Eve library

V2 contracts · source interpretation and evidence

Guide

Revision 1

On this page

Audience: a product or engineering reader who has read V2 product promise and surfaces and needs to trace a documented claim into its actual artifact and checker. Source pin: 1ceae52c7896a59ac74a4429d410f3218f0b91c6.

Visual thesis: a quiet cream-and-ink reading plane, with terracotta emphasis on the evidence boundary; aligned comparisons distinguish independent facts and ordered traces explain real review or generation steps.

Content plan: establish the claim-to-evidence path, resolve references and artifact kinds, examine the floating-point contract across targets and cook routes, then inspect generated descriptors and a reproduced stale-output case. Close by asking for a scoped, reproducible result, not a release certificate.

Interaction thesis: retain slide entrance, source disclosure and sidebar/drawer transitions, with reduced motion. Every relationship must remain understandable in static HTML and PDF. No decorative imagery, new motion or fake interface is needed for this source-reading guide.

Slide-level reading plan#

Each row gives one communication job and takeaway. Tables read by named row, left to right; independent columns have no ordinals. Flows read in their stated order. Titles carry the dominant claim; exact identifiers and worked details stay in aligned rows, notes and pinned source excerpts.

Slide suffix Job, relationship and evidence anchor Emphasis and transition
evidence-path Follow intent → ownership → contract → verification in the architecture convention. Traceability starts an investigation; resolve its notation next.
reference-language Compare four reference forms and their actual document destinations, including named examples. Resolve document before heading; inspect the gate.
doc-validation Compare numbering, links, decision markers and package-name rules in the Python validator. Accepted documentation is not a running package; test the limit.
doc-check-limits Follow an illustrative missing-child reference through parent acceptance and exact-heading review. A green check can leave a missing target; classify the artifact.
target-artifacts Follow named target → disk inspection → owner → qualified evidence. A named path can still be planned; distinguish contract kinds.
contract-kinds Compare an executable Arete Zod value schema with the V2 floating-point JSON and its Python checker. Neither existence nor a local pass establishes consumer integration; reconcile the tree.
module-reconciliation Compare glossary-only names, actual plugins, extra Source modules and the two V2Editor meanings. Vocabulary is not installation; inspect target policy.
compiler-policy Compare platform flags, definitions and build-environment override in the three target classes. Configuration precedes execution evidence; inspect target differences.
target-differences Compare game, server and editor constructors and their duplicated helpers. No explicit editor unity override; trace definitions to consumers.
reserved-definitions Compare declaration, consumption and demonstration as independent evidence questions. Reserved symbols are not active runtime branches; open the JSON.
determinism-identity Map schema/task, targets, definitions and target rules to their inspection role. Requirements are not results; follow delivery fields.
determinism-delivery Map cook, architecture and automation fields to the same contract's delivery boundary. A named route needs implementation review; compare cache routes.
reference-ddc Compare Windows environment, explicit Cook, BuildCookRun and direct wrappers. No false uniform path; absent argument does not prove effective cache.
checker-boundary Compare selected JSON values, cross-file snippets and separate runtime evidence without ordinal markers. A checker supports only its assertions; inspect generation.
grpc-generation Follow proto discovery, supported-signature extraction, generation and read-only freshness checking. Metadata is not transport; examine descriptor fields.
grpc-descriptors Compare method/service fields, lookup and regex parser scope. Options blocks can hide later methods; inspect a real stale baseline.
grpc-freshness-case Compare pinned stored output, isolated regeneration and exact additional services. 44/666 versus 47/675 is a dated mismatch, not release growth; conclude narrowly.
evidence-conclusion Follow inventory → trace → scoped conclusion as an ordered review exercise. Preserve revision, target, environment, result and limitations; proceed to deeper combat/build guides.

The saved sixteen-slide draft remains a historical proposal. Integration adds two substantive comparisons, removes misleading ordinals from two independent comparisons, and qualifies the gRPC interpretation with executed source probes. Eighteen slides follow the learning task, not a count target.

Enlarged-text inspection shortened the freshness-case row labels to “Local output” and “Check result”; the subtitle and notes preserve isolated regeneration. The contract-kind table uses “Executable Zod enum” while notes and the source panel retain CheckInStatusSchema exactly. These changes remove split prose words and a one-letter code-name fragment without shrinking type or changing narration. Word-range regressions cover the two new tables' prose row labels, not every exact code identifier in the library.

The subsequent complete mobile review found fragmented anchor notation, compilerDefinitions, BuildCookRun and a reference-cache path. Reference language, determinism identity and DDC therefore opt into the validated sourceWrap table treatment: native optional breaks at identifier boundaries, a 132–146px mobile row column and 14px cell-right spacing. Each value column still fits beside its row identity. Every source character, note and speech plan remains unchanged. A 20px stress case caught Definitions before the column/spacing allowance was adjusted. An attempted 180px minimum value column failed sticky-identity visibility at 320px and was rejected, not shipped. The browser audit checks source-word geometry and rejects removal of the optional boundaries; PDF checks retain non-whitespace characters and cell order. Other guide HTML is not restyled by this opt-in.

Viewing and acceptance#

Verify all eighteen slides at desktop, compact desktop, mobile initial and full reading sizes, all mobile table columns, embedded desktop/mobile and PDF page fit. Exact-source identifiers may wrap without losing characters; ordinary labels should stay readable without shrinking type. Keep the selected route, source pin and scope qualifications legible. Recheck compact active narration after recording, not only static fit.

This is not an accepted visual benchmark or whole-library migration. Automated checks and model inspection are not willing-reader evidence. Final normal-speed listening and reader review remain required. Use the existing reader protocol, bind final stimuli, and ask without coaching:

  1. A file named contract exists. What would you inspect to establish what it validates and whether a consumer actually uses it?
  2. The floating-point checker passes. What can you conclude about the editor's unity setting and a successful deterministic simulation?
  3. Freshness passes after an RPC options block is added. Could methods still be missing, and what evidence would resolve that question?
  4. The stored header reports fewer services than isolated regeneration. What has this experiment proved, and what has it not proved?

Record unaided answers and accommodations before assessment. Do not invent participant results, comprehension scores or live service/build evidence.

Eight-standard quality ledger · not final acceptance#

Standard Current evidence and remaining acceptance
Concrete benchmark Uses the established Oshun workspace, not a newly accepted reference. The broader representative pilot and willing-reader calibration remain open.
Typography and optical alignment Two new comparisons were inspected at 320px and with 20px body text at 390px. Concise display labels retain exact identifiers in notes. Browser checks enforce a 16px body floor; full font-fallback, page zoom and reader-validated type rules remain open.
Quantitative comparison The stored 44/666 and isolated 47/675 counts share one source pin and generator. The exact three-service/nine-RPC delta is tested against the complete pinned proto archive. This is a dated freshness comparison, not a growth trend, performance measurement or release score.
Visual continuity The same claim → artifact → checker → bounded result vocabulary carries through references, floating-point policy and descriptors. Independent comparisons are unnumbered; actual investigation and generation sequences retain order. The guide is not a complete system-wide entity representation.
Image art direction No photo, generated image or mock UI is justified by these source-reading tasks. Native comparisons and flows carry the evidence; exact sources remain selectable in Explain. Future actual game, editor or operations UI guides need their own screenshot/provenance review.
Actual-size readability Review all eighteen final slides at desktop, compact, mobile, embedded and PDF page fit, plus every compact active-playback state. Narrow/enlarged captures expose content below the first phone viewport; they do not prove initial-viewport fit. Final version-bound delivery evidence belongs in the verification receipt.
Reader comprehension The four neutral tasks above are prepared; no participant observations or comprehension pass exists. Apply the shared agreement, stimulus-freeze and unaided-then-supported protocol.
Editorial subtraction Removed misleading ordinals from independent comparisons and shortened two row labels plus the visible schema identifier after actual-size inspection. Exact facts and speech are preserved. All eighteen notes and transitions were reread; final normal-speed listening remains separate and incomplete.

Assess the four reader tasks only after retaining the unaided answer. Expected boundaries are: a caller must invoke the right validator and its consumer still needs evidence; a source-check pass neither supplies an absent editor unity override nor proves simulation; the regex parser can omit both an options-body method and a later ordinary method while freshness passes; and isolated regeneration establishes a dated metadata mismatch, not network availability or Unreal build success. Equivalent accurate explanations are acceptable. These are source-backed assessment criteria, not coached answers or observed results.