The Frontier (fastbook ch.7–12), mobile fixes, district placards & tour polish

[CodeWithOz]

What this adds

🌉 The Frontier — fastbook chapters 7–12 as a new city area

The map extends south across a river into the Frontier, where the data stops being images. Six new buildings, 23 inspection panels, four new live trainers — every number on screen still comes from real tensors training in the tab. Models were checked against the actual book notebooks, now vendored at reference/fastbook-master.zip (AGENTS.md instructs future agents to read the chapters before building on them):

District / building	Book material
Refinement Row — Refinement Gym	ch.7: normalization (real train-set stats), label smoothing ((1−ε+ε/N) targets), mixup (λ∈0.5–1 per the book's pseudocode), TTA over 5 shifted views — all racing an identically-initialized plain twin on the same batches
Taste Quarter — Taste Cinema	ch.8: the book's DotProductBias (factors + biases + sigmoid_range(0,5.5)) with weight decay on a planted-structure ratings grid; dot-product desk, learned-bias interpretation, factor projector vs planted genres
Table Grove — Decision Arboretum	ch.9: CART trees by variance reduction (the candidate-question audition), a walkable tree, bagged forest with OOB error + importance, and a "regrow the forest" operation showing sampling variance live
Language Lane — Tokenizer Mill + Sentiment Studio	ch.10–11: tokenize → vocab (min-freq) → numericalize pipeline with xxbos/xxunk; live bag-of-words classifier with one inspectable weight per word, verdict updates as you type
Sequence Summit — Echo Tower	ch.12: the book's LMModel2 on human numbers, next-word quiz on the held-out tail, most-common-token baseline, greedy/sampled generation

Frontier datasets are deterministic generators with planted structure + noise — models must rediscover it live, which lets panels honestly compare planted truth vs learned parameters.

📱 Mobile fixes

• D-pad uses one ▲ glyph CSS-rotated per direction (no iOS emoji arrows, no mixed glyph sizes); ▶/⏸/⏭ get text-presentation selectors
• Interiors clamp-follow the avatar like the street camera (no more walking offscreen)
• Panels scroll vertically only: canvases shrink keeping aspect ratio, tables scroll in their own box, long pickers capped
• Drawing canvas gets touch-action: none — dragging paints instead of scrolling

🪧 World & tour polish

• District names are lawn-sign placards (navy board, gold text, posts) with decor cleared beneath; wider grass strips
• Plaza redesigned: a network monument (edges light up while the main model trains) + a tour kiosk — press E beside it to ride a scripted express glide to stop ①; avatar spawns next to it
• "↓ THE FRONTIER" markers on all river crossings; chevrons continue past stop ⑮ to the junction and stop at the final stop ㉑
• Loop pulses are now chips carrying the actual digits of the current mini-batch
• Fixed a pre-existing fmt() bug that truncated round integers everywhere (1700 → "17")

Testing

• npm test: 29/29 (gradient checks, training smoke tests, new frontier scenario/tree/generator tests)
• tsc --noEmit and production build clean
• Walked the city in-browser at mobile (375px) and desktop sizes: all 58 panels open without console errors or horizontal overflow

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features
  • Added Frontier district with six new interactive buildings showcasing advanced ML concepts: collaborative filtering (movie recommendations), sentiment analysis, RNN language modeling, decision trees/random forests, and training refinement techniques.
  • Expanded guided city tour to 21 stops with new "express travel" mode featuring smooth easing.
  • Added interactive visualization panels for each new scenario with real-time exploration tools.

[coderabbitai]

Warning

Review limit reached

@CodeWithOz, we couldn't start this review because you've reached your PR review rate limit.

More reviews will be available in 29 minutes and 42 seconds. Learn how PR review limits work.

Your organization has run out of usage credits. Purchase more credits in the billing tab to continue.

⌛ How to resolve this issue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available.

Please see our Fair Usage Limits Policy for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: c2f4e966-13b9-48dc-b886-2fddebf945ad

📥 Commits

Reviewing files that changed from the base of the PR and between 5549310 and 4ecb3ff.

📒 Files selected for processing (2)

• src/sim/tabular.ts
• tests/frontier.test.ts

Walkthrough

This PR introduces the Frontier, a substantial expansion adding four new machine-learning training scenarios, synthetic dataset generators, interactive UI panels, decision-tree/forest models, world geography expansion, and an express-tour mechanic. The implementation progresses from data utilities through scenario implementations, world integration, comprehensive UI instrumentation, city geography, game mechanics, and testing.

Changes

Frontier ML Models & Interactive Learning Environment

Layer / File(s)	Summary
Synthetic Dataset Generators src/sim/datasets.ts	Deterministic seeded data generators for collaborative filtering (user/movie ratings), text sentiment reviews with tokenization, human-readable number sequences, and tabular regression (rent predictions). All use fixed holdout rules and controlled noise for reproducibility.
Scenario Type & IDs src/sim/scenarios.ts, src/sim/scenarios2.ts	Expanded ScenarioId union to include refinery, collab, sentiment, rnnlm; added scenario module boilerplate with documentation and helper utilities.
Refinery Scenario: Normalization, Label-Smoothing, Mixup, and TTA src/sim/scenarios2.ts	Configurable MNIST digit classifier with optional input normalization, label smoothing, and mixup augmentation. Runs plain twin in parallel for comparison, records loss graphs, and provides test-time augmentation via shifted-image averaging.
Collaborative Filtering Scenario: Matrix Factorization src/sim/scenarios2.ts	User/movie factor matrix factorization with learned biases, sigmoid-scaled 0..5.5 predictions, MSE + L2 loss, and factor/bias inspection helpers.
Sentiment & Language Model Scenarios src/sim/scenarios2.ts	SentimentNet (bag-of-words logistic regression with per-token weights and contribution analysis) and RnnLm (unrolled RNN with context embedding, hidden-state iteration, and temperature-sampled generation).
Decision Trees and Random Forests for Tabular Regression src/sim/tabular.ts	Gradient-free CART-style tree fitting with recursive splits, feature-subset support, and recursive paths. Bootstrap ensemble with seeded PRNGs, out-of-bag error, feature importance via split-gain accumulation, and Arboretum singleton precomputing models on rent data.
World Class: New Trainers and Scenario Access src/sim/world.ts	Four new Trainer instances (refinery, cinema, sentiment, echo) with public getters for underlying scenarios and expanded trainer list for consistent warm-up.
Refinery UI Panels: Training Instrumentation src/ui/panels/refinery.ts	Five interactive panels: Twin Race Bench (refined vs. plain metrics), Normalization Bar (batch statistics), Label-Smoothing Bench (smoothed targets), Mixup Blender (mixed inputs), and Test-Time Augmentation panel.
Collaborative Filtering UI: Ratings and Factors src/ui/panels/collab.ts	Four cinema panels: ratings heatmap, training bench with RMSE, dot-product factor/bias inspection, and factor projector with 2D embedding visualization and bias rankings.
Decision Tree & Forest UI: Regression Inspection src/ui/panels/arbor.ts	Split audition table with MSE/gain, interactive tree path viewer with fold/unfold, and forest metrics panel with out-of-bag explanation and regrow control.
Language & Sentiment UI: Tokenization and Classification src/ui/panels/language.ts	Six panels: tokenization, vocabulary wall, numericalization heatmap, training bench, free-form text classifier with top-token contributions, and learned-weight wall.
RNN Language Model UI: Corpus and Generation src/ui/panels/echo.ts	Four panels: corpus statistics, hidden-state visualization, next-word prediction quiz with top-8 probabilities, and generation interface with greedy/sampled modes.
Panel Registration, Metrics Formatting, and Styling src/ui/panels/index.ts, src/ui/panels/common.ts, src/style.css, src/ui/hud.ts, src/ui/touch.ts, src/ui/widgets.ts	Wires all new panel modules, adds fmtMetric helper for consistent formatting, introduces Frontier widget CSS (chips, trees, path highlighting), updates emoji glyphs for iOS consistency, refines D-pad rotation, and adjusts mobile responsive layout.
Buildings and World Geography Expansion src/world/buildings.ts	Expands trainer type for six identifiers, repositions nine buildings downward 2 tiles, adds six new Frontier buildings (gym, cinema, arbor, tokenmill, sentstudio, echo) with placements and interior stations.
City Map: Monument, Kiosk, and District Signs src/world/city.ts	Expands MAP_H to 65, introduces Monument/Kiosk tiles, defines district-sign system with cleared footprints, rewrites city generation, updates collision, and adds monument/kiosk rendering. Training-loop visualization now renders digit-carrying chips instead of dots.
Game: Express Tour Mechanics and Avatar Movement src/world/game.ts	Adds scripted tour mode: avatar spawn repositioned, travel path and sparkle trail tracking, kiosk-triggered express ride, eased waypoint interpolation with direction updates, trail fading, and center-when-fits camera behavior. Suppresses prompts during travel and adds kiosk-specific interaction text.
Documentation, Guides, and End-to-End Tests AGENTS.md, README.md, src/main.ts, tests/frontier.test.ts, tests/training.test.ts	Expanded AGENTS.md with concept naming and Frontier conventions; updated README with longer tour range and expanded curriculum map; updated welcome overlay narrative; added comprehensive test suite for data generators, scenario training, tree/forest metrics, and refinery TTA.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~75 minutes

The PR introduces substantial new functionality across multiple domains (data generation, ML model implementations, UI panels, world mechanics) with moderate logic density. While individual components follow clear patterns, the breadth of changes (5 new UI modules, tabular models, game mechanics, city expansion) and interconnected dependencies require careful review of each checkpoint. The implementation is self-contained but dense enough to demand thorough inspection of scenario semantics, tree-fitting correctness, UI wiring, and game-state management.

🐰 A Frontier awaits, where digits dance and models play,
Ten buildings born to teach what weights convey,
From chips that train to trees that split with gain,
The city grows—tour ① guides us down the lane! ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 40.82% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main changes: adding the Frontier (chapters 7-12), mobile UI fixes, and district/tour improvements.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/musing-leavitt-bee803

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[coderabbitai]

Actionable comments posted: 3

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/sim/tabular.ts`:
- Around line 187-208: oobMae and predictForest currently divide by
counted/trees.length without guarding against zero, causing NaN; update oobMae
(function oobMae, referencing forest.trees, forest.bags and predictTree) to
return a safe numeric default (e.g. 0) when counted === 0 before doing total /
counted, and update predictForest (function predictForest, which calls
predictTree) to return the same safe default when trees.length === 0 before
dividing; keep existing aggregation logic otherwise.

In `@tests/frontier.test.ts`:
- Around line 28-42: The loadMnistFromDisk() helper lacks validation of the
MNIST file magic number; update loadMnistFromDisk to read and verify the first 4
bytes/magic (e.g., "DLW1") before proceeding with offsets and throw a clear
format/error when it doesn't match so corrupt files fail fast; locate and mirror
the validation approach used in tests/training.test.ts but implement it inside
loadMnistFromDisk (reference function name loadMnistFromDisk and the
DataView/Uint8Array buffer handling) and ensure the error message clearly
indicates an invalid MNIST magic/header.
- Line 156: The test hard-codes the feature index as 1 when asserting feature
importance (expect(Math.max(...imp)).toBe(imp[1])); change it to derive the
index from the feature names instead of a literal: compute sizeIdx =
TAB_FEATS.indexOf("size m²") (and assert sizeIdx !== -1) then use
expect(Math.max(...imp)).toBe(imp[sizeIdx]); this keeps the assertion robust if
TAB_FEATS or feature ordering from makeTabularData()/featureImportance()
changes.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: f21c6bc1-f3f3-4eb2-ac93-f781ef761021

📥 Commits

Reviewing files that changed from the base of the PR and between b2b83a9 and 5549310.

⛔ Files ignored due to path filters (1)

• reference/fastbook-master.zip is excluded by !**/*.zip

📒 Files selected for processing (24)

• AGENTS.md
• README.md
• src/main.ts
• src/sim/datasets.ts
• src/sim/scenarios.ts
• src/sim/scenarios2.ts
• src/sim/tabular.ts
• src/sim/world.ts
• src/style.css
• src/ui/hud.ts
• src/ui/panels/arbor.ts
• src/ui/panels/collab.ts
• src/ui/panels/common.ts
• src/ui/panels/echo.ts
• src/ui/panels/index.ts
• src/ui/panels/language.ts
• src/ui/panels/refinery.ts
• src/ui/touch.ts
• src/ui/widgets.ts
• src/world/buildings.ts
• src/world/city.ts
• src/world/game.ts
• tests/frontier.test.ts
• tests/training.test.ts