# Overnight work queue — authorized by jdc, 2026-10-31

## Final implementation snapshot — 2026-11-01-r5

All requested implementation groups have an additional exact-version automated
check in [FINAL_AUDIT_R5.md](FINAL_AUDIT_R5.md). All 100 current suites pass; the
retained receiver geometry failure is now corrected. Momentum On/Off and the
lower-left dual-unit shot display are included. Original dated entries below
retain their historical results and superseded requests. Current Mini is 7 games,
Icing defaults On, and physical stars are P1 blue/P2 red/P3 green/P4 yellow.

Remaining external follow-up: actual human keeper-outlet confirmation, one-at-a-
time 147 variant human preferences, and Bodin connection/verified installation.
The original faceoff history is unknown and the earlier SanJose two-win save was
not recovered. Historical launch requests are not repeated during headless QA.
The abandoned 39/40-win target stays archived; no Hard win-rate pools resume.

## Prior snapshot — 2026-11-01-r4

Implementation items 1–6 and the retained-variant inventory are complete.
The faceoff appearance is reproduced as native free-point policy on unmodified
old/current ROMs; default-Off Faceoff Dots provides optional painted alignment.
See FACEOFF_DOTS_R4.md. The exact original incident history remains unknown.
All 97 suites execute: 96 pass, retained finisher_hold FAILED, strict gate FAILED.
Local/site ROM, IPS, all 300 tested source hashes, 753 ZIP files, 136 named hooks,
and controller/config/history preservation are verified. Bodin remains pending
until reachable. Physical keeper-outlet verification and variant preference
still require actual human play and are not replaced with invented feedback.

jdc explicitly asked for an inventory of everything still open in the project,
forum and conversation, then continued autonomous work overnight. New gameplay
alternatives should have independent start-menu On/Off options, default Off,
saved separately and locked with the season after its first match. Actual bug
repairs retain their established behavior. Work on one implementation at a time,
qualify it and update the website/source/hex evidence before the next release.
Never close a running RetroArch. The accepted 38-win Hard base is complete; the
stopped39/50 strength investigation stays stopped.

## Ordered implementation and investigation

1. [x] **Current release (2026-10-31-r10):** finish exact-source92-suite QA and publish charged
   passing plus corrected physical stars. Pass Charge Off/On, default Off;
   P1blue/P2red/P3green/P4yellow, goal red. Rebuild, exactr9rollback, matched AI
   controls and original-base IPS already pass. Charged-input/menu/physics and
   physical-color/native-goal render checks pass privately. All92checks ran;91pass, retained finisher failure remains failed. Exact evidence is linked from PASS_CHARGE_COLORS_R10.md.
2. [x] **Pass pickup-delay option (2026-10-31-r11):** investigate the real native pickup/catch
   timing for ordinary B passes, then implement a bounded independent optional
   choice. Released as independent default-Off Pickup Delay; see PICKUP_DELAY_R11.md. Preserve A flips, C shots, CPU keeper requests, downed-player guards,
   existing pass mode/cone/lift/speed/charge and P1–P4 switching. Attribute to
   [Brodeur30 #217609](https://forum.nhl94.com/index.php?/topic/40511-2-guys-one-ai/#comment-217609).
3. [x] **Alternative goalie/rebounds option (2026-11-01):** investigate an optional reachable
   keeper/rebound model without changing the accepted default shot balance.
   Released as independent default-Off Alt Goalies; see ALT_GOALIES_20261101.md.
   Test directional rebounds, catches, slow goalward deflections, both ends and
   co-op outlets. Attribute to
   [floating #217606](https://forum.nhl94.com/index.php?/topic/40511-2-guys-one-ai/#comment-217606).
4. [x] **Remaining shootout parity (2026-11-01-r2):** unique eligible shooters beyond the native
   five-slot rotation, and explicit spinorama handling. Retain released automatic
   start, legal forward loose-puck recovery, backward/stopped-approach handling,
   single-shot/retouch rules and physical-pad ownership. New alternatives should
   be optional; verified against current official rules. Released default-Off
   OT SO Rules for the tied-overtime extension; see STRICT_SO_R2.md.
5. [x] **Optional AI shot decisions (2026-11-01-r3):** investigate the requested long-shot zone,
   safe one-timer choice and lateral approach, using real native choices and
   timeout/pressure fallbacks. Released independent default-Off AI Shot Plan;
   see AI_SHOT_PLAN_R3.md. The lateral keeper exploit itself was already
   repaired in2026-10-30-r3. Preserve the accepted default Hard behavior and all
   native movement/attributes/physics; do not resume win-rate pools.
6. [x] **Native free-point faceoff displacement (2026-11-01-r4):** matching phenotype
   reproduced after actual native queued minor on exact unmodified old/current
   ROMs. Released independent default-Off Faceoff Dots with native regional
   geometry, full ABI, menu/storage/lock and actual foul regressions. Original
   incident history remains unauthenticated. See FACEOFF_DOTS_R4.md.
7. [x] **Retained AI inventory/review queue:** prepare a hash/behavior/result/failure inventory
   and an orderly one-at-a-time human test queue after other work. Inventory
   complete:241 images,244 recorded series,147 review images; see
   RETAINED_AI_VARIANTS.md. Do not invent
   human feedback or silently discard known candidate regressions. Human
   comparison itself waits for jdc to play the variants.
8. [ ] **Delivery/documentation audit:** keep website statuses/authors/post links,
   every hook/hex/source export, exact QA and task list consistent. Replace the
   single current local/site/Bodin release, fresh saves per policy; preserve
   controller/config/history hashes. Bodin remains pending until reachable.

## Open human verification

- The keeper-outlet deadline repair still needs jdc's physical-controller retest;
  automated native checks pass. Keep the distinction explicit.
- Variant difficulty and preference need actual human play; CPU results alone
  cannot determine which variant is hardest for a person.

## Forum audit

Read the complete current thread on2026-10-31 through jdc's#217636, including
Brodeur30#217634. No additional unlisted concrete bug report was found in that
snapshot. Source:
[2 guys one ai](https://forum.nhl94.com/index.php?/topic/40511-2-guys-one-ai/).
Completed requests (settings/PIM, diving possession timer, awareness, receiver
availability, runaway speed, penalty-shot state, short-pass timing, credits and
released direction/cone/lift/speed) are not reopened as new tasks.

## Current evidence caveat

All 100 required current suites pass, including finisher_hold. Earlier failed
matrices and rejected prototypes remain preserved. New failures or missing
coverage may never be silently accepted or labelled passed.

## Final additional implementation

9. [x] Momentum On/Off, default On: actual skill scaling, counters, bars and saved locks.
10. [x] Lower-left dual-unit shot meter replacing the NHL badge, native PP fallback.
11. [x] Retained receiver tracking/braking and Hard containment arrival correction.
12. [x] Exact-source100/100 qualification and 48-group acceptance checklist; final
    local/site/remote artifact status is recorded in FINAL_AUDIT_R5.md.