Dated source reference. Current release status: handoff state. Full original relative paths are in the developer archive.

API development changes

0.2.0.dev57 crowd fullness percentages — November7,2026

Public crowd inputs and readback now use 0–100 percent: 35 means 35% and 100 means full. The API converts launch settings, live targets and guarded preimages to the unchanged native 0–1 Float32 protocol, and converts observations/receipts back to percentages. Python and TypeScript helpers, schema bounds and examples follow the same units. Public crowd state/catalog version2 declares units: "percent"; dev56 callers must multiply their former fraction inputs by100. HTTPv1, ABI26 and the game-tested Crowd.1 DLL are unchanged. Verification.

Private development56 — November7,2026 UTC

Adds per-launch native audience fullness and guarded live attendance set/reset with scalar readback, scene/epoch/revision/preimage checks, durable command identity and lease ownership. GET /v1/crowd and optional snapshot/delta crowd state expose startup verification and eligible native anchors. Python/TypeScript helpers prepare requests without automatic mutation retries. HTTPv1/main ABI26 are retained; the crowd mapping is separately versioned. See usage and exact verification.

Private development51 — October18,2026

Adds rotation plans/durable clock-and-entry observations, original native timeout calls, connected native input discovery and startup side assignment (including explicit CPU clear).510 Windows tests/84 schemas and focused live college/original/keyboard acceptance pass. Physical gamepad multiplayer remains unverified. See the ledger. Public packages remain unchanged.

0.2.0.dev50 external game flow — October18,2026 UTC

ABI25 adds guarded bench-to-court queue/cancel plus native position/pending observations, with durable entry observations separate from immutable queue receipts. Corrected availability2/PE substitution version2 preserves API-owned pending requests against native automatic replacement; the original stoppage/entrance processor still selects the actual entry opportunity. Startup defense adds up to144 per-opponent values to the282-value batch, total426, with complete readback before reveal. Portable ResultInbox commits before ACK and supports independent consumers/lost-ACK recovery/explicit matching-owned finished closure.493 Windows API tests,76 schemas,111 new native cases,1,104 prior cases and27 pause cases pass. Live51w verifies156 startup values and both actual entries; full flow and preservation pass; see the external-flow ledger. Failed51k/result-only passes remain distinct. Guide/evidence. No public package or exact-minute schedule is added.

0.2.0.dev48–49 live coaching/defense — October18,2026 UTC

ABI23 adds paused per-team coaching and six per-opponent defensive fields, including preferred defenders. ABI24 adds an expiring adapter-owned native pause layer plus guarded simulation-dispatch suspension. Counter-only prototype availability1 is refused by current readers; the accepted hook uses availability2/PE simulation-control version2. All writes retain leases, fresh revisions/preimages, run/period and game-thread gates.471 Windows API tests,1,104 defensive/coaching instruction cases,27 pause cases and71 schemas pass. Live49zf verifies42 distinct changes,89 commands, persistence/restoration, frozen clocks/player positions and pause expiry with zero adapter errors/generated input. Guide and exact evidence. Direct substitutions/startup defense remain unfinished; public downloads/stable35 are unchanged.

0.2.0.dev47 per-team coaching — October17,2026

ABI22 appends76 bytes of coaching observations. Managed settings accept nine named coaching choices independently for home and away, adding18 optional entries to the264-value startup batch. Original guarded handler suffixes apply explicit sides before reveal, with complete readback. Settings/snapshots/deltas expose native state and catalog choices; older hooks refuse requesting startup and advertise unavailable coaching.452 Windows API tests,244 native instruction cases and64 schemas pass. College/all282 and original/reversed-side requests reach play without generated input. Native presentation, direct substitutions, exact rotation schedules and defensive matchups remain separate. Evidence and client guide. Existing public ZIPs/stable35 are unchanged.

0.2.0.dev40–46 external match expansion — October17,2026

Private HTTPv1 source now supports one app-owned version1 match request from process launch through native loading, roster preparation, startup controls and pre-play identity binding. The public SDK remains available independently. New per-child rules choose original quarters or college1–20-minute halves, fixed TV marks and scaled OT without editing the original INI. Phase4 binding verifies the loaded ROS, complete actual slot/name mapping, full opening clock, zero stats and event boundary; optional finite match metadata survives results. Uninstantiated plan entries remain explicit. Owned session.stop cancellation abandons without a result; finished requires an already reconciled final result. Durable request/command receipts retain ambiguous Start outcomes and forbid automatic replay.

ABI19 adds264 startup option/slider/controller-preference values through original setters/readback; ABI20 adds native exhibition camera/descriptor selection; ABI21 adds CPU or primary home/away participation through original picker/control commit with matching current/saved roles and input records. The loading cover waits for successful opening readback. Physical device selection, key bindings, multiple humans, live bulk presets and disk persistence remain separate. Normal EXE startup still opens Quick, while managed roster replacement uses covered Home and product startup generates no input.

The optional shared writer exposes39 named roster tables, paged field/query APIs, immutable edits and starting order/native minute preparation. Authored IDs can launch without reupload. API46 exports registered ROS bytes with authenticated SHA-bound download; the SDK independently verifies them and atomically creates a new file without overwriting. No proprietary editor executes, no arbitrary memory/path API is exposed, and file authoring/export is separate from native loading or asset certification.

Latest source47zzp passes441 Windows API tests; native47zze is compiled with warnings-as-errors. All264 presets, app request orchestration, authoring-to-game, college/original rules and2k/broadcast startup cameras pass scoped live checks. Exact-native home/away control commit passes; away has rendered marker and stock export evidence. Combined47zzq passes the expanded/college10/all264/2k/CPU/identity/16MiB-export flow on this exact native DLL;61 schemas validate12 live scenarios/188 commands and unchanged historical results. The active ledger links exact identities/limits and final preservation. Public preview2/API36, editor dev7, stable35 and historical packages remain unchanged. Full rulebook, legal ball-state/universal events, native presentation/extra-field consumers and safe asset hot-swap are not newly claimed.

0.2.0.dev39 native startup and loading — October17,2026

Normal EXE startup uses guarded native patches to skip opening/title waits, preserve required initialization and open Quick Game directly. Managed --launch-game generates no input and retains Home under a neutral or app-configurable loading cover until the requested match reaches native presentation. Version1 JSON supports title/subtitle, colors, local logo/background and window branding. SnapshotABI17 appends16bytes of startup observations; HTTPv1/eventABI4/profile4/court3 remain unchanged. Preflight reads the hook's managed-version export before creating the child; failed broker setup closes that child.

Expanded roster replacement while Quick preview objects were active exposed a native null read. Both broker and native adapter now require inactive Home for roster.load; managed startup prepares there while covered. Start still supports Home/Quick.355 final-source Windows API tests/41 schemas pass. Final manual Quick startup, neutral stock launch and branded905-team launch use the same non-probe DLL. Evidence and limits. Public downloads/stable35 remain unchanged; native in-game presentation and pre-tip identity binding remain separate.

0.2.0.dev38 roster load/direct Start — October16,2026

SnapshotABI16 appends content-bound loaded-roster state and a guarded native launch mailbox. Authenticated binary upload/local import copies standalone ROS files into private storage; roster.load invokes the original decoder/reset tail and match.start invokes the original exhibition factory/loading transition from Home/Quick. Independent SDK/example handles upload→load→13-control preparation→Start, with durable receipts and explicit recovery identities. Optional --launch-game starts a supported Windows game in the signed-in desktop and acknowledges recognized intro/title states. It never navigates team selection.

345 exact-source Windows API tests/39 schemas pass. Stock105-team and expanded905-team original-handler startup pass; fresh unattended905-team flow verifies all13 controls, ten starters, roster order/four minute preferences, college clocks and zero errors. Existing phase10 binding/result consumers remain separate; pre-tip identity binding, custom-asset compatibility and full-mode certification remain open. Public preview2/editor dev7/stable35 are unchanged. Verification and limits.

0.2.0.dev37 match preparation — October16,2026

Append-only snapshot ABI15 adds Game Style, loaded team/uniform row selection and paged read-only catalogs. All nine Game Options settings plus both teams/uniforms are supported. prepare_matchup.py exports versioned run/catalog-bound plans, preflights choices, applies commands in dependency order and returns receipts/readback. Existing eight-control and legacy ABI readers remain compatible. HTTPv1/eventABI4/profile4/court3 are unchanged.

307 exact-source Windows API tests/30 compiled native cases/35 schemas pass. Stock13-control setup and generated905-team final-participant startup pass scoped live checks. Separate source-only roster preparation adds starting order/native minute preferences, with116 writer tests and native starting-five/order/minute readback. Loading/Start remain manual; exact substitutions, pre-tip binding and native extra-field consumption remain future. Uniform menu previews can lag actual match selections. This candidate is isolated, absent from immutable public preview2/editor dev7; stable35/95 saves/hosts remain preserved. Evidence/limits.

0.2.0.dev36 documentation/metadata checkpoint — October15,2026

Current HTTPv1, snapshot ABI14, event ABI4, college profile4 and court definition3 are documented consistently. Python distribution metadata now matches the implementation's existing 0.2.0.dev36 version. No native gameplay, adapter layout, command types, routes or DLL changed.

Capability descriptions now record the latest natural college OT/final-export pass, earlier four automatic TV returns,18 both-end controlled release-contact cases and validated compact8K reference floor. Verification scopes distinguish reference assets/scenarios, older build evidence and arbitrary currently attached binaries/artwork.

Newly normalized profile4 declares final_exports_verified: true for reference export scenarios and replaces coarse unfinished-area labels with specific deferred verification limits. Profiles1–3 retain false and their historical coarse labels. full_rulebook_verified, capability college_rules.complete and arbitrary live texture alignment remain conservative. Result completeness.rule_profile_verified follows export-profile evidence and can become true on new profile4 results; its historical name does not mean full-rulebook certification. No historical result/fixture is rewritten. The canonical schema accepts both stored false and current true for profile4, while retaining false for older profiles. Clients validating against a pinned earlier false-only schema should update the schema; HTTPv1 and the field's boolean type are unchanged.

The additional uninstrumented production full-game smoke and rare live injury/rim-recovery/held-ball reproductions are deferred. Existing controlled tests remain evidence; no new natural gameplay validation is claimed. No public package, publication, game launch or installed-DLL promotion is part of this update.

Earlier checkpoints added pregame preparation (development35/ABI11), college logical periods and bonus (ABI12), explicit media ownership (ABI13) and original court samples (ABI14). Those original packages, manifests, test counts and work-log entries remain historical evidence. See current verification and chronological integration records.