Arcade.it AI Lab

E001 — Infrastructure & first gameplay

Sun, 04 Oct 2026 00:00:00 +0200

FROZENQWEN / COMPUTER VISIONOBSERVED

Validated checkpoint: 2026-10-05. Data last synchronized: 2026-10-05T13:10:46.576063+00:00.
Latest record verified from reviewed published run evidence. Registry counts retain the last full dataset checkpoint; record publication does not refresh the complete dataset. This is a snapshot, not a live counter.

Registry count checkpoint: 2026-10-05T08:28:16.420013+00:00. The latest reviewed record and its evidence update separately from full dataset totals.

Infrastructure and gameplay

A Windows research workstation with three RTX 3090 GPUs and 96 GB RAM runs local Ollama. A MAME bridge observes the emulated game and sends real joystick/fire inputs. Input recording, automatic restarts, video rendering and experiment logging form the initial laboratory loop.

E002 — Perception benchmark

Sun, 04 Oct 2026 00:00:00 +0200

Summary / the selected 9B baseline

9B + history3 + nearest4 → 90% Exact / 1.5% Opposite. The same model with a single 4× frame measured 80% Exact and 12% Opposite. These are horizontal player-localization results, not gameplay scores.

FROZEN v1QWENOBSERVED

A balanced, independently labelled benchmark tests horizontal player localization under different visual conditions.

200 game frames · 24 conditions · 4,800 completed inferences

Selected E002 v1 horizontal localization conditions
Model / input	Exact / 95% CI	Adjacent	Opposite	Median latency
Qwen 3.5 9B 4× nearest-neighbour + 3-frame history	90% 85.9–94.1%	8.5%	1.5%	1.984 s
Qwen 3.5 27B 4× nearest-neighbour + 3-frame history	91% 86.2–95.2%	7.5%	1.5%	4.390 s
Qwen 3.5 35B-A3B 4× nearest-neighbour + 3-frame history	86% 80.3–91.5%	9%	5%	4.031 s
Qwen 3.5 9B 4× nearest-neighbour, single frame	80% 73.7–86.1%	8%	12%	0.938 s

E003 — Controllers & Search

Sun, 04 Oct 2026 00:00:00 +0200

FROZENCOMPUTER VISION / POLICY SEARCHOBSERVED

Registry count checkpoint: 2026-10-05T08:28:16.420013+00:00. The latest reviewed record and its evidence update separately from full dataset totals.

Controllers and valid inputs

The controller branch uses fast computer vision and deterministic policies to produce legal movement and firing inputs. Timing is recorded in emulator frames so a policy can be replayed without relying on wall-clock scheduling.

E004 — Qwen selection & timing

Sun, 04 Oct 2026 00:00:00 +0200

FROZENQWENOBSERVED

Games: 30
Overall verified maximum around this phase: 2,020
Qwen selector maximum, approximately: 1,740

Registry count checkpoint: 2026-10-05T08:28:16.420013+00:00. The latest reviewed record and its evidence update separately from full dataset totals.

Higher-level selection

E004 tested Qwen as a policy selector instead of requesting every joystick action directly. It examined whether selecting a controller and its timing could improve gameplay.

E005 — Policy Search milestone

Sun, 04 Oct 2026 00:00:00 +0200

Search means zero model calls

Automated policy search explores movement and firing schedules. No Qwen or other model is called during a Search run. Candidate policies can be discovered in accelerated research; milestone confirmation runs in real time.

FROZENPOLICY SEARCHREPRODUCED

Verified real-time milestones: 3,590 → 4,450 → 5,050.

The plateau was useful evidence

Final E005 registry: 2,002 games, comprising 1,996 accelerated research games and 6 real-time confirmations. In the earlier 1,980-run sample, 12 overnight batches did not exceed 5,050.

E006 — Diversity Search

Sun, 04 Oct 2026 00:00:00 +0200

ACTIVE at checkpointPOLICY SEARCHREPRODUCED

Diversity Search improved the published verified real-time milestone from 5,050 to 5,750 points. Search uses zero model calls.

A new verified milestone

VERIFIED FROM INGESTED ARTIFACTSAUTOMATED POLICY SEARCHMODEL CALLS: 0REAL-TIMEREPLAY VERIFIED

Run ID: pleiads-20261005-131958-e6ceb4
Experiment / controller: E006 / native5
Score: 5,750 points
Survival: 216.760 s (emulated game time)
Real-time / replay: Confirmed / exact recorded-input replay
Screenshot matches: 2,239 / 2,239 captured samples
Original manifest SHA-256: 9df9cf29b00e0b2ad6f2d209b462bff7f27495a0c01c38b14e83a7c013ec3943
Evidence files: Original manifest · Ordered actions CSV · Replay comparison · Screenshot hashes
Video: Record article and video

All captured samples were checked, at one capture every six emulated frames. Screenshot bytes and game software are excluded. CSV preserves the original TSV action order. This is a laboratory milestone, not a world-record claim.

Impossible Mission E135: fourteenth fragment verified

Mon, 12 Oct 2026 12:24:52 +0000

Nine disclosed laboratory episodes, SNO0/LIFT6, no new death; game incomplete.

Impossible Mission E054: Middle98 furniture SNOOZE refill

Mon, 12 Oct 2026 06:20:50 +0000

Five retained laboratory episodes and separately labelled debit still; no new fragment or death, game incomplete.

Impossible Mission E045: six-note musical SNOOZE

Sun, 11 Oct 2026 17:45:31 +0000

Six-note challenge solved after203 incorrect orders; three laboratory episodes, game incomplete.

Impossible Mission E037: ordinary puzzle merge and Undo

Sun, 11 Oct 2026 11:32:59 +0000

13/36 fragments, 0/9 letters. One disclosed laboratory episode; no new fragment or password letter.

Impossible Mission E036: thirteenth piece and protected return

Sun, 11 Oct 2026 10:08:42 +0000

13/36 pieces, 0/9 letters. Three disclosed laboratory episodes; game incomplete.

Impossible Mission E035: six-note musical SNOOZE

Sun, 11 Oct 2026 09:12:03 +0000

Six-note challenge solved after411 incorrect orders; three laboratory episodes, game incomplete.

Impossible Mission E028: twelfth puzzle piece and safe return

Sat, 10 Oct 2026 21:57:24 +0000

Twelfth fragment collected; all eleven earlier masks retained, SNO0/LIFT6 and physical return verified.

Impossible Mission E027: second five-note musical puzzle solved

Sat, 10 Oct 2026 18:33:42 +0000

Second five-note puzzle solved with ordinary inputs; exact eleven-piece inventory and LIFT6 verified.

Impossible Mission E025: eleventh fragment from the Room29 cabinet

Sat, 10 Oct 2026 14:24:44 +0000

11/36 pieces, SNO0/LIFT4, zero letters. All ten prior masks preserved; ordinary respawn verified.

Impossible Mission E023: tenth fragment from the Room13 stereo

Sat, 10 Oct 2026 10:53:37 +0000

10/36 pieces, SNO0/LIFT4, zero letters. Two labelled laboratory sources preserve all nine previous masks.

Impossible Mission E021: ninth fragment and live robot return

Sat, 10 Oct 2026 08:58:42 +0000

9/36 pieces, SNO0/LIFT4. Three labelled lab episodes, exact prior-mask preservation and a verified inventory.

Impossible Mission E020: eighth piece and staircase return

Sat, 10 Oct 2026 07:37:36 +0000

8/36 pieces, SNO1/LIFT4. Four labelled lab episodes, exact prior-mask preservation and a verified inventory.

Impossible Mission E018: seventh piece and safe return

Sat, 10 Oct 2026 05:58:50 +0000

7/36 pieces, SNO0/LIFT3. One initial lab snapshot, zero intermediate restores, no new death. Verified RGB video and source evidence.

Impossible Mission E001: the first pixel controller works

Fri, 09 Oct 2026 13:37:09 +0000

5/5 local successes. Pixel-only reactive rules, no training, real VICE controls and watchdog cleanup verified.

Pengo M05: 500,350 verified / 500k PASS / 0 MODEL CALLS

Thu, 08 Oct 2026 18:45:40 +0000

Two new whole native reset proofs, independent pixel/round gates and complete new native video.

Pengo M04:301,030 verified /300k PASS /0 MODEL CALLS

Wed, 07 Oct 2026 18:33:33 +0000

Two NEW native reset proofs, result-bound pixel/round gates, whole new native video. Next500k with A10 unchanged.

Pengo M03B: 216,480 verified / 200k checkpoint

Wed, 07 Oct 2026 11:36:11 +0000

Two NEW native reset proofs, pixel-score gate and a complete new native video; A08 target300k remains NUL.

Pengo100k milestone: 100,400 verified

Wed, 07 Oct 2026 04:04:44 +0000

15 chronological rounds, 4 lives;300k next.

Pengo30k milestone:31,040 verified

Wed, 07 Oct 2026 00:50:55 +0000

30k milestone passes two new native resets;100k is next.

Pengo20k milestone:21,210 verified

Wed, 07 Oct 2026 00:14:15 +0000

20k reached through a measured interior diamond bonus; ACT3 remains unfinished.

Pengo E005: causal objective reduction planner

Tue, 06 Oct 2026 20:10:35 +0000

New preregistered causal-event and push-stance strategy, outcome pending.

Pengo E020: reader fix alone insufficient

Tue, 06 Oct 2026 20:10:35 +0000

NUL after counter8→5;3lives,9,490observed points, controlled intervention.

Pengo E004 review: causal hypothesis and the reader-only test

Tue, 06 Oct 2026 19:20:57 +0000

Controlled reader-only comparison takes priority over a redesigned planner.

Pengo E013: fixed routes did not reach the eggs

Tue, 06 Oct 2026 18:12:39 +0000

Four fixed routes expose travel deaths and a cell-controller limit. No search retry.

Pengo E014: a target cell is skipped

Tue, 06 Oct 2026 18:12:39 +0000

Two native replays reproduce y49 to47 while the historical controller waits for y48.

Pengo E015:23 native crush timelines and variable progress delay

Tue, 06 Oct 2026 18:12:39 +0000

Latency gate passes with23 usable events; controller and event reader still gate E005.

Pengo E016: hatching and direct egg removal differ

Tue, 06 Oct 2026 18:12:39 +0000

Matched native controls distinguish transformation from immediate objective reduction.

Pengo E002: risposta alla revisione, 4.868 valutazioni e gate successivi

Tue, 06 Oct 2026 16:44:50 +0000

3.850 punti verificati; costo 7.219.315 FE, almeno 622,203 secondi. Fidelity e N2 fermati ai rispettivi gate.

Pengo 9,790 video review v2: 9,390 alive, ACT 3 death

Tue, 06 Oct 2026 16:27:56 +0000

Permanent death warning, separate score endpoints, fixed evaluation total and exact observed-trace hash scope.

Pengo E004: the missing egg state and a verified 6-to-5 event

Tue, 06 Oct 2026 16:06:45 +0000

Objective diagnosis reveals a six-entry reader limit in ACT 3 and measures 117 frames from hit to count decrement.

Pengo E003: three lives retained, three ACTs not yet completed

Tue, 06 Oct 2026 15:29:42 +0000

The single preregistered Three Acts, Zero Deaths attempt is incomplete. No score search or silent retry follows.

Pengo: a 9,790-point planner trajectory and the ACT 3 death it exposed

Tue, 06 Oct 2026 14:54:49 +0000

Two ACTs completed, an incomplete third ACT, and a reproduced delayed death phase. No full-game or world-record claim.

Pleiads Search milestone: 5,750 verified real-time points

Mon, 05 Oct 2026 11:44:57 +0000

E006 native5 Search reached 5,750 points at real-time speed with zero model calls; exact Lua-input replay and all 2,239 captured frames matched.

Pleiads Search milestone: 5,600 verified real-time points

Mon, 05 Oct 2026 11:28:30 +0000

E006 native5 Search reached 5,600 points at real-time speed with zero model calls; exact Lua-input replay and all 2,168 captured frames matched.

Pleiads Search milestone: 5,540 verified real-time points

Mon, 05 Oct 2026 11:28:15 +0000

E006 native5 Search reached 5,540 points at real-time speed with zero model calls; exact Lua-input replay and all 2,156 captured frames matched.

Publishing a checkpoint without losing its evidence

Mon, 05 Oct 2026 10:48:33 +0000

A research submission becomes public only after its evidence, provenance and editorial context have been reviewed.

5,050 Points — Automated Search Milestone

Sun, 04 Oct 2026 00:00:00 +0200

5,050 Points — Automated Search Milestone

Automated Policy SearchREAL-TIMEMODEL CALLS: 0REPLAY VERIFIED

SEARCH — 0 MODEL CALLS

VIDEO PENDING PUBLICATION

Research footage awaits editorial and rights review.

A real-time Search milestone with deterministic replay verification. The video awaits editorial and rights review; the underlying game graphics and audio are not assumed to be copyright-free.

Date: 2026-10-03
Experiment: E005
Run ID: pleiads-20261003-201906-9c3595
Score: 5,050
Model calls during run: 0
Replay: Exact — 1,898 / 1,898 screenshots matched

The original E005 manifest, ordered actions, metrics and replay hashes are now public through the experiment evidence card. The curated 3:06 snapshot-based video preserves the recorded gameplay and timing, with correct Search attribution.

A plateau changes the next question

Sun, 04 Oct 2026 00:00:00 +0200

The E005 checkpoint revealed that many planned policy switches came after the run had ended. Those candidates could look different on paper while producing the same actions in the game.

E005 documents the plateau. E006 responds with behavioural diversity and effective-policy canonicalisation. The initial site does not claim new E006 results.

The laboratory also learned that stronger visual presentation and history can matter more than model size in a specific perception benchmark. E002 reports the frozen matrix and its hardware limitations.