메뉴
Use when working on SIMPLE's streaming microchunk class-average rejection, simple_microchunked2D lifecycle sentinels, pass-1/pass-2/refchunk/match generation, skip_pass_2 behavior, simple_cluster2D_rejector criteria, or boundaries with model_cavgs_rejection.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when working in SIMPLE's src/main/nano subsystem, including nanoparticle and molecule data structures, graphene and time-series helpers, and the SINGLE-specific domain logic that supports liquid-cell and nanoparticle workflows.
Use when working on SIMPLE's abinitio2D workflow and its cluster2D stages, including 2D search strategies, probabilistic 2D sampling, class-average restoration, distributed partial class-average assembly, online matcher particle I/O, project updates, and streaming/rejection helpers.
Use when working on SIMPLE's model_cavgs_rejection application, simple_cavg_quality feature-space selector, class-average quality selection, quality_mode=analyze outputs, cavgs_quality feature/reference tables, chunk vs pool rejection tuning, or microchunk quality-vector integration.
Use when studying, explaining, or modifying SIMPLE's fractional-update and trailing-reconstruction workflow, including update_frac sampling, sampled/updatecnt bookkeeping, matcher reconstruction I/O, partial even/odd reconstruction handoff, rec3D/volassemble ownership, downsampling compatibility, obsfield mirrors, and stage-to-stage previous-reference contracts.
Use when working in SIMPLE's src/main/stream subsystem, including online/mini-stream orchestration, chunk and microchunk processing, staged pipeline steps, watchers, queue/message definitions, and streaming variants of preprocessing and cluster2D workflows.
Use when working on SIMPLE's refine3D workflow, including CLI/UI entrypoints, commanders, probabilistic pre-alignment, 3D search and matcher strategies, online partial reconstruction, volassemble, automasking, nonuniform filtering, multi-state behavior, distributed execution, and related abinitio3D refinement code paths.
| name | simple-microchunk-rejection |
| description | Use when working on SIMPLE's streaming microchunk class-average rejection, simple_microchunked2D lifecycle sentinels, pass-1/pass-2/refchunk/match generation, skip_pass_2 behavior, simple_cluster2D_rejector criteria, or boundaries with model_cavgs_rejection. |
Use this skill for src/main/stream/simple_microchunked2D.f90, src/main/stream/simple_cluster2D_rejector.f90, and stream integration questions involving model_cavgs_rejection.
doc/microchunk_and_rejection/microchunk_rejection_policy.mddoc/microchunk_and_rejection/microchunk_rejection_model_integration.mddoc/microchunk_and_rejection/model_cavgs_rejection.md when comparing the rule engine with the learned model backend.simple_microchunked2D owns stream lifecycle, sentinel reconstruction, tier generation, match finalization, and combined outputs. simple_cluster2D_rejector owns the live scalar rejection rules. src/main/cavg_quality and model_cavgs_rejection are separate unless a task explicitly integrates them.
Restart state is reconstructed from sentinel files:
ABINITIO2D_FINISHED: chunk job finished.REJECTION_FINISHED: rejection and particle cleanup finished.REJECTION_FAILED: rejection could not be performed safely.COMPLETE: chunk was consumed or finalized.On import, failed chunks count as rejection-complete and complete. Do not consume a source chunk into the next tier until its project file for the next tier has been written successfully.
The live rejector applies cumulative class-average rejection in this order:
res > 40.0;After rejection, rejected classes map to particle state=0, and rejected particles must have class=0 and class_match=0.
The normal ladder is pass-1 -> pass-2 -> refchunk -> match. Pass-1 chunks become complete only after pass-2 projects are written. Pass-2 chunks feed the refchunk but are marked complete only after match chunks are generated. Match finalization updates accepted/rejected counters, writes COMPLETE, and copies project files to the completion directory.
When skip_pass_2 is active, pass-2 import and generation are skipped. Rejection-complete pass-1 chunks feed the refchunk directly, and remaining pass-1 chunks become match chunks directly after the reference stack and box are available.
model_cavgs_rejection is a learned feature-vector backend with hard validity gates, normalized features, model weights, clustering, and threshold controls. It currently exposes chunk_default_v2, chunk_lp4, pool_default_v2, microchunk_p1, and microchunk_p2 presets. Do not replace stream rejection decisions with model output unless the task also preserves stream sentinels, selected/rejected stacks, project state propagation, and match/finalization semantics.