kjh2064/QuantEngineByItz
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
f5b202fb5e4d7d41409d6dd8e459fc34c3f2000d
QuantEngineByItz/spec/formulas/domains/risk.yaml
T
kjh2064 ee3e799de1 feat: 리밸런싱 엔진 V1 + GAS 버그 수정 (2026-06-13) 
주요 변경:
- tools/build_rebalance_engine_v1.py: REBALANCE_ENGINE_V1 신규
  * account_snapshot 직접 합산(_build_snap_position_map) → 소수주 분리 행 병합
  * 레짐 소스 macro.REGIME_PRELIM 최우선 (GAS 와 동일)
- src/gas_adapter_parts/gdf_06_rebalance.gs: runRebalanceSheet_() 신규
  * Logger.log / getSpreadsheet_() 로 run_all 연동 수정
- src/gas_adapter_parts/gdc_01_fetch_fundamentals.gs
  * _mergePositionRecord_(): 소수주 중복 행 합산 신규
  * parseInt → parseFloat (qty, availQty)
- src/gas_adapter_parts/gdf_01_price_metrics.gs
  * 미보유 종목 SELL_READY → WATCH_EXIT_SIGNAL
- spec/41_release_dag.yaml: build_rebalance_sheet 노드 추가 (step_count 63)
- spec/51_formula_lifecycle_registry.yaml: REBALANCE_ENGINE_V1 등록

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-06-13 13:20:14 +09:00

1621 lines
53 KiB
YAML
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
schema_version: formula_domain.v1
source: C:\Temp\data_feed\spec\13_formula_registry.yaml
domain: risk
formulas:
PEG_SCORE_V1:
purpose: 코스닥 종목의 ForwardPER을 EPS 3개년 성장률로 나눠 밸류에이션 타당성 판정. 고PER이라도 고성장이 뒷받침되면
허용.
applicable: 코스닥 상장 종목에만 실행. KOSPI 종목은 이 공식 미적용.
inputs:
- field: forward_pe
unit: ratio
source: spec/12_field_dictionary.yaml:field_dictionary.forward_pe
- field: eps_growth_3y_cagr_pct
unit: percent
source: '컨센서스 3개년 EPS CAGR (예: 30% → 30 입력)'
- field: sector_median_forward_pe
unit: ratio
source: spec/12_field_dictionary.yaml:field_dictionary.sector_median_forward_pe
derived_fields:
peg: forward_pe / eps_growth_3y_cagr_pct
rules:
- if: peg <= 1.5
result: PASS
valuation_gate: OK
quantity_modifier: 1.0
- if: 1.5 < peg <= 2.5
result: CAUTION
valuation_gate: CAUTION
quantity_modifier: 0.7
- if: peg > 2.5
result: REJECT
valuation_gate: REJECT
quantity_modifier: 0.0
fallback:
condition: eps_growth_3y_cagr_pct == DATA_MISSING OR eps_growth_3y_cagr_pct
<= 0
rules:
- if: forward_pe <= sector_median_forward_pe * 2.0
result: PASS
quantity_modifier: 1.0
- if: forward_pe <= sector_median_forward_pe * 3.0
result: CAUTION
quantity_modifier: 0.7
- if: forward_pe > sector_median_forward_pe * 3.0
result: REJECT
quantity_modifier: 0.0
prohibition: EPS 성장률 추정·보간으로 PEG 계산 금지 — 확정 컨센서스 없으면 fallback만 허용
output:
field: peg_gate_result
unit: enum [PASS, CAUTION, REJECT]
required_fields:
- peg
- peg_gate_result
- quantity_modifier
- valuation_gate
missing_policy:
forward_pe: fallback 규칙 적용. DATA_MISSING 표기.
sector_median_forward_pe: fallback 분자 기준 미산출 → CAUTION 보수 처리.
canonical_ref: spec/strategy/stock_model.yaml:stock_model.kosdaq_valuation_gate
owner: quant_team
lifecycle_state: active
input_fields:
- forward_pe
- eps_growth_3y_cagr_pct
- sector_median_forward_pe
output_fields:
- peg_gate_result
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
FINANCIAL_HEALTH_SCORE_V1:
purpose: 'ROE·영업이익률·부채비율·FCF를 결합해 종목의 재무 건전성을 0~20점으로 정량화. 수급·모멘텀 중심 편향을 보완하는
펀더멘털 축. 연간 기준 재무 데이터 사용.
'
inputs:
- field: roe_pct
unit: percent
optional: true
- field: operating_margin_pct
unit: percent
optional: true
- field: debt_to_equity
unit: ratio
optional: true
- field: fcf_b
unit: KRW_100M
optional: true
- field: sector_type
unit: enum
optional: true
note: 금융업(financial) 여부 — D/E 스코어링 건너뜀 판단
components:
profitability:
max_points: 8
source_field: roe_pct
rules:
- if: roe_pct >= 15
points: 8
- if: 10 <= roe_pct < 15
points: 5
- if: 5 <= roe_pct < 10
points: 2
- if: 0 <= roe_pct < 5
points: 0
- if: roe_pct < 0
points: -5
label: 수익성_훼손_페널티
missing_rule: 4pt 중립 처리 (DATA_MISSING_PROFITABILITY 태그)
operating_efficiency:
max_points: 7
source_field: operating_margin_pct
rules:
- if: operating_margin_pct >= 20
points: 7
- if: 10 <= operating_margin_pct < 20
points: 4
- if: 0 <= operating_margin_pct < 10
points: 2
- if: operating_margin_pct < 0
points: 0
label: 영업적자_HF007_발동
missing_rule: 3pt 중립 처리
financial_stability:
max_points: 5
source_field: debt_to_equity
financial_sector_skip: true
rules:
- if: debt_to_equity < 50
points: 5
- if: 50 <= debt_to_equity < 100
points: 3
- if: 100 <= debt_to_equity < 200
points: 1
- if: 200 <= debt_to_equity <= 400
points: 0
- if: debt_to_equity > 400
points: 0
label: 극단_부채_HF008_발동
missing_rule: 2pt 중립 처리
cash_generation:
max_points: 5
source_field: fcf_b
rules:
- if: fcf_b > 0
points: 5
label: 현금_창출
- if: fcf_b <= 0
points: 0
label: 현금_소각_또는_부재
missing_rule: 2pt 중립 처리
expression: "clamp(\n profitability_pts + operating_efficiency_pts +\n financial_stability_pts\
\ + cash_generation_pts,\n min=-5, max=20\n)\n"
output:
field: financial_health_score
unit: points_neg5_to_20
score_interpretation:
18_to_20: 재무 최우량 — ROE 높고 부채 낮고 FCF 창출
12_to_17: 재무 양호
6_to_11: 재무 보통 — 일부 약점 존재
0_to_5: 재무 취약 — 수급 강세여도 진입 신중
negative: 재무 훼손 — 영업적자 또는 ROE 음수. 수급 점수 불문 등급 하향 압력
missing_policy:
all_inputs_missing: 'financial_health_score = 8pt (전체 중립). DATA_MISSING_FHS
태그 필수. 재무 데이터 미제공 자체를 패널티로 처리하지 않음. 단, 코스닥 종목은 all_missing 시 6pt (더 보수적) 적용.
'
partial_missing: 각 컴포넌트별 missing_rule 적용 후 합산
sector_exception:
financial_sector:
definition: 은행·보험·증권·카드·캐피탈·리츠 등 금융업 SIC 분류
treatment: 'debt_to_equity 컴포넌트 건너뜀. financial_stability_pts = 3pt 기본값. ROE·Operating_Margin·FCF
컴포넌트는 동일 적용.
'
canonical_ref: spec/08_scoring_rules.yaml:strategy_score.financial_health
version: 2026-05-18_FINANCIAL_HEALTH_V1
owner: quant_team
lifecycle_state: active
input_fields:
- roe_pct
- operating_margin_pct
- debt_to_equity
- fcf_b
- sector_type
output_fields:
- financial_health_score
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
RS_MOMENTUM_V1:
purpose: 상대강도(RS)와 수급 가속도를 측정하여 상투 진입 방지 및 후발주(Laggard) 조기 식별
inputs:
- field: close_price
unit: KRW_per_share
- field: ma20
unit: KRW_per_share
- field: avg_trade_value_5d
unit: KRW
- field: avg_trade_value_20d
unit: KRW
- field: relative_strength_1m_percentile
unit: percentile
derived_fields:
disparity_20d: close_price / ma20
momentum_acceleration: avg_trade_value_5d / avg_trade_value_20d
rules:
- if: disparity_20d > 1.15
action: TRIGGER_HF009_BLOCK
label: 이격도_과열(상투)
- if: momentum_acceleration < 0.8 AND close_price > ma20
action: TRIM_WARNING
label: 수급_가속도_둔화(설거지_경계)
- if: relative_strength_1m_percentile > 70
action: LAGGARD_SELL_PRIORITY_1
label: 상대강도_최하위(우선감축)
output:
field: alpha_shield_status
unit: enum
owner: quant_team
lifecycle_state: active
input_fields:
- close_price
- ma20
- avg_trade_value_5d
- avg_trade_value_20d
- relative_strength_1m_percentile
output_fields:
- alpha_shield_status
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
OVERSOLD_DELAY_V1:
purpose: 현금 확보 시 '지하실 매도(패닉 셀)' 방지를 위한 데드캣 바운스 대기 알고리즘
inputs:
- field: rsi_14
unit: points
optional: true
- field: current_price
unit: KRW_per_share
- field: cash_shortfall_krw
unit: KRW
rules:
- condition: rsi_14 < 30 AND cash_shortfall_krw > 0
action: 전량 시장가 매도 금지. 25%만 TRIM 실행하고 잔여 수량은 단기 반등(+3% 이상) 시점까지 매도 유예.
label: 과매도_분할탈출(Staged Exit)
output:
field: oversold_exit_strategy
unit: string
owner: quant_team
lifecycle_state: active
input_fields:
- rsi_14
- current_price
- cash_shortfall_krw
output_fields:
- oversold_exit_strategy
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
MEAN_REVERSION_GATE_V1:
purpose: '주가가 MA20 대비 과도하게 상승하면 신규 매수를 하드 블록한다. 에너지 분산(Distribution) 구간에서의 추격
매수(상투 잡기)를 원천 봉쇄.
'
applicable: 매수 주문 생성 전 항상 실행.
inputs:
- field: close_price
unit: KRW_per_share
- field: ma20
unit: KRW_per_share
expression: deviation_ratio = close_price / ma20
output:
field: deviation_ratio
unit: ratio
gates:
- if: deviation_ratio >= 1.15
status: BUY_HARD_BLOCK
rule_id: MRG001
note: HF009_OVEREXTENSION_BLOCK 연동
- if: 1.10 <= deviation_ratio < 1.15
status: BUY_CAUTION
rule_id: MRG001_SOFT
note: 과열 접근 -- 신규 매수 강도 축소
- if: deviation_ratio < 1.10
status: PASS
missing_policy:
ma20: MRG001 DATA_MISSING. MA20 데이터 필요. 매수 보류.
canonical_ref: spec/08_scoring_rules.yaml:hard_filters.HF009_OVEREXTENSION_BLOCK
version: 2026-05-19_ALPHA_SHIELD_V1
owner: quant_team
lifecycle_state: active
input_fields:
- close_price
- ma20
output_fields:
- deviation_ratio
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
ECP_RISK_SCALE_V1:
purpose: 총자산 곡선(Equity Curve) 모멘텀에 따른 리스크 예산 자동 조절
inputs:
- field: total_asset
unit: KRW
- field: total_asset_ma10
unit: KRW
note: 10일 자산 이동평균
rules:
- if: total_asset < total_asset_ma10
action: RISK_BUDGET_HALVE
label: 자산곡선_역배열_방어
- if: total_asset >= total_asset_ma10
action: RISK_BUDGET_NORMAL
label: 자산곡선_정배열_정상
output:
field: equity_curve_status
unit: enum
owner: quant_team
lifecycle_state: active
input_fields:
- total_asset
- total_asset_ma10
output_fields:
- equity_curve_status
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
COMPOSITE_VERDICT_V1:
purpose: 'SS001 등급(A/B/C/D)과 rs_verdict(LEADER/MARKET/LAGGARD/BROKEN/UNKNOWN)를
결합해 종목의 최종 포지션 판정을 내린다. LLM이 "좋아 보인다"류 언어적 판단으로 판정을 변경하는 것을 구조적으로 방지한다.
'
applicable: _addTickerGates_ 내에서 SS001 계산 직후 실행.
inputs:
- field: ss001_grade
unit: enum [A,B,C,D]
- field: rs_verdict
unit: enum [LEADER,MARKET,LAGGARD,BROKEN,UNKNOWN]
matrix: 'GRADE LEADER MARKET LAGGARD BROKEN UNKNOWN
A PRIME_CANDIDATE PRIME_CANDIDATE WATCH_CANDIDATE EXIT_REVIEW WATCH_CANDIDATE
B PRIME_CANDIDATE WATCH_CANDIDATE REDUCE_CANDIDATE EXIT_REVIEW WATCH_CANDIDATE
C WATCH_CANDIDATE REDUCE_CANDIDATE REDUCE_CANDIDATE CLOSE_POSITION REDUCE_CANDIDATE
D REDUCE_CANDIDATE REDUCE_CANDIDATE CLOSE_POSITION CLOSE_POSITION REDUCE_CANDIDATE
'
output:
field: composite_verdict
unit: enum [PRIME_CANDIDATE, WATCH_CANDIDATE, REDUCE_CANDIDATE, EXIT_REVIEW,
CLOSE_POSITION]
action_guidance:
PRIME_CANDIDATE: 코어 유지·추가. RAG_V1 PASS 시 위성 추가 허용.
WATCH_CANDIDATE: 현재 비중 유지. 추가매수 보류.
REDUCE_CANDIDATE: 분할 축소. 5% 이상 비중이면 TRIM 실행.
EXIT_REVIEW: 다음 반등 시 전량 매도 준비.
CLOSE_POSITION: 즉시 정리. calcFinalDecision_ SELL 강제 검토.
missing_policy:
ss001_grade: composite_verdict = WATCH_CANDIDATE (보수적)
rs_verdict: UNKNOWN 컬럼으로 처리
ground_truth: harness
llm_allowed: cite_only
prohibition:
- composite_verdict = CLOSE_POSITION인 종목을 LLM이 'HOLD도 무방'으로 서술 금지
- composite_verdict 없이 종목 정리 우선순위 결정 금지
canonical_ref: spec/13_formula_registry.yaml:RS_VERDICT_V1
version: 2026-05-21_CLA_HARNESS_V1
owner: quant_team
lifecycle_state: active
input_fields:
- ss001_grade
- rs_verdict
output_fields:
- composite_verdict
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
BENCHMARK_RELATIVE_TIMESERIES_V1:
purpose: '종목을 KOSPI 기준 시계열로 평가해 초과낙폭, 반등 회복률, 하락장 베타, RS선 기울기와 brt_verdict를 결정론적으로
산출한다. LLM은 값 인용만 가능하다.
'
applicable: _addTickerGates_에서 RS_VERDICT_V2 이전 실행. 과거 가격 배열이 없으면 RET20/RET60
기반 프록시임을 method에 남긴다.
inputs:
- field: price.ret5D
unit: pct
- field: price.ret20D
unit: pct
- field: price.ret60D
unit: pct
- field: price.close
unit: KRW_per_share
- field: high52w
unit: KRW_per_share
optional: true
- field: globalKospiRet5D_
unit: pct
- field: globalKospiRet20D_
unit: pct
- field: globalKospiRet60D_
unit: pct
- field: globalKospiDrawdown_
unit: pct
derived:
stock_drawdown_from_high_pct: if high52W>0 then max(0,(1-close/high52W)*100)
else null
excess_drawdown_pctp: stock_drawdown_from_high_pct - globalKospiDrawdown_
recovery_ratio_5d: price.ret5D / globalKospiRet5D_ if globalKospiRet5D_ > 0
else null
recovery_ratio_20d: price.ret20D / globalKospiRet20D_ if globalKospiRet20D_
> 0 else null
downside_beta: price.ret20D / globalKospiRet20D_ if globalKospiRet20D_ < 0 else
null
rs_ratio_5d: price.ret5D / globalKospiRet5D_ if globalKospiRet5D_ != 0 else
null
rs_ratio_20d: price.ret20D / globalKospiRet20D_ if globalKospiRet20D_ != 0 else
null
rs_ratio_60d: price.ret60D / globalKospiRet60D_ if globalKospiRet60D_ != 0 else
null
rs_line_20d_slope: (rs_ratio_20d - rs_ratio_5d) / 15 if both available else
(ret20D - k20) / 20
rs_line_60d_slope: (rs_ratio_60d - rs_ratio_20d) / 40 if both available else
(ret60D - k60) / 60
brt_method: RS_RATIO_MULTI_WINDOW_PROXY if rs_ratio_5d and rs_ratio_20d available
else PROXY_FROM_RET20_RET60
verdict_table:
LEADER: excess_drawdown_pctp <= 0 AND recovery_ratio_20d >= 1.20 AND rs_line_20d_slope
> 0
MARKET: excess_drawdown_pctp between -5 and 5 AND recovery_ratio_20d between
0.80 and 1.20
LAGGARD: excess_drawdown_pctp >= 5 OR recovery_ratio_20d < 0.80 OR rs_line_20d_slope
< 0
BROKEN: excess_drawdown_pctp >= 10 AND (recovery_ratio_20d < 0.50 OR rs_line_60d_slope
< 0)
output:
fields:
- stock_drawdown_from_high_pct
- excess_drawdown_pctp
- recovery_ratio_5d
- recovery_ratio_20d
- downside_beta
- rs_ratio_5d
- rs_ratio_20d
- rs_ratio_60d
- rs_line_20d_slope
- rs_line_60d_slope
- brt_verdict
- brt_method
missing_policy: 필수 KOSPI 또는 종목 수익률 누락 시 brt_verdict=UNKNOWN, LLM 대체 계산 금지.
ground_truth: harness
llm_allowed: cite_only
prohibition:
- LLM이 excess_drawdown/recovery_ratio/downside_beta/rs_slope를 직접 계산 금지
- brt_verdict=BROKEN 종목을 낙폭과대 매수 기회로 서술 금지
version: 2026-05-21_BRT_V1_C2
owner: quant_team
lifecycle_state: active
input_fields:
- price.ret5D
- price.ret20D
- price.ret60D
- price.close
- high52w
- globalKospiRet5D_
- globalKospiRet20D_
- globalKospiRet60D_
- globalKospiDrawdown_
output_fields: []
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
RS_VERDICT_V2:
purpose: RS_VERDICT_V1과 BENCHMARK_RELATIVE_TIMESERIES_V1의 brt_verdict를 보수적으로 융합한다.
applicable: BRT_V1 직후. 기존 rs_verdict 필드명은 최종 V2 결과로 유지하고 rs_verdict_v1_raw를 감사용
보존.
inputs:
- field: rs_verdict_v1_raw
unit: enum
- field: brt_verdict
unit: enum [LEADER,MARKET,LAGGARD,BROKEN,UNKNOWN]
fusion_logic:
BROKEN: rs_verdict_v1_raw=BROKEN OR brt_verdict=BROKEN
LAGGARD: rs_verdict_v1_raw=LAGGARD OR brt_verdict=LAGGARD, 단 BROKEN 아님
LEADER: rs_verdict_v1_raw=LEADER AND brt_verdict=LEADER
MARKET: 그 외
special_cases:
- brt_verdict=LEADER AND rs_verdict_v1_raw=LAGGARD -> MARKET
- brt_verdict=BROKEN AND rs_verdict_v1_raw=LEADER -> LAGGARD
output:
field: rs_verdict
additional_fields:
- rs_verdict_v1_raw
- rs_verdict_source
ground_truth: harness
llm_allowed: cite_only
version: 2026-05-21_RS_VERDICT_V2
owner: quant_team
lifecycle_state: active
input_fields:
- rs_verdict_v1_raw
- brt_verdict
output_fields:
- rs_verdict
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
SATELLITE_AGGREGATE_PNL_GATE_V1:
purpose: 위성 합산 평가손익이 코어 수익을 잠식하는 정도를 추적해 위성 전략 실패를 감지한다.
inputs:
- field: position_class
unit: enum [core,satellite]
- field: profit_loss
unit: KRW
computed:
core_total_pnl_krw: sum(profit_loss for core)
satellite_total_pnl_krw: sum(profit_loss for satellite)
satellite_loss_to_core_gain_ratio: abs(min(0,satellite_total_pnl_krw)) / max(core_total_pnl_krw,1)
gates:
PASS: ratio < 0.25
SAPG_ALERT: 0.25 <= ratio < 0.50
SAPG_CRITICAL: ratio >= 0.50
output:
field: sapg_status
additional_fields:
- core_total_pnl_krw
- satellite_total_pnl_krw
- satellite_loss_to_core_gain_ratio
ground_truth: harness
llm_allowed: cite_only
version: 2026-05-21_SAPG_V1
owner: quant_team
lifecycle_state: active
input_fields:
- position_class
- profit_loss
output_fields:
- sapg_status
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
HARNESS_DATA_FRESHNESS_GATE_V1:
purpose: 'harness_context 주요 입력 데이터의 영업일 기준 신선도를 검증한다. 신선도 부족 시 BRT/SAPG/SAQG
결과의 신뢰도 등급을 자동 하향하고 STALE_BLOCK 상태에서 HTS 주문표 생성을 차단한다.
'
applicable: buildHarnessContext_ 가장 먼저 실행 (모든 공식 선행).
inputs:
- field: metadata.generated_at
unit: datetime
- field: metadata.market_date
unit: date
- field: today_date
unit: date
computed:
data_age_business_days: business_days_diff(today_date, metadata.market_date)
freshness_status:
FRESH:
condition: data_age_business_days <= 1
note: 정상
STALE_1D:
condition: data_age_business_days == 2
harness_impact: brt_method에 STALE_1D 태그
STALE_WARN:
condition: data_age_business_days IN [3,4]
harness_impact: brt_verdict LOW, ELIGIBLE -> WATCHLIST_ONLY
STALE_BLOCK:
condition: data_age_business_days >= 5
harness_impact: brt_verdict=DATA_STALE_BLOCKED, HTS 주문표 생성 금지
output:
field: data_freshness_status
additional_fields:
- data_age_business_days
- freshness_degraded_gates
ground_truth: harness
llm_allowed: cite_only
prohibition:
- STALE_BLOCK 상태에서 BRT 결과로 주문 생성 금지
- LLM이 신선도를 임의 판단하거나 사용 가능으로 완화 금지
version: 2026-05-21_HDFG_V1
owner: quant_team
lifecycle_state: active
input_fields:
- metadata.generated_at
- metadata.market_date
- today_date
output_fields:
- data_freshness_status
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
INTRADAY_ACTION_MATRIX_V1:
purpose: '장중 시각(capture_time)에 따라 허용·금지 액션을 테이블로 고정한다. 09:31 캡처임에도 전체 주간 전략 출력(E4
오류) 재발 방지.
'
applicable: DETERMINISTIC_ROUTING_ENGINE_V1 STAGE 1. capture_time 확정 직후.
inputs:
- field: capture_time
unit: HH:MM
note: account_snapshot에서 자동 추출
- field: market_date
unit: date
time_slot_table:
09:00-09:30:
label: OPEN_AUCTION
allowed_actions:
- WATCH_ONLY
blocked_actions:
- BUY
- SELL
- TRIM
09:30-10:30:
label: EARLY_SESSION
allowed_actions:
- TRIM_ONLY
blocked_actions:
- BUY
- SELL_ALL
note: OVERSOLD_REBOUND_SELL 제외
10:30-14:00:
label: MID_SESSION
allowed_actions:
- TRIM
- STAGED_SELL
- WATCH
blocked_actions:
- BUY_NEW
- SELL_ALL
14:00-15:00:
label: LATE_SESSION
allowed_actions:
- TRIM
- STAGED_SELL
blocked_actions:
- BUY_NEW
- SELL_ALL
15:00-15:20:
label: PRE_CLOSE
allowed_actions:
- TRIM_ONLY
blocked_actions:
- BUY
- SELL_ALL
15:20-15:30:
label: CLOSE_VERIFY
allowed_actions:
- ALL_ACTIONS_ALLOWED
blocked_actions: []
note: 종가 근처 재검증 후 실행. 모든 액션 허용.
15:30+:
label: POST_MARKET
allowed_actions:
- REBALANCING_REVIEW
- NEXT_DAY_PLAYBOOK
blocked_actions:
- HTS_IMMEDIATE_EXECUTION
BEFORE_MARKET:
label: PRE_MARKET
allowed_actions:
- PLAYBOOK_DRAFT
blocked_actions:
- HTS_IMMEDIATE_EXECUTION
output:
fields:
- field: time_slot_label
unit: enum
- field: allowed_intraday_actions
unit: list
- field: blocked_intraday_actions
unit: list
ground_truth: harness
llm_allowed: cite_only
prohibition:
- allowed_intraday_actions[]에 없는 주문 유형을 보고서에 생성 금지
- 09:00-10:30 캡처 시 BUY/SELL_ALL 주문표 생성 금지
- TRIM_ONLY 구간에서 주간 전략·종가 예측 등 전체 보고서 출력 금지 — 해당 섹션은 '15:20 재실행 예정' 표기
canonical_ref: AGENTS.md:Direction A3, Direction 0(장중 제약)
version: 2026-05-22_3RD_HARNESS
owner: quant_team
lifecycle_state: active
input_fields:
- capture_time
- market_date
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
ANTI_CHASING_VELOCITY_V1:
purpose: '가격 상승 속도가 국면별 임계값 초과 시 BUY를 결정론적으로 차단한다. N2(거래량 확인)만으로는 막지 못하는 속도 기반
뒷박 매수 원천 차단.
'
applicable: _addTickerGates_ 내 alpha_lead_score 산출 직후.
inputs:
- field: close
unit: KRW_per_share
- field: close_1d_ago
unit: KRW_per_share
- field: close_5d_ago
unit: KRW_per_share
- field: market_regime
unit: enum
computed:
velocity_1d: (close - close_1d_ago) / close_1d_ago * 100
velocity_5d: (close - close_5d_ago) / close_5d_ago * 100
thresholds_by_regime:
EVENT_SHOCK:
v1d_max: 2
v5d_max: 5
verdict: BLOCK_CHASE_SHOCK
RISK_OFF:
v1d_max: 3
v5d_max: 7
verdict: BLOCK_CHASE_RISKOFF
NEUTRAL:
v1d_max: 4
v5d_max: 9
verdict: WARN_CHASE_NEUTRAL
RISK_ON:
v1d_max: 5
v5d_max: 12
verdict: WARN_CHASE_RISKON
CLA:
v1d_max: 6
v5d_max: 15
verdict: WARN_CHASE_CLA
actions:
BLOCK_CHASE: BUY 금지. alpha_lead_score에 -15 페널티 적용.
WARN_CHASE: BUY 주문표에 '속도 추격 경고 — 풀백 대기 권고' 표기.
CLEAR: 정상. 속도 기반 차단 없음.
output:
field: anti_chasing_velocity_status
values:
- BLOCK_CHASE_SHOCK
- BLOCK_CHASE_RISKOFF
- WARN_CHASE_NEUTRAL
- WARN_CHASE_RISKON
- WARN_CHASE_CLA
- CLEAR
additional_fields:
- velocity_1d
- velocity_5d
- velocity_penalty_applied
ground_truth: harness
llm_allowed: cite_only
prohibition:
- BLOCK_CHASE 상태에서 '분위기 좋으니 추격 매수 괜찮다' 우회 서술 금지
- LLM이 velocity 직접 계산 금지 — 하네스 산출값만 인용
canonical_ref: AGENTS.md:Direction B1
version: 2026-05-22_3RD_HARNESS
owner: quant_team
lifecycle_state: active
input_fields:
- close
- close_1d_ago
- close_5d_ago
- market_regime
output_fields:
- anti_chasing_velocity_status
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
DETERMINISTIC_ROUTING_ENGINE_V1:
purpose: '"LLM이 먼저 판단 → 하네스가 검증" 구조를 역전. 9단계 라우팅을 고정 순서로 실행하고 LLM은 최종 결과의 보고관으로만
동작. 라우팅 단계 건너뜀 및 순서 변경 절대 금지.
'
applicable: 모든 분석 보고서 생성 최선행. STAGE 0부터 순서대로 실행.
inputs:
- field: harness_context
unit: json
note: 전체 하네스 컨텍스트 — 모든 STAGE 결과 포함
output:
field: routing_execution_log
schema:
- stage: int
status: PASS|BLOCKED|SKIPPED
output_key: string
elapsed_ms: int
routing_stages:
STAGE_0:
name: HARNESS_DATA_FRESHNESS_GATE_V1
action: STALE_BLOCK → 즉시 중단. 보고서 생성 금지. 데이터 갱신 요청만 출력.
STAGE_1:
name: INTRADAY_ACTION_MATRIX_V1
action: capture_time 기반 허용·금지 액션 테이블 확정. TRIM_ONLY 구간이면 간소화 모드.
STAGE_2:
name: PORTFOLIO_HEALTH_SCORE_V1
action: CRITICAL → 긴급 섹션 필수 출력 후 진행.
STAGE_3:
name: RISK_GATE_CHECKLIST (10개 순서 고정)
gates:
- cash_floor_status
- heat_gate_status
- drawdown_guard_state
- portfolio_drawdown_gate
- portfolio_beta_gate+PORTFOLIO_CORRELATION_GATE
- sector_concentration_gate
- semiconductor_cluster_gate
- position_count_gate
- win_loss_streak_state
- single_position_weight_gate
action: BLOCK 있으면 blocked_actions[] 업데이트.
STAGE_4:
name: SELL_GATE_CHECKLIST (5개)
gates:
- stop_breach_gate
- tp_trigger_gate
- DISTRIBUTION_SELL_DETECTOR_V1
- heat_concentration_gate
- regime_transition_type
action: sell_priority_decision_table 생성 (독립 표 필수).
STAGE_5:
name: CASH_RECOVERY (OPTIMIZER + WATERFALL)
action: CASH_RECOVERY_OPTIMIZER_V1 → SELL_WATERFALL_ENGINE_V1 → SELL_PRICE_SANITY_V1.
STAGE_6:
name: BUY_SCREENING (6개 순서 고정)
gates:
- ANTI_CHASING_VELOCITY_V1
- PULLBACK_ENTRY_TRIGGER_V1
- N2_VOLUME_BREAKOUT
- K1_TRANCHE
- RAG+SFG+SATELLITE_LIFECYCLE
- SECTOR_ROTATION+PRE_DISTRIBUTION
action: buy_candidates_json 확정.
STAGE_7:
name: QUANTITY_FINALIZATION
action: ATR20 기반 atr_qty → regime_size_scale × drawdown_buy_scale → TP_QUANTITY_LADDER
→ SELL_PRICE_SANITY 재검증.
STAGE_8:
name: SHADOW_LEDGER_SEPARATION
action: PASS → HTS 주문표. 비PASS → Shadow Ledger (I4 컬럼명 규칙).
STAGE_9:
name: REPORT_ASSEMBLY (LLM)
action: 하네스 결과를 정해진 양식으로 서술. 숫자 1원·1주 변경 금지. LLM_SERVING_CONSTRAINT_V1 적용.
ground_truth: harness
llm_allowed: STAGE_9 보고서 작성만 허용
prohibition:
- STAGE 건너뜀 금지
- STAGE_0 STALE_BLOCK 시 전체 보고서 생성 금지
- STAGE_9에서 LLM이 숫자 변경 금지 — 보고관(Clerk)으로만 동작
canonical_ref: AGENTS.md:Direction D1, Direction Q(QEH)
version: 2026-05-22_3RD_HARNESS
owner: quant_team
lifecycle_state: active
input_fields:
- harness_context
output_fields:
- routing_execution_log
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
LLM_SERVING_CONSTRAINT_V1:
purpose: 'LLM이 보고서 작성 시 침범 금지 영역 8개를 명시적으로 잠금. HS011 확장판. 30년 실전 전문가의 정밀도는 자유로운
해석이 아닌 규칙 준수에서 나온다.
'
applicable: DETERMINISTIC_ROUTING_ENGINE_V1 STAGE_9 진입 직전 검사.
inputs:
- field: harness_context
unit: json
note: LLM 보고서 생성 전 전체 컨텍스트 검사
output:
field: serving_constraint_check
schema:
violations_detected: list [FB_code]
allowed_actions_taken: list [AL_code]
forbidden_actions:
FB1: 가격·수량 즉석 계산 → DATA_MISSING 표기만 허용 (HS011)
FB2: 하네스 판정 '이번만 예외' 번복
FB3: '''분위기가 좋으니'' 류 감성 서술로 BLOCK 우회'
FB4: rule_id 없는 판단 서술 (근거 공식 ID 없이 결론 금지)
FB5: 매수와 매도를 같은 문단에 연결 서술 (BRT3 재투자 연결 금지)
FB6: 목표 달성률 압박으로 리스크 게이트 완화 서술 (M4 압박 금지)
FB7: 외부 웹 데이터로 prices_json 덮어쓰기 (G3 외부 격리)
FB8: 손절가·익절가 null인 종목에 '보유 유지' 단독 서술
allowed_roles:
AL1: 하네스 결과의 '왜 이 점수인가?' 배경 설명
AL2: 뉴스·이벤트·섹터 흐름 질적 리스크 합성
AL3: '''만약 반도체가 추가 하락하면...'' 시나리오 제시'
AL4: N4 HOLDING_STALE_REVIEW 연동 보유 근거 재확인
ground_truth: harness
llm_allowed: cite_only
prohibition:
- 8개 금지 영역은 어떤 조건에서도 침범 불가
- '위반 항목은 ''[LLM_SERVING_CONSTRAINT: FB{N} 위반]''으로 보고서에 표시'
canonical_ref: AGENTS.md:Direction D2, Direction Q(QEH), HS011
version: 2026-05-22_3RD_HARNESS
owner: quant_team
lifecycle_state: active
input_fields:
- harness_context
output_fields:
- serving_constraint_check
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
HORIZON_ALLOCATION_LOCK_V1:
purpose: 단기/중기/장기 투자 버킷별 비중 상한을 적용해 기간 혼재와 과집중을 차단.
inputs:
- field: invest_horizon
unit: enum
optional: true
- field: market_value_krw
unit: KRW
optional: true
- field: total_asset_krw
unit: KRW
output:
field: horizon_allocation_json
llm_allowed: cite_only
version: 2026-05-25_PROPOSAL53
owner: quant_team
lifecycle_state: active
input_fields:
- invest_horizon
- market_value_krw
- total_asset_krw
output_fields:
- horizon_allocation_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
FUNDAMENTAL_MULTI_FACTOR_SCORE_V2:
purpose: 이익률/성장률/점유율/현금흐름/부채를 종합 점수화해 매수 허용을 잠금.
inputs:
- field: roe_pct
unit: percent
optional: true
- field: opm_pct
unit: percent
optional: true
- field: revenue_growth_pct
unit: percent
optional: true
- field: op_income_growth_pct
unit: percent
optional: true
- field: market_share_proxy_pct
unit: percent
optional: true
- field: operating_cf_krw
unit: KRW
optional: true
- field: free_cf_krw
unit: KRW
optional: true
- field: debt_ratio_pct
unit: percent
optional: true
output:
field: fundamental_multifactor_json
llm_allowed: cite_only
version: 2026-05-25_PROPOSAL54
owner: quant_team
lifecycle_state: active
input_fields:
- roe_pct
- opm_pct
- revenue_growth_pct
- op_income_growth_pct
- market_share_proxy_pct
- operating_cf_krw
- free_cf_krw
- debt_ratio_pct
output_fields:
- fundamental_multifactor_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
MARKET_SHARE_MOMENTUM_PROXY_V1:
purpose: 상대 성장/RS 기반 점유율 모멘텀 프록시를 산출해 공격 매수 여부를 잠금.
inputs:
- field: revenue_growth_pct
unit: percent
optional: true
- field: alpha_lead_score
unit: score
optional: true
output:
field: market_share_proxy_json
llm_allowed: cite_only
version: 2026-05-25_PROPOSAL54
owner: quant_team
lifecycle_state: active
input_fields:
- revenue_growth_pct
- alpha_lead_score
output_fields:
- market_share_proxy_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
ROUTING_EXECUTION_LOG_TABLE_V1:
purpose: 'DETERMINISTIC_ROUTING_ENGINE_V1 11단계(①CV-V2사전검증 ②데이터신선도 ③장중판별 ④포트폴리오상태
⑤거시이벤트동기화 ⑥선제매도레이더 ⑦매수타이밍게이트 ⑧매도우선순위/현금확보 ⑨RS/위성품질 ⑩가격정규화/검증 ⑪LLM서빙)의 실행 로그를
표로 강제 출력한다. GAS 미보고 단계는 결정론 fallback으로 보강. 누락 단계 > 0이면 INCOMPLETE_ROUTING_LOG.
'
inputs:
- field: routing_execution_log
unit: json
- field: _harness_context
unit: json
output:
field: routing_execution_log_v1_json
expected_outputs:
- gate
- stage_coverage_pct
- request_route
llm_allowed: cite_only
version: 2026-05-27_PHASE1
owner: quant_team
lifecycle_state: active
input_fields:
- routing_execution_log
- _harness_context
output_fields:
- routing_execution_log_v1_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
FUNDAMENTAL_RAW_INGEST_V1:
purpose: 'data_feed(Forward_PE/PBR/EPS)와 네이버 금융 fallback을 통해 보유 종목의 펀더멘털 raw 지표를
수집하고 fundamental_raw_v1.json을 생성한다.
'
output:
field: fundamental_raw_v1_json
expected_outputs:
- gate
- coverage_pct
- non_etf_count
llm_allowed: cite_only
version: 2026-05-27_PHASE2
owner: quant_team
lifecycle_state: active
input_fields: []
output_fields:
- fundamental_raw_v1_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
FUNDAMENTAL_MULTIFACTOR_V3:
purpose: 'ROE(25) + OPM(20) + OCF(15) + FCF(15) + Debt(10) + Valuation(15) = 100점
6요소 결정론 공식으로 종목별 펀더멘털 등급을 산출한다. ETF는 별도 분류, 데이터 부족 시 보유 필드 기준 정규화 적용.
'
output:
field: fundamental_multifactor_v3_json
expected_outputs:
- gate
- grade_diverse
- non_etf_count
llm_allowed: cite_only
version: 2026-05-27_PHASE2
owner: quant_team
lifecycle_state: active
input_fields: []
output_fields:
- fundamental_multifactor_v3_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
HORIZON_CLASSIFICATION_V1:
purpose: '펀더멘털 등급 + 이격도 + ATR% + RSI14 기반으로 종목별 투자 기간을 분류한다. LONG/MID/SHORT/ETF/UNKNOWN
결정론 트리. HORIZON_ALLOCATION_LOCK_V1에 주입.
'
output:
field: horizon_classification_v1_json
expected_outputs:
- gate
- classified_pct
- allocation_pct
llm_allowed: cite_only
version: 2026-05-27_PHASE2
owner: quant_team
lifecycle_state: active
input_fields: []
output_fields:
- horizon_classification_v1_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
SMART_MONEY_FLOW_SIGNAL_V2:
purpose: 'Frg_5D/20D + Inst_5D/20D 백분위 기반으로 종목별 스마트머니 흐름을 산출한다. STRONG_INFLOW
/ INFLOW / NEUTRAL / OUTFLOW / STRONG_OUTFLOW 라벨 분산 강제.
'
output:
field: smart_money_flow_signal_v2_json
expected_outputs:
- gate
- label_diversity
- coefficient_of_variation
llm_allowed: cite_only
version: 2026-05-27_PHASE3
owner: quant_team
lifecycle_state: active
input_fields: []
output_fields:
- smart_money_flow_signal_v2_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
GROWTH_RATE_SIGNAL_V1:
purpose: 'EPS YoY / 매출 YoY 기반 성장률 시그널을 결정론적으로 산출한다. HYPER_GROWTH/GROWTH/FLAT/DECLINE/DATA_MISSING
라벨과 단/중/장기 horizon 적합도를 포함한다.
'
output:
field: growth_rate_signal_v1_json
expected_outputs:
- gate
- label_counts
- data_missing_pct
llm_allowed: cite_only
version: 2026-05-27_PHASE2B
owner: quant_team
lifecycle_state: active
input_fields: []
output_fields:
- growth_rate_signal_v1_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
MARKET_SHARE_SIGNAL_V2:
purpose: '실매출 점유율 데이터 없는 환경에서 AvgTradeValue_20D_M 백분위 + 외인/기관 수급 + 20일 모멘텀 3중
프록시로 GAINING/STABLE/LOSING/NO_PEER_DATA를 산출한다. confidence는 항상 LOW(proxy 기반).
실데이터 확보 시 HIGH confidence로 업그레이드 예정.
'
output:
field: market_share_signal_v2_json
expected_outputs:
- gate
- unique_states
- non_etf_scored_count
llm_allowed: cite_only
version: 2026-05-27_PHASE2B
owner: quant_team
lifecycle_state: active
input_fields: []
output_fields:
- market_share_signal_v2_json
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
PREDICTION_ACCURACY_HARNESS_V2:
purpose: '운영 T+1/T+5/T+20 일치율을 90/30/7일 회전 윈도로 산출. calibration_state: CALIBRATED/MONITOR/PAE_CALIBRATION_REQUIRED/BUY_PROPOSAL_FROZEN_RECOMMEND.
'
output:
file: Temp/prediction_accuracy_harness_v2.json
expected_outputs:
- calibration_state
- t5_op_rate
- t5_sample
- window_90d_rate
llm_allowed: cite_only
version: 2026-05-28_PHASE4
owner: quant_team
lifecycle_state: active
input_fields: []
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
VELOCITY_V1:
purpose: '1일/5일 가격 속도를 계산해 뒷박 추격 차단과 풀백 트리거의 입력으로 공급한다.
'
inputs:
- field: close_price
unit: KRW_per_share
- field: previous_close_price
unit: KRW_per_share
- field: ret5d
unit: percent
output:
field: velocity_1d
input_fields:
- close
- prevClose
- ret5d
expected_outputs:
- velocity_1d
- velocity_5d
llm_allowed: cite_only
version: 2026-05-30_PHASE8
owner: quant_team
lifecycle_state: active
output_fields:
- velocity_1d
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
PROFIT_LOCK_STAGE_V1:
purpose: '수익률 구간을 NORMAL/BREAKEVEN/PROFIT_LOCK/APEX 계열로 분류한다.
'
inputs:
- field: profit_pct
unit: percent
output:
field: profit_lock_stage
input_fields:
- profit_pct
expected_outputs:
- stage
llm_allowed: cite_only
version: 2026-05-30_PHASE8
owner: quant_team
lifecycle_state: active
output_fields:
- profit_lock_stage
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
SEMICONDUCTOR_CLUSTER_GATE_V1:
purpose: '반도체 클러스터 집중도와 국면별 차단/감축 여부를 판단한다.
'
inputs:
- field: semiconductor_cluster_json
unit: json
- field: market_regime
unit: enum
output:
field: semiconductor_cluster_gate
input_fields:
- combined_pct
- market_regime
expected_outputs:
- semiconductor_cluster_gate
- combined_pct
llm_allowed: cite_only
version: 2026-05-30_PHASE8
owner: quant_team
lifecycle_state: active
output_fields:
- semiconductor_cluster_gate
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
ANTI_WHIPSAW_GATE_V1:
purpose: '반등/조정 혼선 구간에서 설거지성 매수와 성급한 매도를 차단한다.
'
inputs:
- field: close_price
unit: KRW_per_share
- field: ma20
unit: KRW_per_share
- field: rsi14
unit: points
output:
field: anti_whipsaw_status
input_fields:
- close
- ma20
- rsi14
expected_outputs:
- anti_whipsaw_status
llm_allowed: cite_only
version: 2026-05-30_PHASE8
owner: quant_team
lifecycle_state: active
output_fields:
- anti_whipsaw_status
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
BREAKEVEN_RATCHET_V1:
purpose: '손익분기 이상 구간에서 손절선을 평단 이상으로 올리는 래칫을 산출한다.
'
inputs:
- field: average_cost
unit: KRW_per_share
- field: highest_price_since_entry
unit: KRW_per_share
output:
field: breakeven_stop_price
input_fields:
- average_cost
- highest_close
expected_outputs:
- breakeven_stop_price
llm_allowed: cite_only
version: 2026-05-30_PHASE8
owner: quant_team
lifecycle_state: active
output_fields:
- breakeven_stop_price
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
CAPITAL_STYLE_ALLOCATION_V1:
purpose: '투자성향별 자금 유동성/공격성 가중치와 conviction을 산출한다.
'
inputs:
- field: smart_money_flow_signal_v2_json
unit: json
- field: fundamental_multifactor_v3_json
unit: json
- field: macro_event_ticker_impact_v1_json
unit: json
- field: liquidity_flow_signal_v1_json
unit: json
output:
field: capital_style_conviction
input_fields:
- investor_style
- liquidity_profile
expected_outputs:
- capital_style_conviction
- capital_style_label
llm_allowed: cite_only
version: 2026-05-30_PHASE8
owner: quant_team
lifecycle_state: active
output_fields:
- capital_style_conviction
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
ARTIFACT_FRESHNESS_GATE_V1:
purpose: '하네스 산출물의 타임스탬프를 검증해 신선도 게이트를 산출한다.
'
input_fields:
- artifact_timestamp
- max_age_hours
expected_outputs:
- freshness_gate
- stale_artifacts
llm_allowed: cite_only
version: 2026-06-03_ORPHAN_RECONCILE
owner: quant_team
lifecycle_state: active
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
CANONICAL_ARTIFACT_RESOLVER_V1:
purpose: '동일 의미의 중복 산출물 중 유일 출처를 지정해 단일 진실원장을 고정한다.
'
input_fields:
- artifact_key
- candidate_paths
expected_outputs:
- canonical_path
- duplicate_artifacts
llm_allowed: cite_only
version: 2026-06-03_ORPHAN_RECONCILE
owner: quant_team
lifecycle_state: active
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
COMPLETION_GAP_V1:
purpose: 'pass_100 기준 대비 미충족 항목과 격차를 정량화해 완료 갭 보고서를 산출한다.
'
input_fields:
- pass_100_criteria
- current_metrics
expected_outputs:
- completion_gap_score
- failed_criteria_list
llm_allowed: cite_only
version: 2026-06-03_ORPHAN_RECONCILE
owner: quant_team
lifecycle_state: active
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
COMPREHENSIVE_PROPOSAL_V1:
purpose: '매수·매도·보유·현금확보 전 섹션을 통합한 종합 제안서를 생성한다.
'
input_fields:
- buy_proposals
- sell_proposals
- portfolio_state
expected_outputs:
- comprehensive_proposal
- proposal_id
llm_allowed: cite_only
version: 2026-06-03_ORPHAN_RECONCILE
owner: quant_team
lifecycle_state: active
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
DATA_INTEGRITY_100_LOCK_V1:
purpose: '핵심 데이터 필드의 정합성을 검증해 100% 잠금 게이트를 산출한다 (V2로 대체됨).
'
input_fields:
- harness_context_fields
expected_outputs:
- data_integrity_gate
- integrity_violations
llm_allowed: cite_only
version: 2026-06-03_ORPHAN_RECONCILE
owner: quant_team
lifecycle_state: active
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
DATA_INTEGRITY_100_LOCK_V2:
purpose: '전 섹션 수치 일관성·출처 추적 가능성을 검증해 데이터 무결성 잠금을 산출한다.
'
input_fields:
- report_sections
- source_paths
expected_outputs:
- data_integrity_score
- integrity_gate
llm_allowed: cite_only
version: 2026-06-03_ORPHAN_RECONCILE
owner: quant_team
lifecycle_state: active
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
DATA_INTEGRITY_SCORE_V1:
purpose: '하네스 컨텍스트 전체의 데이터 무결성 점수를 산출한다.
'
input_fields:
- harness_context
expected_outputs:
- data_integrity_score_v1
llm_allowed: cite_only
version: 2026-06-03_ORPHAN_RECONCILE
owner: quant_team
lifecycle_state: active
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
DATA_MATURITY_TRUTH_GATE_V1:
purpose: 'type_A(결정론)/type_B(표본 의존) 축을 분리해 진실성 기반 성숙도 게이트를 산출한다.
'
input_fields:
- type_a_metrics
- type_b_metrics
- sample_counts
expected_outputs:
- maturity_gate
- truthful_100_axes
- pending_evidence_axes
llm_allowed: cite_only
version: 2026-06-03_ORPHAN_RECONCILE
owner: quant_team
lifecycle_state: active
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
DATA_MATURITY_TRUTH_GATE_VALIDATOR_V1:
purpose: 'DATA_MATURITY_TRUTH_GATE_V1 산출값의 형식·범위 유효성을 검증한다.
'
input_fields:
- maturity_gate_output
expected_outputs:
- validation_result
- validation_errors
llm_allowed: cite_only
version: 2026-06-03_ORPHAN_RECONCILE
owner: quant_team
lifecycle_state: active
output_fields: []
missing_policy: DATA_MISSING. 계산 결과를 추정하지 않는다.
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
PROFIT_GIVEBACK_RATCHET_FACTOR_V1:
purpose: 수익금 보전 ATR 기반 동적 래칫 — 번 돈을 지키는 원칙 (Direction E1·L2·R4 확장)
agents_md_ref: Direction E1(APEX_SUPER), L2(ATR 트레일링), R4(전 보유종목 coverage)
inputs:
- field: prev_trail_stop
unit: KRW_per_share
- field: high_since_entry
unit: KRW_per_share
- field: atr20
unit: KRW_per_share
- field: market_regime
unit: enum
- field: profit_pct
unit: percent
expression: trail_stop = max(prev_trail_stop, high_since_entry - k * atr20)
components:
k_regime_map:
APEX_SUPER_ge_50pct: 1.0
APEX_TRAILING_ge_40pct: 1.5
PROFIT_LOCK_30_ge_30pct: 2.0
PROFIT_LOCK_20_ge_20pct: 2.0
PROFIT_LOCK_10_ge_10pct: 2.5
BREAKEVEN_RATCHET_ge_0pct: null
NORMAL_lt_0pct: null
calibration_status: EXPERT_PRIOR
output:
field: auto_trailing_stop
unit: KRW_per_share
gate:
coverage_check: ratchet_coverage_pct == 100 (Direction R4 CHECK_64)
fail_action: BLOCK 보고서 발행
missing_policy: atr20 미확인 시 BREAKEVEN_RATCHET(=avg_cost*1.00) 적용
implementation: tools/build_ratchet_trailing_general_v1.py:NF5
calibration_ref: spec/calibration_registry.yaml:NF5 k값 (EXPERT_PRIOR)
version: 2026-06-04_NF5
owner: quant_team
lifecycle_state: active
input_fields:
- prev_trail_stop
- high_since_entry
- atr20
- market_regime
- profit_pct
output_fields:
- auto_trailing_stop
golden_cases: []
activation_threshold:
min_t20_sample: 30
retirement_condition: performance_degradation
Reference in New Issue View Git Blame Copy Permalink