LoopX 2.0 · Research film · baseline¶
The workbench for long-horizon research. For enterprise teams, research organizations, and independent builders. 2.0 development preview. This is a production baseline, not a release declaration.
Video / 视频 · EN VTT / SRT · 中文字幕 / SRT English · 简体中文
00:00:00.000–00:00:07.000 · The workbench for long-horizon research.¶
2.0 DEVELOPMENT PREVIEW
Loop X two point zero. A workbench for long-horizon research.
00:00:07.000–00:00:25.000 · Delegate the goal. Keep its context.¶
ILLUSTRATED HANDOFF + CODE-PROJECTED 3D + SEPARATE RECORDED RETURN
Carry the goal, constraints, and return route. Let the existing owner adopt the work. Keep corrections on the task. Return the evidence. Peers accept separate scopes. The build uses the research artifact.
00:00:25.000–00:00:29.500 · Make research work visible.¶
AUTHORED WORKSPACE SCENARIO
See work, waiting decisions, and schedules.
00:00:29.500–00:00:33.500 · Make the next action explicit.¶
AUTHORED WORKSPACE SCENARIO
Separate the next task from a pending approval.
00:00:33.500–00:00:37.000 · Keep responsibility visible.¶
RECORDED WORKSPACE · ORIGINAL STATUSES RETAINED
Keep owners and waiting work visible.
00:00:37.000–00:00:42.500 · Trace the research across sessions.¶
RECORDED HISTORY · COUNT RIBBONS REDRAWN TO SCALE
Keep questions, findings, and changes of direction.
00:00:42.500–00:00:52.000 · Inspect the process. Check the source.¶
PACKAGED UI FIXTURE · SYNTHETIC DATA + ILLUSTRATED SOURCE PATH
Read six process judgments. Keep unknowns visible. Trace claims to their source. Measure results and costs separately.
00:00:52.000–00:01:00.400 · Read the result in its scope.¶
ONE HISTORICAL SOLVER RECORD · TWO CPU THREADS
One generated-graph probe improved from forty one point two to eighty five percent. That result applies to the duplication condition shown.
00:01:00.400–00:01:07.600 · Count the cost. Preserve the counterexample.¶
SAME HISTORICAL RECORD · NO INDEPENDENT RERUN
The independent control kept its accuracy. Time rose thirty two percent. Native score gain remains unproven.
00:01:07.600–00:01:13.100 · Carry the evidence into the next step.¶
ILLUSTRATED WORKFLOW · NO NEW TEST EXECUTION
Keep failed paths and their reasons. Choose the next test.
00:01:13.100–00:01:16.700 · Control when research continues.¶
AUTHORED SCENARIO · UNCONFIGURED AND STOPPED STATES
Check continuation, hosts, and channels.
00:01:16.700–00:01:20.700 · Bring your models and tools.¶
MECHANISM + REUSED WATERMARKED SEEDANCE 2.5 CONCEPT
Your runtime brings the models and tools.
00:01:20.700–00:01:26.200 · Research that can continue.¶
OPEN SOURCE · LOCAL FIRST · 2.0 DEVELOPMENT LINE
Loop X. Keep the research, evidence, and next decision together.
Sources and limits¶
Historical record: generated full-size joint-neighborhood graphs, 30% partial-duplication probe, two CPU threads; accuracy 41.2% → 85.0%. Independent-edge accuracy unchanged, time +32%. Native score gain unproven. No independent rerun or LoopX-versus-baseline comparison. The six process judgments are goal alignment, benefit testing, discrimination, information gain, evidence adoption and avoidable repetition. They use synthetic packaged-UI data; uncalibrated, with separate eligibility and no total score. See the assessment contract. Citation paths and teaching choices are illustrated. Recorded workspace and engineering return are separate source cases. English narration uses a generic local synthetic voice. Chinese subtitles translate each spoken cue. A labelled watermarked Seedance 2.5 concept insert is reused; no new remote generation, analysis or benchmark job was launched. The proposed next native test has not been run for this film. Historical source record.
The 18–25s insert reuses a code-projected 3D illustration of separate peer scopes and an artifact dependency. It is not Blender, a physics simulation or the recorded App case. This is the single current film, led by the earlier A product-flow edit. It freezes an artifact for later creative comparisons; future research can change the story, style, duration and tools. No comprehension or causal communication benefit has been established.
The history image is annotated: count ribbons are redrawn from the unchanged 104/2/27 and 39/1/22 records, so the refuted shares are 2/133 (1.50%) and 1/62 (1.61%). All counts sit outside the bars. This is a static-media correction, not a running-App change or independent validation of the recorded findings.