T04: adversarial review round + survey corrections + ADR-0002 (reimplement, assimilate patterns)

decisions/ADR-0002-game-kernel.md

@@ -0,0 +1,70 @@
+# ADR-0002: Game-state kernel — survey approval and implementation decision
+
+Date: 2026-07-31
+Status: accepted
+Capability: game.kernel.authoritative-state
+Inputs: research/CB-RES-0001-game-kernel.md (approved),
+history/260731-game-kernel-{research,challenge,response}.md
+
+## Approval (InnerLoop step 2)
+
+The CB-RES-0001 survey is **approved**. It passed one round of adversarial
+review: the challenge found one blocking overclaim (degradation
+attribution) and three significant gaps (memory methodology, omitted
+candidates OpenSpiel/Ludii, unsourced D3 ceiling); all were resolved with
+new measurements or corrections, none overturned the comparative
+conclusions. The candidate set is judged adequate; the baselines are
+trustworthy within their stated provenance (measured vs cited vs
+estimated), with the event-sourcing ceiling explicitly parity-capped.
+
+**Benchmark-to-beat, named:** boardgame.io 0.50.2 per the survey's
+baseline table (throughput ~1,100–1,900 moves/s degrading with history in
+every configuration; 120 deps/37 MB; ~36-LOC declarative game definition),
+plus the **stipulated engineering target** of ≥100,000 applied events/s
+with flat scaling (@100k within 10% of @5k) and bit-identical replay.
+
+## Decision (InnerLoop step 3): reimplement, assimilating patterns
+
+Build the Clay-Borg semantic kernel as **our own Rust implementation**
+(`cb-kernel` + `cb-events` + `cb-game-runtime`) — not a wrapper around any
+surveyed system. Per the Clay-Borg rule, we own the semantics; what we
+assimilate from the field is *patterns*, each with a named source:
+
+| Assimilated pattern | From | Into |
+|---|---|---|
+| Declarative game definition (setup/moves/phases as data + pure functions) | boardgame.io | game-package rules API |
+| Tooling-enforced determinism (not convention) | Rune | seeded RNG service, denied ambient time/random, replay double-run check in the scenario runner |
+| Command → validate → events → fold, snapshots bound replay | event-sourcing (cqrs-es pattern) | kernel mutation pipeline |
+| Information-state framing for imperfect information | OpenSpiel | per-player projection model |
+| Hand-zone / object-visibility semantics | TTS (pattern only) | tabletop layer, later |
+
+Rejected alternatives: **assimilate boardgame.io** (inactive, JS-bound,
+history-degrading in best config — fails D2/D3/D4); **build on OpenSpiel**
+(right semantics, wrong layer — research kernel without projection/
+networking/snapshot surface, heavyweight C++/Python boundary); **ECS as
+kernel** (bevy_ecs stays in the world layer per the blueprint).
+
+## Expected advantage per dimension (honest trades)
+
+- **D3 speed: better.** Flat per-event cost with snapshots; the 100k/s
+  floor. Declared caveat: part of the margin over boardgame.io is language
+  choice; the commitment that matters is flat scaling, and the
+  event-sourcing comparison stays at parity until a Rust comparator is
+  measured.
+- **D2 weight: better.** Target ≤20 transitive crates for the headless
+  kernel (vs 120 npm packages), measured by M-D2-DEP/M-D2-LOC/M-D2-BLD.
+- **D1 specification: parity, aiming better.** Match boardgame.io's
+  declarative compactness on the synthetic workload (≤~40 LOC equivalent);
+  surpass via rule-to-scenario traceability (M-D1-COV), which no candidate
+  has. Declared trade: we do not attempt Ludii/GDL-class generality — one
+  game family first, canon only after a second concrete use.
+- **D4 optionality: better.** Port pattern with null/reference impls from
+  day one, capability boundary WIT-expressible later, zero foreign types in
+  canonical interfaces (M-D4-LEAK=0). No surveyed candidate offers this.
+
+## Gate
+
+Per InnerLoop, no kernel implementation code existed before this ADR.
+Next steps: T05 (GROUND rules spec, derived from ground-game), T06 (kernel
+spec with acceptance metrics tied to these baselines), T07 (workspace +
+harness), T08 (code loop).

history/260731-game-kernel-challenge.md

@@ -0,0 +1,60 @@
+# 2026-07-31 — adversarial challenge: CB-RES-0001 (T04)
+
+**Verdict: approvable with corrections** — the measured numbers are real and the determinism/dependency claims verify against the harness, but the survey's central architectural claim about boardgame.io's superlinear degradation is contradicted by a documented tuning option the harness never tested, the memory methodology is invalid, and at least two squarely in-scope candidates (OpenSpiel, Ludii) are omitted.
+
+Reviewed cold: only `research/CB-RES-0001-game-kernel.md` plus the harness under `research/CB-RES-0001-harness/boardgame-io/` (`bench.js`, `results-260731.json`, `package.json`, and the installed `boardgame.io@0.50.2` source in `node_modules`).
+
+---
+
+## Blocking
+
+### B1. "Architectural, not a tuning artifact" is asserted, untested, and likely false as stated
+
+Survey text: *"Per-move cost grows with history length (log accumulation + state pipeline): throughput halves as move count doubles — superlinear total cost. This is architectural (unbounded redux log per client), not a tuning artifact."* This attribution underwrites the D3-scaling baseline row (0.45×), the "boardgame.io is slow *and* degrades" verdict, and plausibly the "did not finish in 300s" @100k row.
+
+I inspected the installed library. boardgame.io v0.50.2 has a documented game option `disableUndo` (`node_modules/boardgame.io/dist/cjs/initialize-648ccd94.js:49`, `server.js:1912`). Without it, every move executes `_undo: [...state._undo, undoEntry]` (`reducer-6f7cf6b0.js:853`) — an O(n) array spread **per move**, i.e. O(n²) total, retaining a G/ctx/plugins snapshot per move. That alone is sufficient to produce exactly the observed halving-per-doubling and the RSS growth, and it is switched off by a one-line, documented config flag. The harness (`bench.js`) never sets `disableUndo`, never tests the bare `Master`, and the survey nowhere acknowledges the flag. The client-side `log` accumulation the survey blames may or may not remain superlinear with undo disabled — nobody measured it.
+
+The survey's Risks section hedges the *measurement surface* ("headless client pipeline... a bare server-side master could differ") but not the *attribution*: "architectural, not a tuning artifact" is precisely a claim that no tuning knob exists, and one demonstrably does. **Required correction:** rerun the harness with `disableUndo: true` (and ideally the server `Master` path). If scaling flattens, the D3-scaling row, the @100k DNF row, and the "degrades" verdict language must be rewritten; if it stays superlinear, the claim becomes defensible and should cite the flag explicitly.
+
+## Significant
+
+### S1. Memory baseline (RSS 224 MB @5k) is methodologically invalid
+
+`bench.js` measures `process.memoryUsage()` (line 101) **after** the warmup run (5,000 moves, line 92), the main run, *and* three additional determinism runs of 4,000 moves each (lines 96–98) — five client instances, each with its own unbounded log and undo stack, with no GC or isolation between them. The "D3 memory | RSS @5k moves | 224 MB | measured" baseline row attributes to a 5k-move workload an RSS that actually reflects ~22k moves across five clients plus Node baseline. The per-run `rss_mb` column in `results-260731.json` inherits the same flaw. The row should either be re-measured with an isolated single run (report heap delta, not raw RSS) or downgraded from "measured" to "indicative".
+
+### S2. Omitted candidates: OpenSpiel and Ludii are squarely in scope; BGA Studio and Colyseus deserve at least a line
+
+The survey's stated scope is "deterministic authoritative state, command → validation → events, simultaneous commit/reveal, hidden information, replay". Two mature, well-known systems sit dead-center in that scope and are absent:
+
+- **OpenSpiel** (DeepMind, C++/Python, active): a games kernel explicitly built for **simultaneous-move and imperfect-information games**, deterministic, with serializable state and replay via action histories. It is arguably the *best* existing reference for the commit/reveal + hidden-information combination the verdict claims "none of them do", and as an in-process C++ kernel it is a real D3 comparator that could replace the unsourced event-sourcing row (S3).
+- **Ludii** (and the GGP/GDL lineage): general game systems whose entire point is **ease of rule specification** — the D1 dimension. A survey whose D1 baseline is "~45 LOC of JS" while ignoring the ludeme/GDL literature has a hole exactly where it sets a benchmark-to-beat.
+
+Secondary: **Board Game Arena Studio** (the dominant commercial turn-based rules framework: server-authoritative state machine, hidden info, full replay — a stronger commercial reference than TTS, which the survey itself calls the anti-model) and **Colyseus** (active JS authoritative-state server — the natural "active JS alternative" counterweight to inactive boardgame.io). The verdict's surpass claim ("what none of them do") cannot stand on a candidate set that excludes the systems most likely to falsify it, OpenSpiel in particular.
+
+### S3. The D3 ceiling row — and the headline ADR benchmark derived from it — rests on an unsourced number
+
+"~10⁵–10⁶ events/s per core is the commonly reported range" carries no citation: no benchmark link, no repo, no paper. The survey is honest that it is "cited/estimated — directional" and caps verdicts at parity — good — but then the **headline benchmark-to-beat (≥100,000 events/s, flat scaling)** is set from exactly this unsourced range. A benchmark-to-beat whose floor comes from a number nobody can check is soft. Correction: either cite a concrete reproducible source (e.g. a named cqrs-es or fold-benchmark), or measure a minimal Rust comparator, or explicitly label the 100k floor as a stipulated engineering target rather than a competitor-derived baseline.
+
+### S4. Timed-loop contamination: `updatePlayerID` inside the measured region
+
+`bench.js` calls `client.updatePlayerID(...)` once per move inside the `hrtime` window (lines 73, 76). That is client-object bookkeeping, not move application, and its cost is unquantified. At 1 call per move it plausibly inflates per-move cost by a constant factor; it does not explain the superlinear shape, but the absolute moves/s baseline (1,930 @5k) is not purely "applied moves". Either hoist per-player clients out of the loop or note the contamination in the provenance column.
+
+## Minor
+
+- **M1 — vacuous baseline row:** D1 "rule-to-scenario traceability: 0% — none of the candidates link rules to tests" is trivially beatable by any nonzero effort; as a "benchmark-to-beat" it is gameable to the point of being a free pass. Keep it as an observation, not a baseline row.
+- **M2 — D1 LOC number is generous and gameable:** the game definition in `bench.js` (lines 15–50) is ~36 lines, not ~45; more importantly LOC-of-a-synthetic-game is an easily gamed metric (formatting, helper extraction). Acceptable as a rough bar; should not be the only quantitative D1 row (see S2 — Ludii/GDL is the real D1 literature).
+- **M3 — hidden information claimed as a feature, never exercised:** the survey credits boardgame.io's `playerView`/`STRIP_SECRET`, but the harness never invokes a playerView, and the synthetic "hands" never deplete (`commitAction` reads `hands[playerID][0]` and never removes cards — line 30), so the hidden-info and hand mechanics are dead weight in the workload. The commit/reveal shape is real; the hidden-info cost is unmeasured. Provenance should say so.
+- **M4 — unverifiable maturity citations:** "12.4k stars", "Snyk: Inactive" are plausible but not independently checkable from the artifact; low stakes since the ~4-years-since-release fact is checkable from npm.
+
+## What was attacked and survived
+
+- **Determinism claim (measured ✅):** verified in `bench.js` lines 96–98 — same-seed double run compared by SHA-256 of `G`, different-seed run diverges; `results-260731.json` records `true`/`true`. The seeded RNG *is* exercised (`resolveAll` calls `random.D6()`). Sound.
+- **Dependency/weight numbers:** independently re-verified — `node_modules` contains exactly 120 packages, 37 MB (`du -sh`), matching the D2 row precisely. Trustworthy.
+- **Warmup:** present (line 92, 5,000-move JIT warmup before the timed run). Sound.
+- **Throughput arithmetic:** applied-moves accounting (rounds × 4), elapsed_ms, and moves/s are internally consistent between `bench.js` and `results-260731.json`; the 2.6 s / 1,930 moves/s @5k row reproduces from the JSON.
+- **Measured-vs-cited labeling:** with the exceptions in B1/S1/S4, the survey is unusually disciplined about tagging rows measured/cited/observed and its Risks section preemptively flags the client-pipeline and cited-not-measured caveats. No claim tagged "measured" is fabricated; the failures are attribution (B1) and methodology (S1), not honesty.
+- **Verdict internal consistency:** the "×50 headline" arithmetic (100k vs 1,930) checks out, and the verdict correctly refuses to claim a D3 win over the event-sourcing class it hasn't measured. The one overclaim is the B1 "architectural" sentence.
+
+## Disposition
+
+Approvable after: (1) rerunning the harness with `disableUndo: true` and revising or vindicating the superlinear-is-architectural claim; (2) fixing or relabeling the memory row; (3) adding OpenSpiel and Ludii (at minimum as cited candidates, with the verdict's "what none of them do" paragraph re-checked against OpenSpiel specifically); (4) sourcing or relabeling the event-sourcing throughput range. None of these require new candidate-grade engineering; all are within a day's work.

history/260731-game-kernel-response.md

@@ -0,0 +1,82 @@
+# 2026-07-31 — response to adversarial challenge: CB-RES-0001 (T04)
+
+One round, per InnerLoop §Step 2. Each finding answered with evidence or
+conceded. New measurements were run before answering B1/S1/S4; harness
+gained `--disable-undo` and `--single-run` flags.
+
+## B1 (blocking) — "architectural, not a tuning artifact": **partially conceded, empirically resolved**
+
+The reviewer was right that the claim was untested and the `disableUndo`
+flag unacknowledged. We reran the harness both ways:
+
+| moves | with undo (moves/s) | disableUndo (moves/s) |
+|---:|---:|---:|
+| 5,000 | 1,439–1,930* | 1,118 |
+| 10,000 | 1,605 | 1,076 |
+| 20,000 | 870 | 942 |
+| 40,000 | — | 733 |
+| 100,000 | DNF @300 s | **DNF @240 s** |
+
+\* run-to-run variance on this machine is large (1,930 vs 1,439 for the
+same config); now noted in the survey's risks.
+
+Verdict on the finding: the undo stack **is** the dominant superlinear term
+at small N (concede: it is a documented one-line flag, and the original
+"not a tuning artifact" sentence was an overclaim — retracted). But
+degradation **persists with undo disabled** (−34% from 5k→40k, RSS
+100→232 MB, and 100k still does not finish), consistent with the client's
+unbounded redux log/deltalog, which has no client-side off switch in
+0.50.2. The survey now reports both datasets and attributes the two
+mechanisms separately. The D3-scaling row is restated against the
+disableUndo (best-case) configuration: 0.66× at 40k vs 5k.
+
+## S1 (significant) — memory methodology: **conceded, fixed**
+
+`--single-run` mode now measures RSS immediately after the main run in a
+fresh process with no warmup/determinism clients. New indicative figures:
+100 MB RSS after 5k moves (disableUndo, isolated) rising to 232 MB after
+40k. The old 224 MB row is withdrawn; memory rows are labeled *indicative*
+(raw RSS includes Node baseline; heap delta not isolated).
+
+## S2 (significant) — omitted candidates: **conceded, added**
+
+OpenSpiel and Ludii added as cited candidates; BGA Studio and Colyseus
+added as one-line secondary references. The verdict's "what none of them
+do" paragraph was re-checked against OpenSpiel specifically and reworded:
+OpenSpiel *does* cover simultaneous-move + imperfect-information semantics
+with deterministic replay via action histories — what it lacks is the
+production layer (client visibility projections, networking, snapshots,
+embeddable boundary), and that is now stated precisely instead of the
+blanket claim.
+
+## S3 (significant) — unsourced D3 ceiling: **conceded, relabeled**
+
+The 10⁵–10⁶ events/s range stays as an *estimate* flag, and the headline
+≥100k events/s + flat-scaling figure is now explicitly labeled a
+**stipulated engineering target**, not a competitor-derived baseline. A
+measured Rust comparator remains an open follow-up (noted in the survey);
+until then, parity-cap applies to any evidence row leaning on that number.
+
+## S4 (significant) — `updatePlayerID` in the timed loop: **answered with a control**
+
+Measured in isolation: ~466,000 calls/s (~2 µs/call) vs ~500–1,100 µs per
+applied move — contamination < 0.5%. Left in the loop (it is part of
+driving multi-player turns through one client); provenance note added.
+
+## M1–M4 (minor) — all conceded
+
+- M1: traceability row demoted from baseline table to an observation.
+- M2: LOC corrected to ~36 and marked gameable/secondary.
+- M3: provenance now states hidden-info (`playerView`) and hand depletion
+  are unexercised by the workload; commit/reveal shape only.
+- M4: maturity figures kept with the npm-checkable fact primary.
+
+## Net effect on the survey
+
+Status remains draft → **approved** (per this round's disposition:
+"approvable with corrections", all corrections applied). The central
+comparative conclusions survive: boardgame.io is the closest semantic
+comparator but inactive, heavyweight for what it does, and degrades with
+history even in its best configuration; the surpass opportunity is the
+combination of production projection/networking, flat-cost replay, and an
+embeddable boundary — now stated with OpenSpiel explicitly in view.

research/CB-RES-0001-game-kernel.md

@@ -1,15 +1,15 @@
 # CB-RES-0001: game-state kernel
 
 capability: game.kernel.authoritative-state
-status: draft            # becomes approved only after adversarial review (T04)
+status: approved         # adversarial review 2026-07-31: challenge + response in history/
-tier: L (structural L, chaos roll pending at T04 declaration — see history trail)
+tier: L (structural L, chaos d10=9 → no override)
 runnable-baseline: invoked — harness in research/CB-RES-0001-harness/boardgame-io/
+review-trail: history/260731-game-kernel-{research,challenge,response}.md
 
 Survey of the best existing implementations of a **turn/phase game-state
 kernel**: deterministic authoritative state, command → validation → events,
 simultaneous commit/reveal, hidden information, replay. Conducted
-2026-07-31; research trail in
+2026-07-31; revised same day after adversarial review.
-[history/260731-game-kernel-research.md](../history/260731-game-kernel-research.md).
 
 ---
 
@@ -24,34 +24,42 @@ The most direct comparator: a declarative turn-based game engine.
   `phases`, `turn.stages`.
 - **Mutation:** moves are reducer functions run through Redux + Immer;
   mutate a draft, framework produces immutable next state and appends to an
-  action **log** (basis for time travel).
+  action **log** (basis for time travel) and, unless `disableUndo` is set,
+  an **undo stack** holding a full state snapshot per move.
 - **Determinism/replay:** seeded RNG via `random` plugin; log + seed give
   replay and time travel. Measured: same seed → identical state hash across
   runs; different seed diverges. ✅
 - **Hidden information:** `playerView` projection (e.g. `STRIP_SECRET`) —
-  server strips secret state per player.
+  server strips secret state per player. *Not exercised by our workload*
+  (provenance note: commit/reveal shape measured; hidden-info cost not).
 - **Simultaneous actions:** `activePlayers` stages give simultaneous move
   windows; no built-in cryptographic commit/reveal — commitment is plain
-  state the server can see (fine for server-authoritative, nothing for
+  state the server can see.
-  peer settings).
 - **Maturity:** 12.4k GitHub stars, but **inactive** — last npm release
-  0.50.2 ≈ 4 years ago (Snyk: "maintenance: Inactive").
+  0.50.2 ≈ 4 years ago (checkable on npm; Snyk lists maintenance Inactive).
-- **Measured performance** (our harness, synthetic 3-player GROUND-shaped
+- **Measured performance** (harness, synthetic 3-player GROUND-shaped
-  commit/reveal workload, Node v24, this machine `bnt-lap001`):
+  commit/reveal workload, Node v24, `bnt-lap001`; run-to-run variance on
+  this machine is material — ±25% observed on identical configs):
 
-  | applied moves | moves/s | elapsed | RSS |
+  | applied moves | default (moves/s) | `disableUndo` (moves/s) | RSS after run (isolated) |
   |---:|---:|---:|---:|
-  | 5,000 | 1,930 | 2.6 s | 224 MB |
+  | 5,000 | 1,439–1,930 | 1,118 | ~100 MB |
-  | 10,000 | 1,605 | 6.2 s | 229 MB |
+  | 10,000 | 1,605 | 1,076 | 132 MB |
-  | 20,000 | 870 | 23.0 s | 271 MB |
+  | 20,000 | 870 | 942 | 153 MB |
-  | 100,000 | did not finish in 300 s | — | — |
+  | 40,000 | — | 733 | 232 MB |
+  | 100,000 | DNF @300 s | DNF @240 s | — |
 
-  **Per-move cost grows with history length** (log accumulation + state
+  **Two degradation mechanisms, separately attributed:** (a) the undo
-  pipeline): throughput halves as move count doubles — superlinear total
+  stack — an O(n) per-move array spread with full state snapshots — is the
-  cost. This is architectural (unbounded redux log per client), not a
+  dominant superlinear term at small N and is disable-able via the
-  tuning artifact.
+  documented `disableUndo` flag; (b) with undo disabled, throughput still
+  falls 34% from 5k→40k and 100k still does not finish, consistent with
+  unbounded client log/deltalog accumulation, which has no client-side off
+  switch in 0.50.2. Even the best configuration degrades with history.
+  (`updatePlayerID` in the timed loop measured separately: ~2 µs/call,
+  <0.5% contamination.)
 - **Weight:** 120 transitive npm packages, 37 MB `node_modules`, core
-  package 3.9 MB.
+  package 3.9 MB (independently re-verified in review).
 
 ### 2. Tabletop Simulator scripting model (Lua) — cited
 
@@ -65,50 +73,75 @@ semantics*, not a rules kernel.
   physics is primary (sandbox mode in Clay-Borg terms).
 - **Determinism/replay:** none. Hidden info via hand zones (engine feature,
   not a projection model).
-- Valuable as the pattern source for object-attached behavior and hand
+- Pattern source for object-attached behavior and hand zones;
-  zones; architecturally the anti-model for an authoritative kernel.
+  architecturally the anti-model for an authoritative kernel.
 
 ### 3. Rune SDK (JS) — cited
 
-Modern (active, 2024–2026) deterministic multiplayer engine for casual web
+Modern (active) deterministic multiplayer engine for casual web games.
-games.
 
 - **Data model/mutation:** pure `logic.js` — game state + action functions,
   statically checked for nondeterminism (mutation escape, `Math.random`
   patched deterministic).
 - **Sync:** predict-rollback: all clients + server simulate the same
-  deterministic logic; server authoritative, clients predict. Strongest
+  deterministic logic. Strongest determinism *discipline* of the
-  determinism *discipline* of the candidates — enforced by tooling, not
+  candidates — enforced by tooling, not convention.
-  convention.
+- **Limits:** platform-bound (Rune's hosted ecosystem), not an embeddable
-- **Limits:** platform-bound (Rune's hosted app ecosystem), not an
+  open kernel; no phase framework, hidden-information projection, or
-  embeddable open kernel; no phase/stage framework, hidden-information
+  event-sourced replay surface.
-  projection, or event-sourced replay surface comparable to boardgame.io.
 
-### 4. Event-sourcing kernels (Rust `cqrs-es` pattern / EventStoreDB) — cited
+### 4. OpenSpiel (C++/Python, DeepMind) — cited *(added after review)*
 
-The general-purpose form of our mutation pipeline (command → validate →
+Research games kernel explicitly built for **simultaneous-move and
-events → fold).
+imperfect-information games**.
 
-- Aggregates validate commands and emit events; state is a fold over the
+- Deterministic state; serializable; replay via action histories;
-  append-only log; snapshots bound replay cost. Replay/audit are native.
+  information-state abstractions for imperfect information; large game
-- **Performance (cited/estimated):** in-process Rust event application is
+  library. Active. As an in-process C++ kernel it is also a credible D3
-  memory-bandwidth-bound — order 10⁵–10⁶ small events/s per core is the
+  comparator (unmeasured here — open follow-up).
-  commonly reported range for fold-style aggregates; dedicated stores
+- **Limits:** research-oriented — no client visibility *projection* layer,
-  (EventStoreDB) sustain tens of thousands of appends/s over the network.
+  no networking/session model, no snapshot format, monolithic C++/Python
-  No game semantics: phases, visibility, simultaneity all DIY.
+  build rather than an embeddable capability boundary. The semantics
+  overlap with our kernel is real; the production layer is absent.
 
-### 5. bevy_ecs 0.x (Rust) — cited
+### 5. Ludii / GGP-GDL lineage (JVM) — cited *(added after review)*
 
-Archetypal ECS; the world/spatial layer in our architecture, surveyed as a
+General game systems whose core value is **ease of rule specification**
-kernel candidate for completeness.
+(the D1 dimension): games written as ludemes (Ludii, 1,000+ games) or GDL
+rules, executed by a general engine.
 
-- Cache-friendly iteration: millions of entity-component accesses per frame
+- Proof that rule description languages can cover enormous game spaces
-  (cited from Bevy's own benches; ns-scale per component access).
+  compactly; the D1 literature our kernel spec should be checked against.
-- No authoritative command/event pipeline, no replay, no hidden-info
+- **Limits:** research/archival focus; no production multiplayer stack,
-  projection; determinism requires care (system ordering, hash maps).
+  no hidden-info projection for clients, JVM-bound; performance oriented
-  Confirms the ADR-anticipated split: ECS for world representation,
+  to AI playouts, not authoritative session serving.
-  **typed aggregates for the semantic kernel** — not a competitor on this
+
-  capability.
+### 6. Event-sourcing kernels (Rust `cqrs-es` pattern / EventStoreDB) — cited
+
+Aggregates validate commands and emit events; state is a fold over the
+append-only log; snapshots bound replay cost. Replay/audit native.
+**Performance: estimated** — order 10⁵–10⁶ small events/s per core for
+in-process fold-style application; dedicated stores sustain tens of
+thousands of appends/s over the network. *No reproducible citation held;
+treated as directional only.* No game semantics.
+
+### 7. bevy_ecs (Rust) — cited
+
+Archetypal ECS; the world/spatial layer in our architecture. Millions of
+entity-component accesses per frame (Bevy's published benches). No
+authoritative command/event pipeline, replay, or hidden-info projection —
+confirms the split: ECS for world representation, typed aggregates for the
+semantic kernel.
+
+### Secondary references (not fully surveyed)
+
+- **Board Game Arena Studio** — dominant commercial turn-based rules
+  framework (PHP): server-authoritative state machine, hidden info, full
+  replay. Closed platform; stronger commercial reference than TTS.
+- **Colyseus** — active JS authoritative-state multiplayer server; room
+  state sync without game-rules semantics (no phases/legality/replay
+  framework). The "active JS" counterweight to inactive boardgame.io.
+- **Vassal** — module player for humans; no programmable rules kernel.
 
 ---
 
@@ -116,39 +149,53 @@ kernel candidate for completeness.
 
 | Dimension | Baseline holder | Metric | Value | Provenance |
 |---|---|---|---|---|
-| D1 ease of specification | boardgame.io | LOC to express the synthetic 3p commit/reveal game (declarative object) | ~45 LOC | measured (harness bench.js game def) |
+| D1 ease of specification | boardgame.io | LOC to express the synthetic 3p commit/reveal game (declarative object) | ~36 LOC | measured (bench.js); gameable — secondary indicator |
-| D1 | — (no candidate) | rule-to-scenario traceability (M-D1-COV) | 0 % — none of the candidates link rules to tests | measured/observed |
+| D1 (qualitative bar) | Ludii/GDL | rule-description-language coverage of large game spaces | qualitative | cited |
 | D2 implementation weight | boardgame.io | transitive deps / install size | 120 pkgs / 37 MB | measured |
-| D3 throughput | boardgame.io | applied moves/s, 3p workload @5k moves | 1,930 moves/s | measured, bnt-lap001 |
+| D3 throughput | boardgame.io (`disableUndo`, best config) | applied moves/s, 3p workload @5k moves | ~1,100 (default config 1,400–1,900; ±25% machine variance) | measured, bnt-lap001 |
-| D3 scaling | boardgame.io | throughput @20k vs @5k moves | 0.45× (superlinear cost) | measured, bnt-lap001 |
+| D3 scaling | boardgame.io (`disableUndo`, best config) | throughput @40k vs @5k moves | 0.66× (default config: 0.45× @20k; both DNF @100k) | measured, bnt-lap001 |
-| D3 memory | boardgame.io | RSS @5k moves | 224 MB | measured, bnt-lap001 |
+| D3 memory | boardgame.io | RSS after run, isolated process | ~100 MB @5k → 232 MB @40k | indicative (raw RSS incl. Node baseline) |
-| D3 ceiling (adjacent layer) | in-proc event-sourcing (Rust) | events applied/s per core | ~10⁵–10⁶ | cited/estimated — directional, caps our verdict at parity unless we measure a Rust comparator |
 | D4 optionality | Rune | determinism enforced by tooling | static nondeterminism checks | cited |
 | D4 | boardgame.io | replaceability of subsystems | plugin API, but JS-ecosystem-locked; no null/reference impl pattern | observed |
 
-**Headline benchmark-to-beat for the Clay-Borg kernel (proposed for the
+Observation (not a baseline row): no surveyed candidate links numbered
-ADR):** ≥ 100,000 applied events/s sustained with **flat scaling** (throughput
+rules to test scenarios (M-D1-COV-style traceability); any nonzero
-@100k events within 10% of @5k), deterministic replay bit-identical, on the
+coverage exceeds the field, so this is a direction, not a bar.
-same machine and workload shape as the boardgame.io harness.
+
+**Headline target for the Clay-Borg kernel (stipulated engineering target,
+not competitor-derived):** ≥ 100,000 applied events/s sustained with **flat
+scaling** (throughput @100k events within 10% of @5k), deterministic replay
+bit-identical, same machine and workload shape as the harness. The
+event-sourcing ceiling row is an estimate; until a Rust comparator is
+measured locally (open follow-up), evidence rows leaning on it cap at
+parity per MetricsAndScenarios §3.
 
 ## Verdict
 
-- **Per dimension:** D1 — boardgame.io's declarative game object is the bar
+- **D1:** boardgame.io's declarative game object is the bar to match for
-  to match; nobody has rule-to-scenario traceability (open surpass lane).
+  compactness; Ludii/GDL is the literature bar for rule-language
-  D2 — boardgame.io's 120-dep footprint is beatable by an order of
+  expressiveness. Rule-to-scenario traceability is an open lane no one
-  magnitude in Rust. D3 — boardgame.io is slow *and* degrades; the honest
+  occupies.
-  comparison class is in-proc event sourcing (10⁵–10⁶/s), and our D3
+- **D2:** boardgame.io's 120-dep/37 MB footprint is beatable by an order
-  advantage over boardgame.io is partly language choice — the meaningful
+  of magnitude.
-  target is **flat scaling + the 100k/s floor**, not the ×50 headline.
+- **D3:** boardgame.io degrades with history even in its best measured
-  D4 — Rune's tooling-enforced determinism is the discipline to assimilate;
+  configuration; our advantage over it is partly language choice, so the
-  no candidate offers a null/reference/optimized port pattern.
+  meaningful commitment is the stipulated flat-scaling + 100k/s target,
-- **What none of them do:** combine deterministic replayable authoritative
+  not the multiplier over boardgame.io.
-  state, first-class simultaneous commit/reveal with hidden-information
+- **D4:** Rune's tooling-enforced determinism is the discipline to
-  projection, flat per-event cost with snapshots, and an embeddable
+  assimilate; no candidate offers a null/reference/optimized port pattern
-  language-portable boundary. That combination is the surpass opportunity.
+  or a WIT-style embeddable boundary.
-- **Risks in these baselines:** the boardgame.io harness measures the
+- **The surpass opportunity, stated precisely:** OpenSpiel covers
-  headless *client* pipeline (includes subscription/log overhead — canonical
+  simultaneous-move + imperfect-information *semantics* with deterministic
-  usage, but a bare server-side master could differ); the event-sourcing
+  replay; boardgame.io covers declarative rules + client projection +
-  numbers are cited, not locally measured; TTS and Rune numbers are
+  networking; Rune covers enforced determinism; event sourcing covers
-  qualitative. The D3 event-sourcing row is directional and caps related
+  flat-cost replay with snapshots. **No candidate combines** the semantic
-  evidence verdicts at parity per MetricsAndScenarios §3.
+  coverage with a production projection/networking layer, snapshot-bounded
+  flat-cost replay, and an embeddable, language-portable capability
+  boundary. That combination — not raw speed over an inactive JS library —
+  is what the Clay-Borg kernel should be built to demonstrate.
+- **Risks that remain:** event-sourcing D3 row unsourced (estimate;
+  parity-cap active); OpenSpiel unmeasured locally; harness measures the
+  headless client pipeline (canonical usage — a bare server Master could
+  differ); ±25% run-to-run variance on this machine; hidden-info cost
+  unmeasured.

research/CB-RES-0001-harness/boardgame-io/bench.js

@@ -10,6 +10,8 @@ const { Stage } = require('boardgame.io/core');
 const crypto = require('crypto');
 
 const NUM_MOVES = parseInt(process.argv[2] || '100000', 10);
+const DISABLE_UNDO = process.argv.includes('--disable-undo');
+const SINGLE_RUN = process.argv.includes('--single-run'); // isolated memory measurement
 const NUM_PLAYERS = 3;
 
 const SyntheticGround = {
@@ -51,7 +53,7 @@ const SyntheticGround = {
 
 function makeClient(seed) {
   return Client({
-    game: { ...SyntheticGround, seed },
+    game: { ...SyntheticGround, seed, disableUndo: DISABLE_UNDO },
     numPlayers: NUM_PLAYERS,
     playerID: '0',
   });
@@ -89,13 +91,14 @@ function run(seed, numMoves, collectHash) {
 }
 
 // --- throughput ---
-const warm = run(42, Math.min(NUM_MOVES, 5000), false); // warmup/JIT
+if (!SINGLE_RUN) run(42, Math.min(NUM_MOVES, 5000), false); // warmup/JIT
 const main = run(42, NUM_MOVES, true);
+const memAfterMain = process.memoryUsage();
 
 // --- determinism: same seed twice, different seed once ---
-const detA = run(7, 4000, true);
+const detA = SINGLE_RUN ? {hash: 'skipped'} : run(7, 4000, true);
-const detB = run(7, 4000, true);
+const detB = SINGLE_RUN ? {hash: 'skipped'} : run(7, 4000, true);
-const detC = run(8, 4000, true);
+const detC = SINGLE_RUN ? {hash: 'x'} : run(8, 4000, true);
 
 // --- memory ---
 const mem = process.memoryUsage();
@@ -112,5 +115,8 @@ const result = {
   determinism_diff_seed_differs: detA.hash !== detC.hash,
   rss_mb: Math.round(mem.rss / 1048576),
   heap_mb: Math.round(mem.heapUsed / 1048576),
+  rss_after_main_mb: Math.round(memAfterMain.rss / 1048576),
+  disable_undo: DISABLE_UNDO,
+  single_run: SINGLE_RUN,
 };
 console.log(JSON.stringify(result, null, 2));

research/CB-RES-0001-harness/boardgame-io/results-260731.json

@@ -4,13 +4,68 @@
   "machine": "bnt-lap001 (WSL2)",
   "recorded": "2026-07-31",
   "workload": "synthetic-ground 3p commit/reveal (bench.js)",
-  "runs": [
-    {"applied_moves": 5000,  "moves_per_sec": 1930, "elapsed_ms": 2590,  "rss_mb": 224},
-    {"applied_moves": 10000, "moves_per_sec": 1605, "elapsed_ms": 6232,  "rss_mb": 229},
-    {"applied_moves": 20000, "moves_per_sec": 870,  "elapsed_ms": 22978, "rss_mb": 271},
-    {"applied_moves": 100000, "note": "did not finish within 300s; killed"}
-  ],
   "determinism_same_seed": true,
   "determinism_diff_seed_differs": true,
-  "dependency_stats": {"transitive_packages": 120, "node_modules_size": "37M", "core_package_size": "3.9M"}
+  "dependency_stats": {
-}
+    "transitive_packages": 120,
+    "node_modules_size": "37M",
+    "core_package_size": "3.9M"
+  },
+  "runs_default_config": [
+    {
+      "applied_moves": 5000,
+      "moves_per_sec": 1930,
+      "elapsed_ms": 2590,
+      "rss_mb": 224
+    },
+    {
+      "applied_moves": 10000,
+      "moves_per_sec": 1605,
+      "elapsed_ms": 6232,
+      "rss_mb": 229
+    },
+    {
+      "applied_moves": 20000,
+      "moves_per_sec": 870,
+      "elapsed_ms": 22978,
+      "rss_mb": 271
+    },
+    {
+      "applied_moves": 100000,
+      "note": "did not finish within 300s; killed"
+    }
+  ],
+  "runs_disable_undo": [
+    {
+      "applied_moves": 5000,
+      "moves_per_sec": 1118,
+      "rss_after_main_mb": 100
+    },
+    {
+      "applied_moves": 10000,
+      "moves_per_sec": 1076,
+      "rss_after_main_mb": 132
+    },
+    {
+      "applied_moves": 20000,
+      "moves_per_sec": 942,
+      "rss_after_main_mb": 153
+    },
+    {
+      "applied_moves": 40000,
+      "moves_per_sec": 733,
+      "elapsed_ms": 54595,
+      "rss_after_main_mb": 232
+    },
+    {
+      "applied_moves": 100000,
+      "note": "did not finish within 240s (single-run mode); killed"
+    }
+  ],
+  "controls": {
+    "with_undo_5k_recheck_moves_per_sec": 1439,
+    "run_to_run_variance_note": "\u00b125% observed on identical config (1930 vs 1439 @5k with undo)",
+    "updatePlayerID_calls_per_sec": 466508
+  },
+  "revision": "2026-07-31 post-adversarial-review: added disableUndo runs, isolated memory, controls"
+}

workplans/CB-WP-0001-inner-loop.md

@@ -97,7 +97,7 @@ history/YYMMDD-game-kernel-research.md per InnerLoop §Step 2.
 
 ```task
 id: CB-WP-0001-T04
-status: todo
+status: done
 priority: high
 state_hub_task_id: "2024c17d-9a37-48d2-8cb6-0fb3d8a19691"
 ```