# Impossible Mission E028: twelfth puzzle fragment

Verified result:12 of36 pieces, SNOOZES0/LIFT INITS6, zero password letters.
All eleven previous nonblack32x16 inventory masks match exactly, and one new
mask is present. No new death is credited; three earlier ordinary deaths remain.
Independent inventory HUD1:06:38. The game is incomplete.

Room22's layout matches the historical room/node25; no fresh terminal number
was read. The agent crosses two clear24px gaps on the upper50 platform and
reaches the right lift at187.5. A physical calibration observes50to98to50,
using short released direction pulses to start motion and neutral input until
the next supported stop. It then descends to98, aligns with the small right
object208/98 and displays Searching. Two half-second search slices have explicit
neutral pauses in the same continuous source episode. The result appears in
the second pause. The actor returns toTop50, crosses the upper gaps leftward
and physically exits through the originalLeft50 doorway. No SNO or LIFT code
is spent.

The white card is initially recorded as unclassified. Only independent pocket
scanning identifies it as a fragment:14 memory clicks in each direction find
twelve distinct masks, including all eleven previous hashes. New mask SHA256:
fa44ae3a51b53818b16c2e8834ab041fe2a2f49e892a793e769ad5a9089937e6.
The observed twelve-piece sheet derives from actual rendered pocket pixels.
The current horizontal/vertical-flip analysis finds zero nonoverlapping
quartets; no assembled letter is inferred. Further fragments are needed.

Three retained initial laboratory episodes are disclosed: upper route, lift
calibration, search/exit/inventory. Each has one initial private source load
and zero intermediate restores. Three whole unpromoted Room22 tests are
omitted: a split walking pose, an occluded moving lift plate and a truncated
reward-card guard mask caused conservative stops. Their data remain archived;
no partial-search state is transferred between controllers or loads. The
successful search completes before the lift retreat, so persistence across
longer search-and-retreat cycles is still unproven. This is not uninterrupted
normal-boot gameplay or proof that the game is finished.

The scoped avatar fallback is checked against18 real walking frames and61
archived room layouts. The local guard distinguishes the actual moving plate
and32px displayed reward card from new nearby colors. Firing heading remains
unknown; successful traversal validates the shown route only. No robot-jump
or general robot-avoidance capability is claimed.

Replay:45.12s,25fps,960x800;633 actual RGB observations,
30.956914 measured PAL seconds, native320x200 field enlarged threefold.
Zero-order hold at1x, no interpolation. Decision pauses are omitted; each
source has a labelled2s initial hold and terminal RGB frame. The observed
reward and verified inventory each have an explicit4s display hold. Full
decode and seven actual-MP4 frame checks pass;57 frozen
source entries. Retained-source execution wall charge:75.139149s, distinct
from replay time and excluding omitted trials.

Policy: rendered RGB, ordinary inputs and observed memory. Separate declared
emulator raster diagnostics account for timing. Input limits, watchdog and
neutral cleanup remain active; zero agent GPU allocation. No gameplay RAM
coordinates, hidden contents, trainer or game-memory writes. No ROM, private
save-state, raw RAM dump or executable is distributed.

Video SHA256:a5294749062e893e40e370bcc378f7cc7c33da04e70b20e1279546c3bd0a6bed
YouTube:https://www.youtube.com/watch?v=T-2s-Ek9_VQ

Further ordinary furniture searches, robot avoidance, code resupply, password
assembly and reproducible final-screen evidence remain necessary. E027 and
older evidence, Pengo/Pleiads and lab catalogs are preserved.