentation
 
+- **Benchmark exclusions audit** (#836): [pinned enumeration and per-shape dispositions](docs/benchmarks/836-exclusions/README.md) reconcile 156 unselected native IDs, retain 110 manual row-frame cases, and distinguish linear full-prefix UDF accumulation from bounded replay. No new latency measurements or runtime changes.
+
 - Retire tracker #383 in favor of the v3.0.0 milestone and its release gates. Ordinary/window value-handling consistency is completed by #789.
 
 ### Performance

@@ -6,7 +6,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 Minigraf is a tiny, portable **bi-temporal graph database with Datalog queries** written in Rust. Designed as the embedded graph memory layer for AI agents, mobile apps, and the browser — built on the SQLite philosophy: embedded, single-file, reliable, with time travel.
 
-See `ROADMAP.md` for planned work and current direction. See `CHANGELOG.md` for completed-release history.
+See `ROADMAP.md` for planned work and current direction. See `CHANGELOG.md` for completed-release history. The [#836 benchmark audit](docs/benchmarks/836-exclusions/README.md) records the 156 unselected native IDs and per-shape manual dispositions; it changes no runtime behavior or test counts.
 
 ## Core Philosophy - CRITICAL
 

@@ -32,7 +32,7 @@ For example, `(lag ?salary :over (:partition-by ?dept :order-by ?hire-date))` re
 
 The unreleased v3.0.0 branch supports ordinary `(avg ?score)` (#777). It skips nil and nonnumeric values, counts only numbers, and returns a Float average or nil for an existing group without numbers. No matching rows produces no result row. It uses the same exact-total-then-divide arithmetic as window `avg`; a finite total can overflow before division. V2 requires `:over` for `avg`.
 
-The unreleased v3.0.0 branch adds bounded sliding windows such as `(avg ?salary :over (:partition-by ?dept :order-by ?hire-date :rows 2 preceding))` (#183).
+The unreleased v3.0.0 branch adds bounded sliding windows such as `(avg ?salary :over (:partition-by ?dept :order-by ?hire-date :rows 2 preceding))` (#183). See the [benchmark coverage and manual-case dispositions](docs/benchmarks/836-exclusions/README.md) for the limits of nightly coverage (#836).
 
 On v3.0.0, an ascending value-only query such as `[:find ?v :where [?e :score ?v] :order-by [?v] :limit 10]` can read directly in index order and stop early when all committed `:score` history has one value kind (integer, float, or Boolean) and its source range has no pending writes. A query whose full order cannot be proved uses the existing sort (#641). Built-in `count`, `count-distinct`, `sum`, and `avg` can release completed groups when the pinned source proves their grouping order (#781). `min`/`max` also qualify when their arguments are proved Integer/Float/Boolean/nil; `sum-distinct` requires Integer/nil. A single Integer/Boolean/Float key (including mixed numeric groups) is supported. A longer effective tuple, including hidden `:with` keys, must reduce to one varying Integer/Boolean slot plus repeated occurrences and proved Integer/Boolean/nil constants. Any requested final Sort is retained for `:with`. Small grouped point queries pay proof startup cost; see the [measurements](docs/benchmarks/781-ordered-aggregate.md). Memory follows the active group and batches; distinct members within one group still grow. A limit can stop after enough emitted groups, but dropped min/max groups do not count and a retained Sort drains the input. Callbacks, `collect-set`, aggregate+window queries and unproved arguments keep full-input evaluation. When every built-in window uses the same proved nonconstant partition key, the engine can also finish and release one partition at a time. Its memory follows the largest partition plus bounded batches; one growing partition still grows. Unpartitioned windows, incompatible keys, aggregate queries, callbacks and `collect-set` keep full-input evaluation. A surrounding sort or distinct operation can still require more memory.
 

@@ -102,7 +102,7 @@ Scope — query API:
 
 - Optional `:distinct` result tuples, preserving bag output by default (#493; done on `v3`)
 - `lag` and `lead` window functions for the previous/next row within an ordered partition (#182; v3.0.0)
-- Bounded sliding row frames, `:rows N preceding`, for window aggregates (#183; v3.0.0, implemented on `v3`; builds on #182)
+- Bounded sliding row frames, `:rows N preceding`, for window aggregates (#183; v3.0.0, implemented on `v3`; builds on #182). [#836 audit evidence](docs/benchmarks/836-exclusions/README.md) retains the 110 wider cases as manual/deferred; it does not expand nightly coverage.
 - Cursor API: `Minigraf::query` and `PreparedQuery::query` return an owned `Cursor` that delivers rows in batches, with the answer fixed at open (#432; Rust API done on `v3`). The binding cursor objects follow. Behind the cursor, the streaming engine (#473) runs every query and produces rows as they are pulled: patterns, expressions and pseudo-attributes (#544), joins and `:db/valid-at` (#545), `not` and `or` (#546), aggregates and windows (#547), rules (#548), and transaction reads (#550); the executor's query path is gone (#551; done on `v3`). Operators do not spill: memory follows operator state, with an opt-in per-query limit (`QRY-012`, #549).
 - Projected UUID lookup batches walk dense DICT ranges (#530; implemented on `v3`), keeping temporary memory bounded to a batch and query-cache charges in projection order.
 - Native `:limit N`, `:order-by` and `:offset`, with a top-k `Sort` whose memory follows `N` (#306; done on `v3`, with the 10M-fact acceptance tests moved from #473; spec `docs/superpowers/specs/2026-10-24-limit-order-by-design.md`). #641 targets v3.0.0 in three slices: snapshot-bound order proofs and Sort omission for committed scalar ranges, then one-group count/count-distinct and partition-at-a-time windows. All three slices are implemented in the dependent #641 PRs for `v3`. #781 extends ordered grouping to sum/avg, fixed-scalar min/max, Integer/nil sum-distinct, and effective tuples reducible to one ordered Integer/Boolean slot plus scalar constants/repetitions, including hidden `:with` keys. Every `:with` Sort is retained. Unproved keys/arguments, callbacks, `collect-set`, rules and aggregate+window queries retain full-input evaluation. Compatible built-in windows with one proved nonconstant partition key hold one partition and bounded batches; unpartitioned/incompatible windows, aggregate queries, callbacks and `collect-set` retain full-input evaluation. See `docs/superpowers/specs/2026-10-28-641-input-order-design.md` and `docs/superpowers/specs/2026-10-29-781-ordered-aggregate-design.md`.

@@ -732,6 +732,12 @@ UDF case remains excluded for quadratic callback cost. An untimed local check
 of the shortlist uses `cargo bench --bench minigraf_bench -- --test '<filter>'`
 with the exact filter above.
 
+The [#836 disposition report](benchmarks/836-exclusions/README.md) retains the
+110 manual cases and reconciles all 156 unselected registrations at a fresh pin.
+It records per-shape costs, not new timings. In particular, full-prefix UDF
+frames use a running accumulator; only bounded replay has the quadratic
+half-frame cost. The one-partition prefix remains manual.
+
 ## Performance Envelope (v3.0.0, #394)
 
 **Version**: `v3` branch (format v8), unreleased · **Date**: 2026-10-11 · **Command**: `cargo run --profile envelope --example perf_envelope -- --scales 100k,1m,10m` · Intel Core i7-1065G7 (4 cores / 8 threads, 16 GB RAM), Rust 1.94.0, Linux 6.12.108, on AC power and otherwise idle; the `.graph` file on an SSD with the OS page cache warm.

@@ -130,7 +130,7 @@ execution was not repeated; its 2,066-test inventory was verified.
 
 - **Partition windows** (#641, third slice; `window_order_tests.rs`, `order/tests.rs`, `order_tests.rs`, `engine/tests.rs`): compatible multiple windows with independent orders, peers, frames and neighbors; scalar/mixed-number, entity, Ref and Keyword adjacency with comparator-specific fallbacks; boundaries on and across batch edges; empty/selective ranges; callback/CollectSet/aggregate-query exclusions; bounded early-limit reads and output coalescing; pending/staged/working-reader/prepared/cursor pinning; fixed versus growing partitions; separately tracked Window/IdSpace peaks; internal input and reference-result cache eviction, including a failing retention mutation; lookahead ownership, low budgets, allocator refusal, close/drop/rewind and defensive `INT-032` before and after returned batches. The committed `MemoryBackend` fixture runs on native, WASI and browser targets.
 
-- **Bounded row frames** (`tests/window_functions_test.rs`, `src/query/datalog/aggregate/tests.rs`, #183): `:rows N preceding` parsing and errors, physical rows with tied order keys, partition reset, descending order, width zero and maximum, all five built-in aggregates against a replay oracle with numeric edge cases, UDFs through prepared queries and cursors, grouping with `:with`, distinct output, memory-limit refusal for long min/max strings, and final `:order-by` matching. A `test-hooks` case checks answer bags with clause reordering on and off.
+- **Bounded row frames** (`tests/window_functions_test.rs`, `src/query/datalog/aggregate/tests.rs`, #183): `:rows N preceding` parsing and errors, physical rows with tied order keys, partition reset, descending order, width zero and maximum, all five built-in aggregates against a replay oracle with numeric edge cases, UDFs through prepared queries and cursors, grouping with `:with`, distinct output, memory-limit refusal for long min/max strings, and final `:order-by` matching. A `test-hooks` case checks answer bags with clause reordering on and off. The [#836 benchmark audit](benchmarks/836-exclusions/README.md) reconciles registration/selection and documents deferred shapes; enumeration is not a query-correctness run and adds no Rust tests or timing claims.
 
 - **Optional distinct output** (`tests/distinct_test.rs`, #493): bag defaults, whole tuples, section syntax, cross-batch and committed/pending values, chunked write transactions, prepared execute/cursors, joins/rules/disjunction, aggregates and hidden `:with` groups, window peers and partitioned row numbers, temporal filters, typed equality and canonical float/set output, memory limits (including retained set payloads) and cursor cleanup. A meter unit test verifies each cell/key/value estimate includes set elements and their strings. With `test-hooks`, checks both reorder settings and hashed/correlated subplans. An engine unit test checks counters and state release.
 

@@ -362,7 +362,7 @@ shortlist. No manual timing or release sign-off is implied by this close-out.
 
 | ID | Disposition | Remaining scope and reason |
 | --- | --- | --- |
-| X01 | MANUAL / EXCLUDED registered cases | Of 583 registered native IDs, 156 are unselected: 110 `window_rows/` cases are MANUAL wider width/function/partition sweeps; four history depth-500 IDs and six retract 1k/100k/1m IDs are EXCLUDED redundant intermediate/scaling points; six pending `btree_lookup/` IDs are EXCLUDED duplicates of C01; 30 synthetic SIMD IDs are EXCLUDED experimental non-engine kernels. Exact expansions are in Appendix C. Full-prefix one-partition UDF replay is expensive and stays manual; the unregistered 100k UDF half-frame is also excluded for quadratic callback cost |
+| X01 | MANUAL / EXCLUDED registered cases | Of 583 registered native IDs, 156 are unselected: 110 `window_rows/` cases are MANUAL wider width/function/partition sweeps; four history depth-500 IDs and six retract 1k/100k/1m IDs are EXCLUDED redundant intermediate/scaling points; six pending `btree_lookup/` IDs are EXCLUDED duplicates of C01; 30 synthetic SIMD IDs are EXCLUDED experimental non-engine kernels. Exact expansions are in Appendix C. The [#836 pinned reconciliation and per-shape dispositions](benchmarks/836-exclusions/README.md) retain all exclusions. Full-prefix one-partition UDF execution stays manual but uses a linear running accumulator, not quadratic replay; the unregistered 100k UDF half-frames remain excluded for replay cost |
 | X02 | ENVELOPE / acceptance, execution unverified | Retained-state/RSS/scaling bounds, nonlinear/multistratum recursion and derived-fact exhaustion, million-fact transactions and custom chunk bounds are not Bencher memory metrics. Observe the revised v3 smoke schedule or dispatch it and inspect exact SHA/test names/bounds before claiming these enforced in nightly CI. Accounting itself has no disabled-engine control; unlimited only disables the finite cap |
 | X03 | MANUAL migration | Native migration uses one small frozen v7 fixture, not a large dictionary/scale sweep. Browser migration and migration with pinned-reader churn have no V timing; IndexedDB open/write coverage is not migration coverage |
 | X04 | MANUAL binding checks | JVM/JNI, Android ART/ABI/device storage, Swift ARC/Apple storage and WASI need distinct runtime evidence/release sign-off. Options/error paths, foreign FactLog/LogWriter conversion, foreign preparation/registration latency and browser pinned-reader churn remain unmeasured. Python shared UniFFI coverage does not measure other host runtimes; no binding RSS bound is asserted |
@@ -1008,6 +1008,9 @@ binding/python/aggregate/1k
 
 The enumeration/selected-set difference is exactly **156** IDs. These have no
 V analysis or upload. Counts below are not omitted-feature counts.
+The [#836 follow-up](benchmarks/836-exclusions/README.md) re-enumerates a later
+pinned revision and records an explicit cost/disposition for every manual shape;
+it adds no latency measurements and leaves this historical validation unchanged.
 
 | Prefix / expansion | Count | Disposition and reason |
 | --- | ---: | --- |