Eve library

V2 network-quality teaching assignment

Guide

Revision 1

On this page

This review assigns one complete documentation unit to the existing twelve-slide guide. It does not complete the exhaustive presentation objective. The source is “Network quality, handshake, and graceful fallback,” lines 262–309 of V2/architecture/rollback-netcode-and-tag-team.md, at revision 1ceae52c7896a59ac74a4429d410f3218f0b91c6.

The assignment binds the source-unit hash and all twelve assembled slide hashes, including inherited chapter/release and composition data. The guide already contains the explanations; this change records the full-unit editorial review and updates teaching coverage.

Source review#

The review covers the whole grade table, three config thresholds, wireless wording, consecutive-pressure claim, eight-node flowchart, offer/decline description, score expression, diagnostic fields/endpoint and both follow-up links. The pinned analyzer, config/policy/decision/stamp predicates, rollback session and Blueprint wrappers qualify several architecture claims:

Source topic Explanation in the guide
Handshake and wireless Mean RTT/loss and maximum jitter feed ordered Poor/Borderline/Good branches. RTT/jitter comparisons are strict, loss is inclusive, empty input has its own path, and the dedicated wired recommendation appears in the Good branch. The two-sample worked case returns Borderline/Hybrid with ranked eligibility.
Diagram and mode selection The architecture arrows describe intended behavior. Analyzer results and forwarding wrappers do not establish a session mode change. The inspected session initializes Rollback and exposes a getter; the guide does not infer behavior of all assets or external integrations.
Sustained pressure The helper scans consecutive trailing entries strictly above the configured cap. A newest non-pressure entry interrupts it. 1,800 entries imply 30 seconds only with one entry per 60 Hz frame; sample labels do not measure elapsed time.
Offer, acceptance and validation Empty samples, extreme RTT, sustained pressure and stable mode have separate branches. False acceptance is not a separate observed decline; offered/no-contest flags can coexist. Repeated evaluation carries no once-only offer history. Accepted stable Hybrid can fail the independent decision predicate.
Score Each RTT/jitter/loss ratio is clamped before the 25/25/40 weights are applied; rollback above eight adds ten. Penalties are averaged before subtraction and the final clamp. Extreme RTT alone yields 75 despite a Poor handshake. No account identifier or persistence is supplied by the score function.
Diagnostics and follow-up The match identifier, aggregates, score, grade and export flags form a returned record. Ops.NetworkHealth.V2 is an assigned endpoint name. Neither proves storage or delivery. Independent predicate counterexamples and the investigation slide explain what online-operations/telemetry consumers still need to establish.

All twelve slides contribute to this one unit. Referenced C++ files, generated duplicates, other architecture sections and the follow-up documents do not gain teaching assignments through association.

Verification and inheritance#

Current assignment validation, 16 inventory regressions, eight source/render regressions, builder/preview freshness, repository discovery, workflow coverage and scoped ESLint/Prettier pass. The complete-teaching gate returns one as required: 32 reviewed, 10,018 structural and 103,294 unassigned documentation units remain an incomplete corpus. No workflow assignment is added; 3,718 workflow units remain unassigned. The coverage delta is exactly one newly reviewed unit on this architecture page.

The baseline is the published shared-table repair, c20471d8a60faf0e3fcf3b1b36252533425c3306. Its full 421-slide bank, 40-guide center, typography/zoom/source/print regressions and all-421 audio checks are historical executions retained in the shared-table receipt. Inheritance requires byte equality for the delivered HTML/PDFs, content, renderer/test inputs, catalog, scripts and recordings. This metadata review does not relabel them as newly executed browser or audio checks.

state.json records observed exits, current metadata and unchanged-artifact hashes, the exact source-unit identity and pinned source-file hashes. Retained logs omit PID-prefixed host-process rows and normalize trailing whitespace, with original/retained hashes and transformation counts recorded. This read-only checker validates those bindings against the current checkout:

sh
node docs/presentations/presentation-center/verification/v2-network-quality-teaching-2026-09-07/verify-retained.mjs

For fresh execution, use the prepared presentation Python environment and run the following commands separately under the resource supervisor:

sh
python tools/presentations/run-supervised.py python -m unittest discover -s tools/presentations/tests -p 'test_inventory.py'
python tools/presentations/run-supervised.py python -m unittest discover -s tools/presentations/tests -p 'test_v2_network_quality_draft.py'
python tools/presentations/check-center-coverage.py --check
python tools/presentations/check-center-coverage.py --check --require-complete
python tools/presentations/build-eve-oshun.py --check --preview
python tools/presentations/inventory-repository.py --check
python tools/presentations/check-workflow-coverage.py --check

The complete gate's exit one is an expected incomplete result. Every other listed command must exit zero. Hash integrity does not replace these executions or editorial review. As the project advances, changed inputs require new evidence; this receipt remains a dated record.

The same distinction applies to the adapted-host native cases retained with the guide integration. This task does not run Unreal, a game, sockets, a player dialog, account persistence or operations delivery. Source review and assignment integrity are not reader comprehension, normal-speed listening or final craft acceptance.