The 25-guide, 204-slide Eve route received a complete sequence review. The earlier seven-specimen repair is preserved in its historical receipt; the remaining twelve-guide defects listed there are addressed by this pass. The per-slide ledger records every retain/change decision and binds it to the expanded slide content.
This is model editorial and optical review, supported by actual browser/PDF evidence. It does not claim willing-reader acceptance or completion of the separate whole-library section 10 program. The route keeps product vocabulary and evidence before work, decisions before delivery, specialized surfaces before practice, then architectural explanation and its dated runtime companion.
| Guide | Sequence decision and resulting treatment |
|---|---|
| What you can do with Eve | Open with the Enrollment investigation, then distinguish feature evidence, work and walkthrough progress. The capability map now answers an actual reader task before introducing the chapter route. |
| Read and explore the product graph | Containment establishes domain/feature/flow vocabulary before typed incoming links, work projections and the actual Explorer selection. Tara is the meditation domain; the selected Enrollment id stays consistent. |
| Analyze coverage, impact and backlog | Direct coverage and the uncovered sibling precede waivers, detail queries and freshness. Playback is mapped and Enrollment remains unverified. The versioned traceability case separates structure from missing runtime evidence. |
| Capture and manage work | The concrete task payload precedes kinds, preparation, leasing and review. The actual ten-state Work board anchors recognition; all five recovery/exit transitions retain their exact source and destination states. The PDF enlarges the in-review task instead of shrinking the complete board. |
| Discuss, decide and preserve the rationale | Capture discussion before proposing an ADR, accepting it, superseding it and exporting it. The Enrollment ADR exposes options and consequences; the successor points back to the superseded decision. |
| Follow execution, verification and delivery | Inspect evidence and the lease before submission, merge observation and artifact verification. The lease example and separate verification/retry branches replace the long reference-table run; a same-count/different-hash case makes manifest limits concrete. |
| Investigate operator state | Read time scope before failure diagnostics, incident follow-up and serving configuration. The incident specimen distinguishes a recorded response from a live outage; three model configurations are alternatives, with no containment or fallback arrow. |
| Analyze the studio workbenches | Understand registered views before inspecting Tara, related evidence, Isis, readiness and evaluation. A source-shaped concept dossier and executed hypothetical evaluator output break the five-table run. Revisions expose metadata and claim count, not script/claim text. |
| Act on studio records and deliver content briefs | Read current state before proposing, confirming, scheduling and handing off. One Enrollment guide carries through the brief, concept, planned date and delivery account. A saved date and dispatch report remain separate from publication. |
| Walk the features and track member progress | Begin and mark a run before interpreting counts, catalog drift and recovery. The native 5 visited/3 skipped/2 pending chart retains the denominator; an implemented drift notice shows the same declared run. Skips are processed, not successful completion. |
| Understand the shared member capabilities | Establish Lilith as the member identity and active room/account before room-specific reads and confirmed writes. Keep one guide: Tara, Nisaba, Nyx, Arete and Metis share the same scope/effect reading task, with short source registries useful as references. |
| Practice the complete product-to-delivery workflow | Carry tara.example.enrollment/example-enrollment through observation, four owners, explicit counterfactuals, extension and final review. Counterexamples are alternative scenarios, not a fabricated observed history. The final record names unresolved publication/acceptance evidence. |
| Charter and architecture | Define Eve and its evidence boundary before four planes, the request graph and the reading route. Enlarge both architecture graphs without changing nodes or directed relationships; conceptual structure remains distinct from dated implementation. |
| Identity, surfaces and invocation | Move from an invocation request to supported surfaces, context collection, availability and restoration. Retain short comparisons where the reader must choose between entry conditions, and sequences where actual steps follow one another. |
| Context, memory and grounding | Separate current context, memory and grounding before governance, retrieval and continuity. Enlarge the evidence-to-answer graph; both supported claims and insufficient evidence remain visible outcomes. |
| Context, memory and evidence lifecycles | Use a selected passage and reading preference to establish purposes, then compaction, governed recall, deletion and claim support. The closing store map earns its table because deletion rights differ by store. |
| Safety, trust and authority | Introduce origin/authority before the retrieved-page attack, policy order, interruption, grants and retained evidence. Conclude with threat coverage and distinct failure outcomes; no inventory or refusal is promoted to proven protection. |
| Authority, safety and consequential decisions | Resolve precedence before crisis continuation, specific grants and independent resource/consequence checks. The disclosure table precedes the adversarial exercise; explicit emphasis remains only where the authored takeaway warrants it. |
| Governed execution | Distinguish intent, confirmation, execution and acceptance before the changed-target case. The enlarged artifact graph exposes verification/revision loops; the final failure table provides bounded recovery actions. |
| From request to verified result | Trace a brief revision through routing, tool contract, versioned approval, typed result and attributable work. Execution/verifier separation precedes unknown-outcome recovery, so the reader has the evidence needed to reason about retries. |
| Models, quality and cost | Set the task-family quality floor before illustrative cost arithmetic, serving legs and measurement. Retain registry and cost tables for exact comparison; finish with a failed-stage diagnosis before escalation. |
| Model selection, evaluation and change control | Allocate intelligence, diagnose failure, retain release evidence, bound escalation and control model-visible changes before economics. Rename the final category header Cost dimension; its rows are not questions. |
| Operations and acceptance | Start from user-visible health and a lost write reply before degraded dependencies, telemetry, recovery and maintenance. Ownership and charter gaps precede runtime admission, planning quality and completion criteria. |
| Operating and changing Eve safely | Move from dependency health to honest fallback, traceability, containment, operating cadence, change ownership and a release exercise. Tables map genuinely different boundaries; flows describe ordered responses. |
| Current runtime: authority, voice and inspection | Declare the dated code snapshot before authority, isolation, impact, injection measurements, admission, voice and capture. The final draft-revision case integrates the architectural route without implying current deployment from dated code. |
The changed explanations use case studies, source-shaped records, real pinned React UI, a complete recovery map, gated branches and quantitative accounting. Useful registries/comparisons remain tables. All 204 slides remove the repeated masthead; unordered columns have no artificial step numbers. The additional Cost dimension correction closes the last category-only Question header found in the Eve manuscripts. Explicit emphasis on verification/admission remains where the authored brief explains its meaning.
The completion receipt records final automated coverage, optical captures and publication inputs. Reader review continues to distinguish neutral tasks from unperformed human sessions.