13 KiB
13 KiB
Research Wiki 索引
自动生成,更新时间:2026-07-16 02:01 UTC
design (34)
- 2026-07-06-core-agent-adapters-llm-design
design:2026-07-06-core-agent-adapters-llm-design - 2026-07-07-app-harness-design
design:2026-07-07-app-harness-design - 2026-07-07-core-evolution-design
design:2026-07-07-core-evolution-design - 2026-07-08-tree-repair-resilience-design
design:2026-07-08-tree-repair-resilience-design - 2026-07-09-paper-main-figure-design
design:2026-07-09-paper-main-figure-design - 2026-07-11-agent-runtime-fixes-design
design:2026-07-11-agent-runtime-fixes-design - 2026-07-11-batch-tree-build-design
design:2026-07-11-batch-tree-build-design - 2026-07-11-question-gen-v2-design
design:2026-07-11-question-gen-v2-design - 2026-07-12-per-category-pool-strategy-design
design:2026-07-12-per-category-pool-strategy-design - Action Recognition 单题型首次训练实验设计
design:2026-07-14-action-recognition-training-design - main.py 推理入口 + 初始 Prompt 集设计
design:2026-07-09-main-inference-entry-design - main.py 推理入口 + 初始 Prompt 集设计
design:main-inference-entry - Per-Category Pool Strategy 设计
design:per-category-pool-strategy - question-gen v2 — AR grounded-contrastive 重构(双模式生成 + 独立 VLM grounding + 生成端对抗)
design:2026-07-15-question-gen-v2-grounded-contrastive-design - question-gen v3 — 构造优先范式重构(全孪生 + 四家族真独立 + 坍缩度量)
design:2026-07-15-question-gen-v3-construction-paradigm-design - Spec-1 Agent 执行环境修复(解析容错+步级重试+摘要附实体)
design:agent-runtime-fixes - Spec-2 建树批量并行入口
design:batch-tree-build - Spec-3 出题管线 v2(失败机理靶向+逐题质量门)
design:question-gen-v2 - 修复诊断 tree_data 断链 bug(TRM4→TRM5 迁移 regression)
design:fix-diagnosis-tree-data-link - 出题模块迁移设计(question_gen)
design:2026-07-07-question-gen-design - 出题管线 TaskTypeStrategy 拆分设计(Clean Architecture)
design:2026-07-14-task-type-strategy-design - 出题质量提升 Phase A — Grounded 单题质量(候选池 + VLM 视觉打分 selector + 单维反事实)
design:2026-07-14-grounded-question-gen-phaseA-design - 出题质量提升 Phase B — Agent 对抗过滤(独立后置过滤层)
design:2026-07-14-adversarial-question-gen-phaseB-design - 建树修复管线:熔断根因修复 + 断点续跑 + 并发改造
design:tree-repair-resilience - 建树模块竖切设计:数据结构 + 建树 + 修复 + 迁移
design:2026-07-07-tree-module-design - 建树模块竖切设计:数据结构 + 建树 + 修复 + 迁移
design:tree-module-vertical-slice - 搜索 Agent 装配层设计(app/search/)
design:2026-07-07-search-module-design - 结果驱动的视频级 train/val/test 切分(替代自造题模块)
design:2026-07-15-results-driven-video-split-design - 结果驱动的视频级 train/val/test 切分(替代自造题模块)
design:results-driven-video-split - 训练池 Maintenance 正确题自动补入机制
design:2026-07-14-maintenance-pool-design - 训练池 Maintenance 正确题自动补入机制
design:maintenance-pool - 论文主图:Self-Evolving Search Agent 推理训练闭环
design:paper-main-figure - 赛题生成工具设计
design:question-gen-synth - 赛题生成工具设计(Question Generation Synthesis)
design:2026-07-09-question-gen-synth-design
finding (10)
- 2026-07-11-benchmark-failure-taxonomy
finding:2026-07-11-benchmark-failure-taxonomy - 2026-07-11-question-gen-calibration-analysis
finding:2026-07-11-question-gen-calibration-analysis - 2026-07-14-ar30-too-easy-analysis
finding:2026-07-14-ar30-too-easy-analysis - 2026-07-14-hard-distractor-literature
finding:2026-07-14-hard-distractor-literature - 2026-07-14-question-quality-gap-analysis
finding:2026-07-14-question-quality-gap-analysis - 2026-07-15-question-gen-v2-diagnosis-and-strategy
finding:2026-07-15-question-gen-v2-diagnosis-and-strategy - Harness 评估: Spec-1 修复验证 (infer_spec1check)
finding:eval-spec1check - Harness 评估: Spec-2 批量并行建树
finding:eval-spec2-batch-tree-build - v3 §9 帧感知抽取机制 — 小样本实证验证结果
finding:2026-07-15-v3-frame-perception-spike-validation - 出题范式转变 — 从"生成-打分-过滤"转向"构造优先 + 两正交独立信号"(六篇原文深读)
finding:2026-07-15-question-gen-paradigm-shift-construction-over-filtering
plan (40)
- 2026-07-06-core-agent-adapters-llm
plan:2026-07-06-core-agent-adapters-llm - 2026-07-07-app-harness
plan:2026-07-07-app-harness - 2026-07-07-core-evolution
plan:2026-07-07-core-evolution - 2026-07-07-question-gen
plan:2026-07-07-question-gen - 2026-07-07-tree-module-vertical-slice
plan:2026-07-07-tree-module-vertical-slice - 2026-07-09-main-inference-entry
plan:2026-07-09-main-inference-entry - 2026-07-09-question-gen-synth
plan:2026-07-09-question-gen-synth - 2026-07-09-tree-repair-resilience
plan:2026-07-09-tree-repair-resilience - 2026-07-11-agent-runtime-fixes
plan:2026-07-11-agent-runtime-fixes - 2026-07-11-batch-tree-build
plan:2026-07-11-batch-tree-build - 2026-07-12-per-category-pool-strategy
plan:2026-07-12-per-category-pool-strategy - 2026-07-14-action-recognition-strategy
plan:2026-07-14-action-recognition-strategy - 2026-07-14-action-recognition-training
plan:2026-07-14-action-recognition-training - 2026-07-14-adversarial-question-gen-phaseB-plan
plan:2026-07-14-adversarial-question-gen-phaseB-plan - 2026-07-14-grounded-question-gen-phaseA-plan
plan:2026-07-14-grounded-question-gen-phaseA-plan - 2026-07-14-maintenance-pool
plan:2026-07-14-maintenance-pool - 2026-07-14-task-type-strategy-framework
plan:2026-07-14-task-type-strategy-framework - 2026-07-15-question-gen-v3-phase1-contract
plan:2026-07-15-question-gen-v3-phase1-contract - 2026-07-15-results-driven-video-split
plan:2026-07-15-results-driven-video-split - Action Recognition 单题型首次训练实验计划
plan:action-recognition-training - ActionRecognitionStrategy 特化实现计划 (Plan B)
plan:action-recognition-strategy - Adversarial Question-Gen Phase B
plan:adversarial-question-gen-phaseB - app/harness/ 训练循环编排层实现计划
plan:app-harness - app/search/ 搜索 Agent 装配层实现计划
plan:2026-07-07-search-module - core/agent/ + adapters/llm 基础设施实现计划
plan:core-agent-adapters-llm - Grounded Question-Gen Phase A
plan:grounded-question-gen-phaseA - main.py 推理入口 + 初始 Prompt 集实现计划
plan:main-inference-entry - Maintenance Pool 自动补入实现计划
plan:maintenance-pool - Per-Category Pool Strategy 实现计划
plan:per-category-pool-strategy - question_gen 模块实现计划
plan:question-gen - Spec-1 Agent 执行环境修复实现计划
plan:agent-runtime-fixes-plan - Spec-2 建树批量并行实现计划
plan:batch-tree-build-plan - TaskTypeStrategy 框架实现计划 (Plan A)
plan:task-type-strategy-framework - 出题管线 v2 实现计划
plan:2026-07-11-question-gen-v2 - 实现计划: 修复诊断 tree_data 断链 bug
plan:fix-diagnosis-tree-data-link-plan - 建树修复管线三项改造实现计划
plan:tree-repair-resilience - 建树模块竖切实现计划
plan:tree-module-vertical-slice - 结果驱动的视频级切分实现计划
plan:results-driven-video-split-plan - 赛题生成工具实现计划
plan:question-gen-synth - 项目基础设施初始化计划
plan:infrastructure-setup
review (1)
- question-gen v2 设计对抗审核 — 六路独立核验(四层病灶闭合度 + 契约一致性)
review:2026-07-15-question-gen-v2-adversarial-audit
schema (6)
- 表结构 v3 出题日志/观测(unit_verdict / collapse_metrics / quarantine / facts / resume)
schema:v3-question-gen-logging - 表结构: adversarial_verdicts(Phase B agent 门判定)
schema:adversarial-verdicts - 表结构: baseline_diagnosis(逐题诊断信号)
schema:baseline-diagnosis - 表结构: question_gen_items(逐题门判定)
schema:question-gen-items - 表结构: question_gen_runs(出题批次)
schema:question-gen-runs - 表结构: split_manifest(切分冻结溯源)
schema:split-manifest
metric (7)
- 出题四门总拦截率
metric:qgen-gate-rejection-rate - 指标: 48格(task_type×error_type)多样性覆盖率
metric:split-cell-coverage - 指标: floor 达标率 + test 代表性偏差
metric:split-floor-and-representativeness - 指标: 诊断信号 tier 分布(T0/T1/T2/uncertain 占比)
metric:split-signal-tier-distribution - 最终接受率
metric:qgen-acceptance-rate - 答案位置分布均匀性
metric:qgen-answer-uniformity - 重出收敛率(3轮内过门)
metric:qgen-regen-convergence