Skip to content

Connected Mini-Projects

This book is designed to produce working judgment, not recognition. The practical track connects 36 mini-projects to the same companion workspace. You repeatedly extend one system from strict Rust domain types to a CLI, service, multimodal evidence pipeline, AI harness, evaluation system, model router, fine-tuning workflow, and production incident drill.

The mini-projects are not detached demos. Their types, fixtures, receipts, and failure investigations feed Forge, Signal Room, Cutroom, and the final Mosaic platform.

predict ─► trace ─► implement ─► break ─► diagnose ─► improve ─► explain
▲ │
└──────────── rerun the same acceptance evidence ────────────────┘
  1. Predict the accepted input, state transition, bounded resource, expected trace, and failing mutation before running the code.
  2. Trace one input through types, transformations, effects, receipts, and terminal state.
  3. Implement a narrow behavior that requires an engineering decision.
  4. Break it with a supplied malformed fixture, failpoint, timeout, stale generation, unauthorized effect, or adversarial input.
  5. Diagnose the first violated invariant rather than the last visible symptom.
  6. Improve one responsible layer and rerun unchanged cases.
  7. Explain the evidence, rejected alternative, limitation, and recovery path.
ProjectBuildWhat survives into later work
MP-01 · Validated Task EnvelopeStrict IDs and task statesEvery run and API boundary
MP-02 · Source Span WorkbenchUnicode-safe normalizationRestorable citations
MP-03 · Runtime Provider RegistryCapability-aware dispatchModel routing
MP-04 · Concurrent Budget LedgerReserve/commit/releaseHarness and worker admission
MP-05 · Isolated Native BoundaryAudited unsafe wrapperLocal inference and media
MP-06 · Forge Command SpineStable CLI behaviorForge interface
MP-07 · Bounded Async WorkerCancellation and backpressureServices and workers
MP-08 · Streaming Upload and ResumeBounded input and ordered eventsMosaic API
MP-09 · Durable Run RecoveryLeases and idempotencyRestart-safe execution
MP-10 · Tenant-Safe Artifact StoreScoped content addressingEvidence and outputs
MP-11 · Observable Service RolloutHealth, drain, rollbackOperations
MP-12 · Causal Inference NotebookToken/tensor/sampling mechanicsHonest model contracts
MP-13 · Inference Capacity PlannerCache/device capacityRouting and admission
MP-14 · Capability-Neutral RouteScripted/local/remote adaptersProvider independence
MP-15 · Four-Modality IntakeText/image/audio/video envelopesSignal Room and Cutroom
MP-16 · Provenance and Locator GraphLineage and source locationsVerifiable evidence
MP-17 · Timeline and GeometryReversible time/region mapsMultimodal reasoning
MP-18 · Retrieval and CitationFusion, diversity, citationsGrounded answers
MP-19 · Adversarial LaboratoryMutations and fail-closed oraclesSecurity regression corpus
MP-20 · Behavior Contract CompilerCompletion evaluationDefinition of done
MP-21 · Context BudgeterVersioned selection manifestControlled model input
MP-22 · Verified Memory Write GateEvidence, consent, conflicts, expirySafe durable memory
MP-23 · Authorized Tool LoopCapabilities, exact approvals, effect receiptsBounded agency
MP-24 · Replayable Model LoopStopping, repair, replayForge harness
MP-25 · Dataset and Split BuilderVersioned cases and splitsEvaluation corpus
MP-26 · Grader Calibration ClinicAgreement and abstentionTrusted grading
MP-27 · Multimodal Evaluation PackModality-specific gradersSignal Room/Cutroom gates
MP-28 · Statistical Release GatePaired slices and intervalsRelease decisions
MP-29 · Decomposed Incident TriageNarrow subproblemsStrong small-model workflow
MP-30 · Generate–Check–RepairIndependent feedback and searchTest-time compute
MP-31 · Difficulty-Aware CascadeCalibrated escalationQuality/cost frontier
MP-32 · Data Governance PipelineRights, consent, contaminationTrainable corpus
MP-33 · Tiny SFT and LoRAAdapter experimentTuning comparison
MP-34 · Preference AuditPreference quality and gatesSafe optimization
MP-35 · Capability Red TeamInjection and exfiltration drillsSecurity release gate
MP-36 · Safe Degradation IncidentOverload and recoveryFinal operational proof

MP-20 through MP-24 are fully packaged labs. Earlier chapters already contain runnable companions and exercises; they will receive the same formal lab packaging as the expansion proceeds. The table distinguishes the committed sequence from what is implemented today.

Every lab names:

  • a stable lab ID and version;
  • prerequisite chapters and labs;
  • files you are expected to change;
  • a focused command and full regression command;
  • normal, boundary, failure, and security fixtures where applicable;
  • exact or environment-sensitive expected evidence;
  • supplied mutations and a root-cause investigation;
  • a design-review question and completion rubric.

Your learner record should include the source revision, acceptance result, artifact or trace identity, diagnosed mutation, measurement environment, design choice, rejected alternative, known limitation, and the next dependency unlocked.

LevelTypical workProof
1 · RecallClassify a boundary or failureCorrect vocabulary
2 · PredictTrace ownership, state, or costAccurate pre-run prediction
3 · RepairExtend code or fix a mutationFocused regression
4 · BuildComplete and measure a mini-projectArtifact plus receipts
5 · DefendReview a threat model or architectureEvidence-backed design memo

Major chapters cover levels 1–3. Each chapter cluster culminates in levels 4–5. Solutions explain the invariant and test strategy without turning the exercise into transcription.

Begin with MP-20 · Behavior Contract Compiler when you reach Harness Engineering. It is intentionally provider-free: the first practical lesson is that a model saying “done” is an observation, not proof of completion.