State at the handoff
The installable baseline is API 0.2.0.dev58, TypeScript SDK 0.1.0-dev.3, HTTP v1, snapshot ABI26, event ABI4, college profile4 and court definition3. Native DLL SHA-256 is a24ff0c66051a3a9e46b1cdfcef740b73436446169786d2d865652dc7d823453.
| Area | Delivered behavior | Acceptance and remaining limits |
|---|---|---|
| Bootstrap and saves | Offline startup/onboarding bypass, generic local player, portable saves, guarded loader installation and rollback | Preserve valid existing profile/settings; unknown loaders and changed files are refused |
| App-owned match startup | Roster intake, preparation, choices, rules, identity and native launch behind a loading cover | Private per-session broker/state/token; preparation must complete before reveal |
| Live API and results | Copied snapshots, events, commands, exclusive leases, reconciled immutable final results, independent consumers and durable ACK | Unknown semantic/legal fields remain unknown; main and rule snapshots are not one native transaction |
| Match controls | Clock-based rotation intent, observed native substitutions, timeout calls, coaching and connected-input/team assignment | Physical multiple-gamepad coverage is unavailable; intent and actual native entry are distinct |
| College and rule extensions | Scoped college mechanics plus experimental timing/reset/bonus/FT/classification/native C SDK capabilities | No universal NCAA/NBA event certification; changed FT-count game proof remains limited |
| Roster expansion | Native save/reload with 906 teams, 13,680 players and 3,790 uniforms in the documented capacity experiment | That fixture is a scoped capacity result, not a promise for every mod or maximum configuration |
| Roster authoring | Independent Python/C# writer, structured fields, companion metadata, CSV/grid edits, browser collaboration, accounts and backups | Native code/IFF/career semantics are not universally mapped; keep ROS and matching companion together |
| Crowd attendance | Launch/live fullness and reset with guarded native application | Public values are 0–100 percent; distant/near audience included, static baked spectators/audio separate |
| Visiting crowd | Launch/live away share, authentic team clothing/palettes, stable weighted whole-row allocation | Global split only; no named sections or behind-bench targeting; row sizes quantize actual allocation |
| Ultrawide | Hooked 21:9/32:9 framing and scoped UI/input adjustments | Windowed 3440x1440/5120x1440 evidence; physical monitor/fullscreen/reset certification unavailable |
| Native metadata and match assets | Pre-reveal frozen metadata and admitted immutable asset overlays | Selected IFF reads verified; arbitrary hot-swap and all asset types are separate |
| Presentation | Portable composition, software drawing, authoring/bundles, data/cue bridge, scoped native owners and extensive resource/callback research | Full framework unfinished; exact ESPN design unaccepted; packet mask zero at RG |
Exact sealed dev58 evidence
The original game finished naturally 11 away / 12 home with 3,588 schema checks and 19 applied commands. The college game finished naturally 5 away / 6 home with 1,888 schema checks and 19 applied commands. Both delivered equal results to independent consumers and both ACKs, restored 103 files and released custody. Twelve separately reviewed paused backbuffer images have narrower camera/artwork scope; two image runs failed their later overall harness gates and remain failed runs.
The actual sealed package passed 666 Windows API tests, 139 schemas / 52 operations, 30 SDK cases, standalone examples and ten installer/rollback groups. See evidence/analysis/crowd-away78/sealed-package78.json and native-acceptance.json in the developer archive. These are dated inherited receipts, not new execution during this handoff.
Current presentation checkpoint
November 9, 15:07 UTC: packet75rg freezes 817 inputs, passes 149 Windows groups, builds ten native profiles and passes 666 API tests. The identical resource-probe DLL passes original EG and college EH after 30.098/45.046 seconds of verified clock movement. Both capture the original shared-quad CRT allocation and all 48,672 index bytes with zero unavailable DrawUP inputs, and pass mixed-target/direct-parameter joins. Both restore 99 files and release custody.
This proves bounded copied input/reference behavior. Complete callback/resource/source/free/context closure, semantic ownership/cues, complete visuals/lifecycle and final author delivery remain open. Live working files may contain later unqualified changes; only the frozen RG checkpoint inherits its receipt.