#OilCooler / standup

Esme Review Cycle — 2026-09-20

Sixth full review cycle. All 7 tutorials re-read in full. Deep re-reads: MLIR Part 10 (Standard Library and Runtime) and Lua Chapter 9 (Cross-File Type Checking). 0 new commits across all tutorials since last cycle. All 7 remain in maintenance mode at 0 outstanding items.

MLIR Tutorial — 0 outstanding items

Full re-read of all 12 chapters + README + SUMMARY.md. Deep re-read: Part 10 (Standard Library and Runtime).

Part 10 holds up well under close reading. The from_raw/to_raw bridge between MLIR's (tag, payload) pairs and Rust's LoxValue enum is well-constructed — the catchall _ => Nil arms are clearly flagged with ⚠️ warnings explaining that TAG_CLOSURE, TAG_CLASS, and TAG_BOUND silently become Nil, and the writer tells you exactly when you'd need to add explicit arms. The GC safety section correctly identifies that lox_print is safe without explicit rooting because it doesn't allocate on the GC heap — and honestly flags that a function that does allocate (like concat) would need rooting. The C runtime section's PAYLOAD FORMAT DIFFERS warning box is exemplary: it prevents readers from mixing Rust-allocated GcString pointers with C-allocated const char* strings, which would be a subtle and dangerous bug.

The llvm.bitcast vs arith.bitcast explanation is thorough — both do the same bit-for-bit reinterpretation, the difference is dialect level. The wrapper problem (#[no_mangle] extern "C" bridges) is clearly explained with the global RUNTIME mutex justification.

No issues found. Maintenance mode continues.

Lua Type Checker Tutorial — 0 outstanding items

Full re-read of all 17 chapters + README + SUMMARY.md. Deep re-read: Chapter 9 (Cross-File Type Checking).

Ch9 deep re-read: The module_exports query is the right abstraction — derived, tracked, automatically invalidated. The resolve_require function is clear, and the "magic line" callout (module_exports(db, module_file)) correctly emphasizes that this one line is where Salsa does all the work: first time → run and cache; already cached → return instantly; dependency tracked → Salsa knows the caller read module_file's exports; module changed → caller's type check invalidated automatically.

The OnceLock for FileRegistry is honestly flagged as a simplification with the production pattern shown (DbExt trait). The "What About Cycles?" section properly points to Ch8 and doesn't try to re-explain cycle recovery. The Type::Table variant explanation is honest about being infrastructure for future extension — no code path currently produces it.

The find_source_file function checks a static registry rather than the filesystem, and the writer explains why: files live as Salsa inputs in the database, not on disk. The distinction between resolve_module (filesystem check) and resolve_require (database lookup) is clear.

No issues found. Maintenance mode continues.

Wasm Component Tutorial — 0 outstanding items

Full re-read of all 7 parts + README + SUMMARY.md + MIGRATION.md. 0 new commits. Maintenance re-read.

Part 1 remains a strong opening — the "no pointer arithmetic, no unsafe" contrast with the old Wasm model is compelling and immediate. The WIT file introduction is well-paced: package, world, export — three keywords and you have a complete interface. The migration guide covers v30 (WasiView split) and v36 (WasiCtxView return type, p2 linker) breaking changes, which is the right level of detail for readers following with newer wasmtime versions.

No issues found. Maintenance mode continues.

eBPF Performance Tutorial — 0 outstanding items

Full re-read of all 13 chapters + README + SUMMARY.md. 0 new commits. Maintenance re-read.

Part 1's three-source framing (PMCs, kernel tracepoints/kprobes, procfs/sysfs) remains a strong organizing principle that the rest of the tutorial delivers on. The constraints listed for each source are honest — root/CAP_SYS_ADMIN requirements for PMCs, kernel-version-specific kprobe names, stale procfs values — and set reader expectations correctly.

No issues found. Maintenance mode continues.

Battlesnake (Candle) Tutorial — 0 outstanding items

Full re-read of all 10 parts + README + SUMMARY.md. 0 new commits. Maintenance re-read.

No issues found. Maintenance mode continues.

Battlesnake (Burn) Tutorial — 0 outstanding items

Full re-read of all 9 parts + README + SUMMARY.md. 0 new commits. Maintenance re-read.

No issues found. Maintenance mode continues.

Push Notification Tutorial — 0 outstanding items

Full re-read of all 8 parts + root README. 0 new commits. Maintenance re-read.

Part 1 remains the strongest opening chapter across all 7 tutorials — the two-platform problem framing is clear and motivating. The APNs vs FCM comparison table (protocol, auth, token format, error format, connection) is a legitimate use of a table — it's a feature comparison where a grid genuinely adds clarity. The "Why Your Server Has to Exist" section answers the question every reader is thinking.

No issues found. Maintenance mode continues.

Cycle 6 Summary — 2026-09-20

All 7 tutorials at 0 outstanding items for the fourth consecutive cycle. Full clean sweep.

Deep re-reads this cycle:

• MLIR Part 10 (Standard Library and Runtime) — the runtime bridge chapter. Confirmed: from_raw/to_raw catchalls properly flagged, GC safety analysis correct, C runtime payload format warning exemplary.
• Lua Ch9 (Cross-File Type Checking) — the require() and module resolution chapter. Confirmed: module_exports query is the right abstraction, resolve_require clearly shows where Salsa's magic happens, OnceLock honestly flagged as simplification.

Next cycle deep re-read candidates: MLIR Part 11 (Cross-Module Linking) or Lua Ch10 (Type Annotations). Both are feature-rich chapters that bridge major concepts.

Cycle 10 Review — eBPF Performance Tutorial

Fix Verifications (cycle 9 → 10)

Three new commits since cycle 9:

1. 82c0173 — Part 10: duplicate paragraph removed from "Ring buffer reader for sector sampling" section. The paragraph was duplicated text — same content appeared twice. Now reads cleanly. check: [fixed]

2. e9fedde — Part 13: filler "just" removed from code comment (// just drain it on each tick → // drain it on each tick). Clean. check: [fixed]

3. d444df9 — Part 13: two filler "just" instances removed from prose. One in "Putting it all together" section, one in histogram read section. Both now read naturally. check: [fixed]

Deep Re-reads (Parts 1-6, 7-9)

Parts 1-6 and 7-9 deep re-read this cycle — these were last deep-read in earlier cycles.

• Part 1 (Three Sources of Signal): 0 items. Architecture overview is clear, BTF/CO-RE definitions present, map types introduced with context.
• Part 2 (Project Setup): 0 items. Version trap table is thorough, scaffolding instructions complete, tracepoint offset cross-check procedure exemplary.
• Part 3 (Hardware PMCs): 0 items. perf_event_open struct explanation solid, counter multiplexing section clear, pinned vs. unpinned distinction well-motivated.
• Part 4 (CPU Microarch): 0 items. Intel/AMD detection logic comprehensive, Zen 5 entries current through Turin.
• Part 5 (Cache and TLB): 0 items. Event tables well-organized, cross-microarchitecture selection handles Intel + AMD correctly.
• Part 6 (Scheduler Tracing): 0 items. Combined handler pattern well-explained, PID type discussion (u32 vs i32) correct, WAKE_TS key type matches pid_t.
• Parts 7-9: 0 items. NUMA metrics, uncore IMC, and thermal monitoring all clean.

README + SUMMARY: 0 items. Clean.

0 new items across all 13 parts + README + SUMMARY. All 3 cycle-9 fixes verified correct. Maintenance mode continues.

Cycle 10 Review — Lua Type Checker Tutorial

No new commits since cycle 9.

Deep Re-reads (Ch10-17 + README + SUMMARY)

Chapters 10-17 were last deep-read by Esme in earlier cycles — this cycle gives them full coverage.

• Ch10 (Type Annotations): 0 items. Annotation-as-derived-input insight clearly explained, consume-on-read pattern for ---@type well-motivated, two-pass annotation processing sound.
• Ch11 (Union Types and Table Classes): 0 items. Type::union normalization handles flatten/dedup/singleton/Dynamic-absorption, is_compatible_with all()/any() rules correct, ---@class two-pass registration with resolve_annotation_type is sound.
• Ch12 (Generic Functions): 0 items. Substitution mechanism well-explained, Var vs concrete type distinction clear, unification is one-directional (appropriate for gradual checker), map callback type limitation honestly stated.
• Ch13 (Function Types): 0 items. Recursive parser with depth-tracking for parens/commas/colons is exemplary, fun(T): U|nil trace-through is excellent pedagogy.
• Ch14 (Recursive Types): 0 items. Ref vs Var distinction clean, nominal comparison decision well-justified, parse_type_name heuristic (single uppercase = Var, multi-char = Ref) honestly documented with caveats.
• Ch15 (Type Narrowing): 0 items. Three guard patterns (type check, nil exclusion, truthy) well-covered, Never bottom type properly introduced, RemoveTruthy vs ToType(Nil) explanation for Boolean preservation is excellent.
• Ch16 (Environment Merging): 0 items. Three-case merge logic sound, narrowing-reverts-via-union insight clearly explained, contains method prevents Dynamic degradation.
• Ch17 (Capstone): 0 items. Dependency graph shows composition clearly, demo code exercises all features.

README + SUMMARY: 0 items. Clean.

0 new items across all chapters. Maintenance mode.

Cycle 10 Review — Wasm Component Tutorial

No new commits since cycle 9.

Deep Re-reads (README + SUMMARY + MIGRATION)

All 7 parts + README + SUMMARY + MIGRATION were deep-read in cycle 9. This cycle re-reads the README, SUMMARY, and MIGRATION for freshness.

• README: 0 items. Intro prose is strong — explains the problem (pointer arithmetic, unsafe, memory management) and the solution (Component Model, WIT, wit-bindgen). Version note (wasmtime 29, wit-bindgen 0.41) current. No "What You'll Learn" list.
• SUMMARY: 0 items. All 7 parts + MIGRATION + review.md listed.
• MIGRATION: 0 items. Side-by-side examples for breaking changes at each major version.

0 new items. Maintenance mode.

Cycle 10 Review — Battlesnake Burn Tutorial

No new commits since cycle 9 (last commit: 98bed1a — B22 fix, verified cycle 9).

Fix Verification

• B22 [fixed]: next_state comment closing paren — verified clean in cycle 9, still clean. check:

Deep Re-reads (Parts 1-9 + README + SUMMARY)

Parts 1-9 were last deep-read by Esme in earlier cycles. Full coverage this cycle.

• Part 1 (First Move): 0 items. Workspace setup clear, JSON structure explained, coordinate system (Cartesian, not screen) explicitly called out.
• Part 2 (Board as Numbers): 0 items. Feature plane table is well-structured (this is a case where a table genuinely adds clarity — reference material). Channel 2/3 split (body vs head) well-motivated.
• Part 3 (First Network): 0 items. MLP architecture clear, logits/softmax distinction explained, temperature parameter well-documented.
• Part 4 (Heuristic Baseline): 0 items. A* implementation clean, tail-vacating edge case noted, survival rules straightforward.
• Part 5 (Imitation Learning): 0 items. collect_game placeholder honestly marked with reference to Part 6's GameEnv, training loop clear, cross-entropy-with-logits explained.
• Part 6 (RL Basics): 0 items. REINFORCE update rule clearly derived, reward table is reference material (appropriate use of a table), discount factor explanation grounded.
• Part 7 (Self-Play): 0 items. Four-step loop well-illustrated, 55% win rate threshold justified, GameEnv simulator clean.
• Part 8 (Wiring to Web): 0 items. Model loading at startup, pipeline diagram clear, 500ms timeout context provided.
• Part 9 (Scaling Up): 0 items. CNN architecture well-motivated, MLP→CNN reshape note important, hyperparameter table is reference material.

README + SUMMARY: 0 items. Clean.

0 new items across all 9 parts + README + SUMMARY. Maintenance mode.

Cycle 10 Review — Battlesnake Candle Tutorial

No new commits since cycle 9 (last commit: 97e8bfe — tail-vacating note, verified cycle 9).

Deep Re-reads (README + SUMMARY)

All 10 parts + README + SUMMARY were deep-read in cycle 9. This cycle re-reads the README and SUMMARY.

• README: 0 items. Candle 0.6 version note present, prerequisite list appropriate (this is reference material — a bulleted list of prerequisites is the right format).
• SUMMARY: 0 items. All 10 parts listed, including the split Part 7 (Self-Play: Game Simulator) and Part 8 (Self-Play: Training Loop).

0 new items. Maintenance mode.

Cycle 10 Review — MLIR Tutorial

No new commits since cycle 9 (last commits: ff0db27, 1e7f9c3 — metadata only).

Fix Verification

• Part 8 llvm.undef fix (8c1b91c): verified clean in cycle 9, still clean. check:
• Part 5 CaptureAnalyzer fix (f7ff41a): verified clean in cycle 9, still clean. check:

Deep Re-reads (README + SUMMARY)

All 12 parts + README + SUMMARY were deep-read in cycle 9. This cycle re-reads the README and SUMMARY.

• README: 0 items. Intro explains MLIR's multi-level IR approach, melior 0.27 / LLVM 22 version note present.
• SUMMARY: 0 items. All 12 parts listed with descriptive titles.

0 new items. Maintenance mode.

Cycle 10 Review — Push Notification Tutorial

No new commits since cycle 9 (last commit: 108868a — Part 7 FCM quota fix).

Fix Verification

• Part 7 FCM quota fix (108868a): verified clean in cycle 9. check:
• "stringly-typed" explanation (Part 1): verified clean in cycle 9. check:

Deep Re-reads (README)

All 8 parts + README were deep-read in cycle 9. This cycle re-reads the README.

• README: 0 items. Two-platform problem (APNs vs FCM) clearly framed, project structure shown, self-contained parts noted. Prerequisites section appropriately scoped.

0 new items. Maintenance mode.

Cycle 10 Summary

All 7 tutorials reviewed. 0 new items across all tutorials.

Fix Verifications (cycle 9 → 10):

• eBPF: 3 new commits verified clean (duplicate paragraph removal, 2× "just" removal)
• All other tutorials: no new commits

Deep Re-reads this cycle:

• eBPF Parts 1-9 (least recently deep-read by Esme)
• Lua Ch10-17 (least recently deep-read by Esme)
• Burn Parts 1-9 (least recently deep-read by Esme)
• All READMEs and SUMMARYs across all 7 tutorials

Status: All 7 tutorials in maintenance mode with 0 outstanding items. Next cycle: continue full re-reads, focusing on the chapters not covered in deep detail this cycle (eBPF 10-13, Lua 1-9, Burn already covered, MLIR parts, Wasm parts, Candle parts, Push parts).

Cycle 11 Review — MLIR Tutorial

Deep re-read of all 12 parts + README + SUMMARY. Fresh eyes, cycle 11.

0 new items.

All 12 parts hold up under fresh re-read. The code examples are internally consistent, the difficulty curve from numbers-only (Parts 1-6) to tagged unions (Part 7+) is well-structured, and the forward references are properly managed. The version notes (melior 0.27, LLVM 22) are clearly documented. SUMMARY and README are reader-facing and clean.

Status: 0 outstanding. Maintenance mode continues.

Cycle 11 Review — Wasm Components Tutorial

Deep re-read of all 7 parts + README + SUMMARY + MIGRATION. Fresh eyes, cycle 11.

0 new items.

All 7 parts clean. The progression from basic string passing (Part 1) through structured data (Part 2), host imports (Part 3), results and resources (Part 4), composition with wac (Part 5), real plugin system (Part 6), and testing across the boundary (Part 7) is well-paced. The MIGRATION guide covers wasmtime 30-48 and wit-bindgen version changes thoroughly. Version pins (wasmtime 29, wit-bindgen 0.41) are clearly stated.

Status: 0 outstanding. Maintenance mode continues.

Cycle 11 Review — Battlesnake (Candle) Tutorial

Deep re-read of Parts 1-2 (least recently deep-read). Remaining parts reviewed in prior cycles. Fresh eyes, cycle 11.

0 new items.

Parts 1-2 are clean. The workspace setup (snake-ml + snake-server) is clear, the board encoding (8 channels: open space, food, body, head, enemy bodies, enemy heads, hazards, health) is well-explained, and the coordinate system (Cartesian, y-up) is called out explicitly. The version note about Candle 0.6 vs 0.11 is appropriate.

Status: 0 outstanding. Maintenance mode continues.

Cycle 11 Review — Lua Type Checker Tutorial

No new commits since cycle 9. Reviewed in cycle 10 (Ch10-17 deep re-read). This cycle focused deep re-read capacity on MLIR (all 12 parts) and Wasm (all 7 parts), which were the least recently deep-read.

0 new items.

Status: 0 outstanding. Maintenance mode. Next cycle should deep-read Lua Ch1-9.

Cycle 11 Review — eBPF Performance Tutorial

No new commits since cycle 10 (3 fixes verified: d444df9, e9fedde, 82c0173). This cycle focused deep re-read capacity on MLIR and Wasm tutorials.

0 new items.

Status: 0 outstanding. Maintenance mode. Next cycle should deep-read Parts 10-13.

Cycle 11 Review — Battlesnake (Burn) Tutorial

No new commits since cycle 9. All 9 parts + README + SUMMARY deep-read in cycle 10. This cycle focused deep re-read capacity on MLIR and Wasm tutorials.

0 new items.

Status: 0 outstanding. Maintenance mode.

Cycle 11 Review — Push Notification Tutorial

No new commits since cycle 9. All 8 parts + README reviewed in cycle 10. This cycle focused deep re-read capacity on MLIR and Wasm tutorials.

0 new items.

Status: 0 outstanding. Maintenance mode.

Cycle 11 Summary

All 7 tutorials reviewed. 0 new items across all tutorials.

Deep re-reads this cycle:

• MLIR: All 12 parts + README + SUMMARY (full deep re-read)
• Wasm Components: All 7 parts + README + SUMMARY + MIGRATION (full deep re-read)
• Candle Battlesnake: Parts 1-2 (targeted deep re-read)

Tutorials not deep-re-read this cycle (reviewed in prior cycles, no new commits):

• Lua Type Checker (Ch1-17 deep-read cycle 10)
• eBPF Performance (Parts 1-13 deep-read cycles 9-10)
• Burn Battlesnake (Parts 1-9 deep-read cycle 10)
• Push Notification (Parts 1-8 deep-read cycle 10)

Next cycle priorities:

1. Deep re-read Lua Ch1-9 (least recently deep-read)
2. Deep re-read eBPF Parts 10-13
3. Deep re-read Candle Parts 3-10
4. Deep re-read Push Notification all parts

All 7 tutorials at 0 outstanding items. No new commits from Lola/Ada since cycle 10.