# Keeper catch handoff and requested defaults — 2026-10-30-r6 jdc reported a dark-blue keeper marker after a catch and requested CPU control (light-blue marker) with outlets to the requesting P1/P2/P3/P4 skater. His latest start-menu screenshot shows Auto lines, Fighting Off, Hard, Auto Replays Off, Crit Shots Off and Icing Off. This release follows that latest screenshot; r5's earlier explicit Icing On request remains recorded in its historical notes. ## Reproduction and change The r5 input recording contains a Y press before a later catch. The recovered native pre-catch state already has P1 controlling keeper5. At replay frame9757, r5 accepts the puck but leaves control5 and human flag0800 (flags8800). Its existing pickup guard stops a new automatic keeper assignment but does not reverse manual control established before the catch. The same native state and input prefix on r6 accept the same catch, return P1 to a free skater (object4 in this scene) and remove keeper human ownership. The native AI job flag is armed (8200), then processed normally (8000). The native marker uses this CPU ownership; no palette or artwork is recolored. No game positions, velocities, scoring, clock or catch results are forced in this reported-scene regression. The checkpoint was recovered from the actual r5 initial checkpoint and recorded inputs using the same core/devices; the session started without pre-existing SRAM. No input queries are missing. Subsequent branches change only controller input and are not independent games. `goalie_ai.s` calls the **288-byte `goalie_catch_control.s` at2DBC00** from its approved cooperative-keeper path. The new call is at **2FE46E**. The original 15322 pickup hook remains. The helper runs only after an accepted owned catch, with Manual goalie control and a previously human-controlled keeper. It uses the native **C0F8** assignment routine to release the keeper and select the nearest eligible free skater. Other humans, boxed/unavailable players and keepers are excluded. P3/P4 temporary bank mirrors are updated consistently. Native pass/recovery flags are preserved. Every register, CCR and stack value is restored; no new persistent RAM is allocated. If nobody can receive control, native object-1 safely detaches the pad while the keeper still becomes AI. A pad without an eligible skater cannot request an outlet until a receiver is available. Opposing human teams, CPU-only keepers, shootouts, penalty shots and stopped play keep their existing rules. Explicit Y pressed after the catch can still select the keeper. The ordinary outlet publisher now lets a fresh B reach the validated keeper request path before ordinary switching. An accepted AI-keeper request keeps that pad's current skater. Otherwise the existing immediate nearest-puck switch continues. The adapter's raw short-tap publisher and first-valid-request priority remain. Physical A is logical Genesis B in jdc's current controller mapping. The r4 nonnegative-deadline grace remains; expired/queued whistles are not revived. ## Defaults and compatibility `crowd.s` initializes Hard; `seasonplayers.s` uses Hard only for an empty season without a valid difficulty record. Played legacy fallback remains Normal. `replaymenu.s` uses Off for absent/corrupt Replay/Crit preferences while honoring valid saved On/Off records, without writing fallback bytes at boot. `slots.s` gives new/restarted formats Hard and Icing Off. Period length, Manual keeper control, penalties and Auto lines retain their defaults. Edmonton stays club10; an ongoing season keeps its saved club. Existing S/T records and played legacy saves retain native Icing On; valid U records retain their own choice. Only an empty season without a valid record receives the new Icing Off default. The normal post-first-result locks are unchanged. `--legacy-goalie-catch --legacy-menu-defaults --version 2026-10-30-r5` reproduces r5 byte-for-byte. Both flags are required for that complete rollback. ## Exact-build validation - **44** native handoff ABI/guard cases cover both teams, every active pad/count, pass flags, no free receiver, orphan/stale ownership, Auto keeper, opposing humans and special/stopped modes. All registers/CCR/stack, all12 whole objects and all8 control/side words are checked. These are synthetic native fixtures. - **148** native adapter cases pass, including **68 added** pre-controlled-catch handoff/current-requester/hold cases: every pairing of keeper-owning pad and requesting pad for1–4 teammates, both teams, no takeover or stolen controller. The original ordinary two-pad, Y and deadline suites also pass. - **5** reported-catch replay branches reproduce r5's retained human ownership, then verify current CPU catch, no-input hold and requests to the current skater. - **6** native fresh/default/new-format/saved-choice cases check the actual value renderer. The 34-case Icing and existing Crit/Replay compatibility suites pass. - All **75 shared suites executed;74 pass**. The inherited `finisher_hold` failure remains: native1.9046278288761302 vs held2.777268439788352; early one-timers2vs2, four Normal/Easy controls unchanged. Strict release gate stays **FAILED**. This retains the user-selected38-win playtest's documented AI limitation. - Forty current no-B scenes (24000 frame samples) have identical puck, ownership, match-clock, score and native action/control timelines. **35 raw traces are exact to fresh r5;15 observations across5 traces are not raw-byte-identical**. Each difference is confined to one actor's X/Y, the goal-line prediction cache or sprite-list cursor, with both neighbouring whole rows exact. Coordinates must match an adjacent reference coordinate; derived cache/cursor values must stay within the adjacent native range. All other fields remain exact. These are consistent with video sampling during native updates; no complete raw invariance is claimed. The historical observer calls BEEC `shot_clock`, but the native disassembly identifies it as puckcross6 crossing prediction. Fresh r5/r4/r3 controls agree exactly; fresh r2 controls reproduce the immutable original traces. Binary scope proofs and all raw diagnostic failures remain. - Independent rebuild, exact r5 rollback and original-base IPS reconstruction pass. Original MD5:8f863ecb108c1d213953bada9b98f4f2. The expanded low-level B test now models P3/P4's actual physical side/control words as well as their temporary P1/P2 mirrors. The earlier synthetic setup left the physical words empty, which cannot represent a valid P3/P4 outlet. Real four-pad gameplay independently checks the resulting ownership and target. Global-option tests seed explicit saved On for edit flows and independently check the newly requested Off fallback. Diagnostic failures remain recorded. ROM SHA256:74f099b10459e000b4b50870e41617983641445f83c813328dba019b6b02fd7d; CRC32:25BBB406. Fresh progress per standing release policy; configs/remaps/history retained. The faceoff alignment report remains open; the Hard strength search stays stopped. ## Report provenance All75 shared checks ran on this exact image. Four affected checks were rerun on the same image after the fixture/scope corrections described above;71 unaffected implementations and results remain unchanged. The final report binds the current source manifest and explicitly records this refresh. Timing experiments were private and were discarded; final goalie_ai.s retains the small cooperative catch call and the original operand forms. The finisher comparison also rebuilds the exact current-source no-hold ablation, so the new catch hook's timing is shared by both sides; its four Normal/Easy controls remain strict raw comparisons. The replay audit retains all40 cases before reporting a comparison failure. Native prediction-cache naming/updates were checked against the primary source: https://github.com/Chaos81/nhl94-disassembly/blob/main/src/ram_addrs.inc and its `findpc` routine in nhl94.bin.lst. Raw recordings/states remain private.