cs249r_book/labs/LAB_DEEP_MIGRATION_CHECKLIST.md

Lab Deep Migration Checklist

This checklist tracks the remaining work after the catalog-wide baseline migration. The baseline state is complete: every catalog lab has a track plan, four canonical track variants, and a notebook-level track/report surface. The remaining work is the deeper pass where each lab's actual parts, plots, numbers, reports, and reflections become track-specific.

Canonical tracks:

• Mobile: iPhone
• Tiny: Oura Ring
• Edge: RoboTaxi
• Cloud: Cloud Fleet

Source-of-truth rule:

• Do not add hardware, model, system, cost, memory, latency, energy, or fleet facts as notebook-local constants unless they are temporary UI labels.
• If a fact is computed or displayed as evidence, put it in MLSysIM or a shared mlsysbook_labs registry/helper first.
• Notebook code may compose facts, run solvers, render plots, and collect student decisions, but the underlying reference data should be shared.

Global Completion Gates

• Worktree exists at /Users/VJ/GitHub/MLSysBook-labs on branch codex/labs.
• Canonical track registry exists in mlsysbook_labs.
• Oura Ring and RoboTaxi hardware entries exist in MLSysIM.
• Every catalog lab has a *.track-plan.md file.
• Every catalog lab has baseline variants for all four tracks.
• Every catalog lab has a notebook-level track/report surface.
• Static tests guard track plan coverage and track/report surface coverage.
• Define shared plot/modality catalog for reusable lab evidence displays.
• Define shared per-part section helpers for:
  • what students need to know,
  • prediction lock,
  • track scenario,
  • computed evidence,
  • checkpoint reflection,
  • source trace,
  • report payload.
• Add shared report schema checks for required fields across all deep-migrated labs.
• Add shared tests that every deep-migrated lab's variants resolve all referenced hardware, model, system, and infrastructure refs.
• Add browser-facing smoke coverage for at least one representative lab per track.
• Keep labs/LAB_IMPLEMENTATION_NOTES.md updated after every lab slice.

Per-Lab Definition Of Done

For each lab:

• Read the current notebook and the matching *.track-plan.md.
• Identify the core pedagogical spine: the one decision students should learn to make.
• Decide which parts are invariant across tracks and which parts should change.
• Replace baseline variant text with lab-specific variants where needed.
• Add or reuse source-of-truth facts in MLSysIM or mlsysbook_labs.
• Add or reuse solver/helper APIs for computed evidence.
• Wire selected track into the notebook's parts, plots, tables, and narrative.
• Ensure every part has a consistent internal structure:
  • Objective
  • What you need to know
  • Prediction or design choice
  • Track-specific scenario
  • Computed evidence
  • Checkpoint reflection
• Ensure the synthesis/report includes:
  • selected track,
  • predictions,
  • computed evidence summary,
  • final decision,
  • reflection,
  • residual risk,
  • source trace.
• Add or update tests for any shared helper, solver, registry, or report contract.
• Run focused tests plus static checks.
• Commit the lab or coherent small batch.
• Mark checklist items complete.

Recommended Implementation Order

• 1. V2-11 Edge Intelligence: closest to device-specific constraints.
• 2. V1-11 Hardware Roofline: pressure-tests hardware registry and roofline plots.
• 3. V2-10 Inference Economy: connects latency, cost, batching, and deployment target.
• 4. V1-12 Benchmarking Trap: standardizes benchmark plots and report evidence.
• 5. V1-13 Tail Latency Trap: extends serving and SLA evidence.
• 6. V1-14 Silent Degradation: adds monitoring and drift evidence.
• 7. V2-06 Collective Communication: upgrade existing rich wrapper to canonical tracks.
• 8. Finish remaining Volume I labs by dependency order.
• 9. Finish remaining Volume II labs by dependency order.
• 10. Run catalog-wide QA and final cleanup.

Volume I Labs

V1-00 - The Architect's Portal

Path: labs/vol1/lab_00_introduction.py

Current status:

• Deep orientation structure exists.
• Track selection exists.
• Local report export exists.
• Track profile references come from mlsysbook_labs.

Remaining tasks:

• Replace any remaining notebook-local track display facts with profile-derived fields.
• Confirm the track picker writes a ledger value consumed by all later labs.
• Add a regression test that Lab 00 exposes all four canonical track IDs.
• Confirm the downloaded report includes track ID, hardware ref, system ref, and source policy.
• Update the track-plan file if Lab 00 becomes the canonical place for student track selection.

V1-01 - The AI Triad

Path: labs/vol1/lab_01_ml_intro.py

Current status:

• Baseline track/report panel installed.

Deep migration tasks:

• Map Data, Algorithm, and Machine axes to each canonical track.
• Add track-specific D-A-M examples:
  • iPhone: on-device vision or text classification under memory and latency limits.
  • Oura Ring: always-on sensing under energy and SRAM/flash limits.
  • RoboTaxi: perception stack under safety and real-time latency limits.
  • Cloud Fleet: hosted model service under throughput, cost, and reliability limits.
• Move any displayed device/model facts to shared registries.
• Replace generic baseline scenario text with per-track narrative.
• Add a small track-specific constraint table.
• Update report evidence to include the selected D-A-M bottleneck.
• Add tests for any new helper that summarizes track constraints.

V1-02 - Physics of Deployment

Path: labs/vol1/lab_02_ml_systems.py

Current status:

• Baseline track/report panel installed.

Deep migration tasks:

• Tie the deployment physics lesson to track-specific resource budgets.
• Add a source-of-truth helper for deployment envelope summaries if one does not exist.
• Compute or display memory, latency, energy, or cost only through shared APIs.
• Give each track a different binding constraint:
  • iPhone: thermal/latency envelope.
  • Oura Ring: battery and memory envelope.
  • RoboTaxi: deterministic latency and sensor throughput.
  • Cloud Fleet: cost, utilization, and failure domain.
• Update parts so the same conceptual lesson produces different track conclusions.
• Add report fields for binding physical constraint and mitigation.

V1-03 - Constraint Tax

Path: labs/vol1/lab_03_ml_workflow.py

Current status:

• Baseline track/report panel installed.

Deep migration tasks:

• Identify where the workflow creates hidden constraint tax for each track.
• Add shared workflow-stage descriptors if they are reused by later labs.
• Add track-specific examples for data collection, validation, deployment, and monitoring.
• Make the final decision differ by track rather than only by generic workflow stage.
• Ensure the report captures the most expensive workflow constraint and the mitigation plan.
• Add tests for any shared workflow or ledger serialization helper.

V1-04 - Data Gravity

Path: labs/vol1/lab_04_data_engr.py

Current status:

• Baseline track/report panel installed.

Deep migration tasks:

• Define track-specific data source, data rate, retention, and privacy assumptions.
• Move reusable data-rate or storage assumptions to shared registries/helpers.
• Add evidence for where data should be processed:
  • iPhone: local preprocessing vs upload.
  • Oura Ring: summary features vs raw sensor streams.
  • RoboTaxi: local sensor fusion vs fleet upload.
  • Cloud Fleet: warehouse/lake/feature-store placement.
• Update plots/tables to show storage, bandwidth, or freshness tradeoffs.
• Update report evidence with selected data placement and residual data risk.

V1-05 - Activation Tax

Path: labs/vol1/lab_05_nn_compute.py

Current status:

• Baseline track/report panel installed.

Deep migration tasks:

• Tie neural computation concepts to per-track activation memory and compute budgets.
• Add shared helper for activation footprint if not already available.
• Use model refs from the shared variant registry.
• Show how batch size, precision, and activation shape affect each track.
• Add track-specific failure state:
  • iPhone: thermal or memory pressure.
  • Oura Ring: SRAM/energy overflow.
  • RoboTaxi: latency miss.
  • Cloud Fleet: utilization/cost miss.
• Include activation-footprint evidence in the report.

V1-06 - Architecture Tax

Path: labs/vol1/lab_06_nn_arch.py

Current status:

• Deep track-aware architecture migration installed.

Deep migration tasks:

• Connect architecture choices to device and service constraints.
• Add or reuse model architecture descriptors in mlsysbook_labs.
• Compare track-appropriate model families instead of one generic model list.
• Show why a model that is accurate in one track fails in another.
• Add report fields for architecture choice, rejected alternatives, and dominant constraint.
• Add tests for model-family registry lookups if introduced.

V1-07 - Framework Tax

Path: labs/vol1/lab_07_ml_frameworks.py

Current status:

• Deep track-aware framework/runtime migration installed.

Deep migration tasks:

• Make framework/runtime choice track-specific.
• Add shared runtime/deployment-target catalog if needed.
• Track examples:
  • iPhone: Core ML style deployment and operator support.
  • Oura Ring: TFLite Micro or MCU-oriented runtime.
  • RoboTaxi: TensorRT/accelerated edge runtime.
  • Cloud Fleet: server runtime with batching and observability.
• Display operator coverage, packaging, and portability tradeoffs from shared facts.
• Update report with selected runtime and deployment risk.

V1-08 - Training Gauntlet

Path: labs/vol1/lab_08_model_train.py

Current status:

• Deep track-aware training migration installed.

Deep migration tasks:

• Separate training environment from deployment environment for each track.
• Add shared training budget/cost helper if needed.
• Track examples:
  • iPhone: train centrally, personalize lightly on device if appropriate.
  • Oura Ring: train centrally, deploy tiny model, maybe adapt thresholds.
  • RoboTaxi: simulation/fleet retraining loop.
  • Cloud Fleet: large-scale distributed training and evaluation.
• Show compute, data, and evaluation bottlenecks by track.
• Update report with training strategy and deployment handoff risk.

V1-09 - Selection Paradox

Path: labs/vol1/lab_09_data_selection.py

Current status:

• Deep track-aware data-selection migration installed.

Deep migration tasks:

• Make data/model selection criteria depend on track constraints.
• Add shared candidate-selection helper if useful across labs.
• Include track-specific tradeoffs between accuracy, robustness, latency, memory, and cost.
• Ensure plots compare candidates with track-specific feasibility boundaries.
• Update report with selected candidate, rejected candidate, and reason.
• Add tests for candidate feasibility logic if introduced.

V1-10 - Compression Paradox

Path: labs/vol1/lab_10_model_compress.py

Current status:

• Deep pilot migration exists.
• Compression candidate solver exists in MLSysIM.
• Track-specific variants exist.
• Report synthesis includes computed evidence and reflection fields.

Remaining tasks:

• Audit device precision facts, especially iPhone int8 support, and decide whether to add missing hardware capabilities to MLSysIM.
• Confirm all four track variants have hand-authored guardrails, validation tests, and residual risks.
• Add a visual regression or data-shape test for the compression evidence table if useful.
• Use V1-10 as the reference pattern for later deep migrations.

V1-11 - Hardware Roofline

Path: labs/vol1/lab_11_hw_accel.py

Current status:

• Deep track-aware notebook migration installed.
• Track selector, track context, source trace, and local report export installed.
• Shared roofline helper backs GEMM, fusion, and roofline evidence.
• All displayed hardware roofline facts resolve through canonical track variants and MLSysIM refs.

Deep migration tasks:

• Build track-specific roofline evidence using hardware registry facts.
• Confirm no missing hardware capabilities are needed for this slice:
  • peak ops,
  • memory bandwidth,
  • memory capacity,
  • accelerator type,
  • supported precision.
• Add shared roofline helper if current calculations are notebook-local.
• Track examples:
  • iPhone: Neural Engine/GPU/CPU boundary.
  • Oura Ring: MCU/DSP-style tiny compute boundary.
  • RoboTaxi: edge accelerator throughput and deterministic latency.
  • Cloud Fleet: GPU roofline and utilization.
• Update report with compute-bound vs memory-bound diagnosis.
• Add tests for roofline helper and hardware ref resolution.

V1-12 - Benchmarking Trap

Path: labs/vol1/lab_12_perf_bench.py

Current status:

• Deep track-aware notebook migration installed.
• Track selector, track context, source trace, and local report export installed.
• Shared benchmarking helper backs Amdahl, sustained benchmark, multi-metric, and tail-latency evidence.

Deep migration tasks:

• Standardize benchmark modalities for latency, throughput, memory, energy, and cost.
• Add shared benchmark-result schema if needed.
• Give each track a different "bad benchmark" trap.
• Add track-specific benchmark plots and failure states.
• Ensure report captures benchmark setup, misleading metric, corrected metric, and conclusion.
• Add tests for benchmark schema/report serialization.

V1-13 - Tail Latency Trap

Path: labs/vol1/lab_13_model_serving.py

Current status:

• Baseline track/report panel installed.

Deep migration tasks:

• Tie serving choices to track-specific SLA and request pattern.
• Add shared latency distribution helper if current logic is notebook-local.
• Track examples:
  • iPhone: interactive on-device response.
  • Oura Ring: periodic background inference.
  • RoboTaxi: hard real-time perception budget.
  • Cloud Fleet: p95/p99 service SLO.
• Display tail latency distributions and mitigation options by track.
• Update report with SLA, tail driver, mitigation, and residual risk.
• Add tests for latency helper/report fields.

V1-14 - Silent Degradation

Path: labs/vol1/lab_14_ml_ops.py

Current status:

• Baseline track/report panel installed.

Deep migration tasks:

• Make monitoring signals and remediation options track-specific.
• Add shared drift/degradation scenario catalog if useful.
• Track examples:
  • iPhone: app version/device OS drift.
  • Oura Ring: sensor placement or physiology drift.
  • RoboTaxi: weather/geography drift.
  • Cloud Fleet: traffic mix or upstream data drift.
• Add plots for metric drift, alert thresholds, and false alarms.
• Update report with monitoring plan, trigger, action, and residual risk.
• Add tests for operations helper/report fields.

V1-15 - No Free Fairness

Path: labs/vol1/lab_15_responsible_engr.py

Current status:

• Deep track-aware responsibility structure installed.
• Track selector, source trace, ledger save, and local report export installed.

Deep migration tasks:

• Make fairness/responsibility tradeoffs specific to each deployment context.
• Add shared metric-policy helper if the lab needs comparable definitions.
• Tie track scenarios to realistic stakeholders and harm models.
• Ensure any group metric examples come from a shared synthetic dataset/helper.
• Update report with selected metric, tradeoff, mitigation, and residual risk.
• Add tests for report fields and synthetic dataset shape if introduced.

V1-16 - The Architect's Audit

Path: labs/vol1/lab_16_ml_conclusion.py

Current status:

• Deep track-aware capstone structure installed.
• Track selector, source trace, ledger replay, sensitivity audit, memo report export installed.

Deep migration tasks:

• Make the capstone synthesize the selected track across prior Volume I decisions.
• Add ledger reader/helper if cross-lab decisions need normalized access.
• Track examples should summarize the student's chosen device/service constraints.
• Add final architecture report fields that reference prior lab evidence.
• Confirm missing ledger entries degrade gracefully.
• Add tests for any ledger summary helper.

Volume II Labs

V2-01 - The Scale Illusion

Path: labs/vol2/lab_01_introduction.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Reframe scale as track-specific rather than only cloud-scale.
• Add shared scale-envelope helper if reused.
• Track examples:
  • iPhone: millions of installed devices.
  • Oura Ring: always-on fleet telemetry with tiny payloads.
  • RoboTaxi: city-scale autonomous fleet operations.
  • Cloud Fleet: GPU/service fleet scaling.
• Update parts so scale changes reliability, coordination, and cost differently by track.
• Update report with the selected scale failure mode.

V2-02 - The Compute Infrastructure Wall

Path: labs/vol2/lab_02_compute_infra.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Tie compute infrastructure limits to track-specific hardware/system refs.
• Add missing infrastructure facts to MLSysIM if displayed.
• Compare compute, bandwidth, memory, and utilization limits by track.
• Show where adding more compute stops helping.
• Update report with infrastructure bottleneck and mitigation.
• Add tests for any shared compute infrastructure helper.

V2-03 - Network Fabric Design

Path: labs/vol2/lab_03_communication.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Make communication fabric meaningful for every track.
• Add or reuse network/fabric refs in MLSysIM.
• Track examples:
  • iPhone: device-to-cloud uplink constraints.
  • Oura Ring: BLE/mobile relay payload constraints.
  • RoboTaxi: vehicle-edge-cloud synchronization.
  • Cloud Fleet: east-west datacenter fabric.
• Display bandwidth, latency, payload, and retry tradeoffs.
• Update report with selected communication strategy and residual risk.

V2-04 - The Data Pipeline Wall

Path: labs/vol2/lab_04_data_storage.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Connect data storage and freshness to each track.
• Add shared pipeline/storage budget helper if needed.
• Track examples:
  • iPhone: privacy-preserving local cache and upload policy.
  • Oura Ring: compressed time-series summaries.
  • RoboTaxi: high-volume sensor logs and incident upload.
  • Cloud Fleet: feature store and training data lake.
• Add plots/tables for storage growth, freshness, and bandwidth.
• Update report with data pipeline decision and residual risk.

V2-05 - The Parallelism Puzzle

Path: labs/vol2/lab_05_dist_train.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Make parallelism choice relevant to track deployment lifecycle.
• Distinguish training-time parallelism from inference-time deployment target.
• Add shared parallelism strategy descriptors if useful.
• Track examples:
  • iPhone/Oura Ring/RoboTaxi: centralized or federated training feeding edge deployment.
  • Cloud Fleet: data/model/pipeline parallelism in service of large model training.
• Update report with training strategy, scaling bottleneck, and residual risk.
• Add tests for any shared parallelism helper.

V2-06 - Collective Communication

Path: labs/vol2/lab_06_collective_communication.py

Current status:

• Rich shared wrapper exists.
• Baseline track/report panel installed.
• MLSysIM collective communication physics functions are already used.

Deep migration tasks:

• Upgrade the existing rich wrapper to canonical track variants.
• Decide which tracks use true collectives and which use communication analogs.
• Track examples:
  • iPhone: federated update aggregation payloads.
  • Oura Ring: tiny update/sensor summaries through a phone relay.
  • RoboTaxi: fleet update synchronization and map/model rollout.
  • Cloud Fleet: GPU collective algorithms.
• Add source-of-truth communication payload assumptions if displayed.
• Ensure the report uses canonical track profile and variant fields.
• Add tests for track-specific communication scenario selection.

V2-07 - When Failure Is Routine

Path: labs/vol2/lab_07_fault_tolerance.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Make failure modes and recovery decisions track-specific.
• Add shared reliability/failure-budget helper if needed.
• Track examples:
  • iPhone: offline mode, app crashes, OS updates.
  • Oura Ring: battery depletion, sensor dropout, sync gaps.
  • RoboTaxi: sensor faults, degraded mode, safety fallback.
  • Cloud Fleet: node failures and retry storms.
• Display failure-rate, recovery-time, or availability evidence.
• Update report with failure budget and recovery decision.

V2-08 - The Scheduling Trap

Path: labs/vol2/lab_08_fleet_orch.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Connect scheduling/orchestration to each track's fleet shape.
• Add shared fleet/orchestration descriptors if needed.
• Track examples:
  • iPhone: staged rollout and device eligibility.
  • Oura Ring: background job timing and battery-aware scheduling.
  • RoboTaxi: vehicle dispatch, update windows, and safety constraints.
  • Cloud Fleet: GPU scheduling and bin packing.
• Display utilization, queueing, rollout, or availability tradeoffs.
• Update report with scheduling policy and residual risk.

V2-09 - The Optimization Trap

Path: labs/vol2/lab_09_perf_engineering.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Make optimization targets differ by track.
• Add shared performance-counter or bottleneck taxonomy helper if useful.
• Track examples:
  • iPhone: optimize latency without thermal regression.
  • Oura Ring: optimize energy without missing sensing events.
  • RoboTaxi: optimize latency while preserving safety margin.
  • Cloud Fleet: optimize throughput/cost without p99 regression.
• Update plots to show local optimum vs system optimum.
• Update report with chosen optimization and unintended side effect.

V2-10 - The Inference Economy

Path: labs/vol2/lab_10_inference.py

Current status:

• Deep track-aware notebook migration installed.
• Track selector, track context, source trace, and local report export installed.
• Shared inference-economy helper backs cost crossover, state/cache capacity, batching, and serving-plan evidence.

Deep migration tasks:

• Build track-specific inference economics.
• Add shared inference cost/latency helper if current logic is notebook-local.
• Track examples:
  • iPhone: local inference cost is battery/thermal/UX.
  • Oura Ring: local inference cost is energy and duty cycle.
  • RoboTaxi: local inference cost is latency and safety margin.
  • Cloud Fleet: service inference cost is dollars, utilization, and p99.
• Display batching, quantization, caching, or placement tradeoffs by track.
• Update report with inference placement and economic constraint.
• Add tests for cost/latency helper.

V2-11 - The Edge Thermodynamics Lab

Path: labs/vol2/lab_11_edge_intelligence.py

Current status:

• Hand-authored variants exist.
• Baseline track/report panel installed.
• Deep track-aware notebook migration installed.

Deep migration tasks:

• Make this the first full device-track deep migration after V1-10.
• Move any displayed device battery, memory, latency, or energy facts into MLSysIM/shared helpers.
• Add a shared edge energy or duty-cycle helper if needed.
• Wire each notebook part to the selected canonical track.
• Track examples:
  • iPhone: on-device adaptation under thermal and battery limits.
  • Oura Ring: always-on sensing and tiny inference energy budget.
  • RoboTaxi: edge perception and vehicle compute envelope.
  • Cloud Fleet: compare edge offload against centralized inference.
• Add track-specific plots for memory, battery/energy, and update payloads.
• Update report with edge placement decision and thermodynamic residual risk.
• Add tests for helper math and variant field completeness.

V2-12 - The Silent Fleet

Path: labs/vol2/lab_12_ops_scale.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Make fleet observability and operations track-specific.
• Add shared fleet health/telemetry helper if needed.
• Track examples:
  • iPhone: app/device version segments.
  • Oura Ring: sensor quality and sync coverage.
  • RoboTaxi: route/geography/weather fleet slices.
  • Cloud Fleet: service shards, regions, and model versions.
• Display fleet health dashboard evidence by track.
• Update report with monitoring slice, action threshold, and residual risk.

V2-13 - The Price of Privacy

Path: labs/vol2/lab_13_security_privacy.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Make security/privacy controls track-specific.
• Add shared privacy/security control catalog if needed.
• Track examples:
  • iPhone: local processing and permission boundary.
  • Oura Ring: health-derived sensor data and consent.
  • RoboTaxi: location/video logs and incident retention.
  • Cloud Fleet: tenant isolation and data governance.
• Display accuracy, latency, cost, or utility impact of privacy controls.
• Update report with selected privacy control and residual risk.

V2-14 - The Robustness Budget

Path: labs/vol2/lab_14_robust_ai.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Tie robustness budget to track-specific failure consequences.
• Add shared robustness scenario/helper if needed.
• Track examples:
  • iPhone: varied lighting/device/user context.
  • Oura Ring: sensor noise and physiological variation.
  • RoboTaxi: weather, occlusion, and out-of-distribution scenes.
  • Cloud Fleet: prompt/user/data distribution shift.
• Display robustness-cost tradeoffs.
• Update report with robustness budget and unresolved hazard.

V2-15 - The Carbon Budget

Path: labs/vol2/lab_15_sustainable_ai.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Make sustainability accounting track-specific.
• Add shared energy/carbon helper if the lab displays computed carbon evidence.
• Track examples:
  • iPhone: battery energy and charging externality.
  • Oura Ring: tiny battery and lifecycle duty cycle.
  • RoboTaxi: vehicle compute energy and fleet update cadence.
  • Cloud Fleet: datacenter energy, utilization, and region mix.
• Display energy/carbon tradeoffs with source trace.
• Update report with carbon budget decision and residual risk.

V2-16 - The Fairness Budget

Path: labs/vol2/lab_16_responsible_ai.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, and local report export installed.

Deep migration tasks:

• Make fairness tradeoffs specific to each deployment track.
• Reuse or extend the fairness metric helper from V1-15 if created.
• Track examples:
  • iPhone: accessibility and device/user variation.
  • Oura Ring: physiological and demographic variation.
  • RoboTaxi: neighborhood, pedestrian, weather, and safety exposure.
  • Cloud Fleet: service quality across user cohorts/regions.
• Display fairness/utility/safety tradeoffs.
• Update report with fairness budget, mitigation, and residual risk.

V2-17 - The Fleet Synthesis

Path: labs/vol2/lab_17_fleet_synthesis.py

Current status:

• Deep shared system-design renderer installed.
• Typed system-design variants cover all canonical tracks.
• Track selector, source trace, decision frontier, scaling curve, reflection, ledger save, prior Volume II decision summary, and local report export installed.

Deep migration tasks:

• Make this the Volume II capstone for the selected canonical track.
• Add or reuse ledger summary helper to gather prior Volume II decisions.
• Track examples should synthesize scale, infrastructure, communication, failure, operations, privacy, robustness, sustainability, and responsibility.
• Confirm missing ledger entries degrade gracefully.
• Generate a final fleet architecture report.
• Add tests for fleet synthesis report schema and ledger fallback.

Final Catalog QA

• Run python3 -m pytest labs/tests/test_static.py -q.
• Run all shared helper tests in labs/tests.
• Run relevant MLSysIM tests separately from labs/tests.
• Verify wheels/mlsysbook_labs-0.1.0-py3-none-any.whl is rebuilt after shared helper changes.
• Verify wheels/mlsysim-0.1.2-py3-none-any.whl is rebuilt after MLSysIM changes.
• Verify every lab renders in a browser smoke pass.
• Verify every lab's report can be generated locally.
• Verify each track has at least one representative visual/evidence modality.
• Verify no lab embeds source-of-truth hardware/model/system facts locally.
• Commit final cleanup.