Data Freshness
✓Metrics data7.19.26 (0d old)
Source: dashboards/dashboard_data.js · Mtime: 0d ago
✓Recommendation events0d old
Source: data/processed/recommendation_events.csv
Top Buys (0)
Top BF Combo
FCF Efficency + EPS-ROA PE Score + SM Watson 3
Open BF Combo Dashboard →★Win rate100%
Lift: +78.6pp vs 21.4% baseline · n=17
·Combo metricsFCF Efficency + EPS-ROA PE Score + SM Watson 3
Each metric screens independently against its threshold (AND-gate). Tickers must pass ALL rules simultaneously.
Active Gate (5 rules)
Live from mde_rules_config.RULES_6
Open Gate Alternatives →✓?dist=? · ?
sm_col: 65 · scripts_dist: ? · direction: ?
✓?dist=? · ?
sm_col: 31 · scripts_dist: ? · direction: ?
✓?dist=? · ?
sm_col: 59 · scripts_dist: ? · direction: ?
✓?dist=? · ?
sm_col: 52 · scripts_dist: ? · direction: ?
✓?dist=? · ?
sm_col: 57 · scripts_dist: ? · direction: ?
Health Check
⛔Errors2
Errors must be 0 before any pipeline run. Open metric_discovery_engine/data/processed/health_check_report.html for details.
⚠Warnings66
Review warnings before next pipeline run. They do not block execution but indicate drift, staleness, or governance gaps.
Gate Alternatives (14 alts)
Above current: 0/14 · current composite: 54.3
Open Gate Alternatives Dashboard →·Current Gate (Production Baseline)composite=54.3 · v12 — User Curated Lean Gate
gate_id: 949 · sel=95 · ctrl=4.2 · mps=66.6 · dir=100 · pass_all=6 · n_rules=5
★SIL-Driven Fixcomposite=54.3 · Δ +0.0 · 📌 pinned
gate_id: 949 · group: quick_swap · n_rules: 5 · pass_all: 6 · No swap-outs recommended — gate is healthy at this severity tier · top3: PLTR, ALAB, VICR
✓Top-4-SIL + FCF Qualitycomposite=53.6 · Δ -0.7
gate_id: 1687 · group: deeper_overhaul · n_rules: 5 · pass_all: 15 · Auto: top 4 non-gate L2 metrics by sil_score + FCF Quality retained from current gate · top3: PLTR, CRDO, TER
✓Top-4-SIL + RSI-Net EBITDAcomposite=53.1 · Δ -1.2
gate_id: 1009 · group: deeper_overhaul · n_rules: 5 · pass_all: 2 · Auto: top 4 non-gate L2 metrics by sil_score + RSI-Net EBITDA retained from current gate · top3: PLTR, CRDO, TGTX
✓Top-4-SIL + ROIC-EBITDA QCcomposite=52.0 · Δ -2.3
gate_id: 3039 · group: deeper_overhaul · n_rules: 5 · pass_all: 7 · Auto: top 4 non-gate L2 metrics by sil_score + ROIC-EBITDA QC retained from current gate · top3: PLTR, CRDO, TER
✓Top-4-SIL + FCF Efficiencycomposite=51.8 · Δ -2.5
gate_id: 4876 · group: deeper_overhaul · n_rules: 5 · pass_all: 13 · Auto: top 4 non-gate L2 metrics by sil_score + FCF Efficiency retained from current gate · top3: PLTR, CRDO, TER
✓Raw Foundationscomposite=49.9 · Δ -4.4
gate_id: 744 · group: experimental · n_rules: 5 · pass_all: 1 · Pure L1 gate — raw ratios only, control-lab directions, universe-percentile thresholds · top3: DJCO, ELSE, ETON
✓Mixed Tiercomposite=48.7 · Δ -5.6
gate_id: 7301 · group: experimental · n_rules: 4 · pass_all: 14 · Best L1 signal anchored by top L2 quality checks — tests if raw signals add lift · top3: NSP, ELSE, LSPD
✓Top-4-SIL + Avg EPS Growthcomposite=48.3 · Δ -6.0
gate_id: 4452 · group: deeper_overhaul · n_rules: 5 · pass_all: 13 · Auto: top 4 non-gate L2 metrics by sil_score + Avg EPS Growth retained from current gate · top3: PLTR, CRDO, TER
✓Independent Spreadcomposite=42.4 · Δ -11.9
gate_id: 4218 · group: deeper_overhaul · n_rules: 4 · pass_all: 127 · Best balance of MPS grade, signal independence, coverage, and direction confidence · top3: ALAB, VICR, CPSH
✓Top Grade Fivecomposite=39.6 · Δ -14.7
gate_id: 7487 · group: deeper_overhaul · n_rules: 4 · pass_all: 387 · Maximize average MPS score — all 5 rules A-grade, zero B/F rules · top3: OSS, CPSH, UROY
Data Pulls
See references/METRIC_UPDATE_RUNBOOK.md
✓finviz_pull2026-07-18 (2d ago)
Source: data/processed/finviz_pull_*.csv · Cadence: weekly (warn ≥7d)
✓ratio_pull2026-07-19 (1d ago)
Source: data/processed/ratio_pull_*.csv · Cadence: weekly (warn ≥7d)
✓newresearch2026-07-20 (0d ago)
Source: data/processed/newresearch_*.csv · Cadence: weekly (warn ≥7d)
Control Lab
UNDERPERFORMING — treatment lags control by -13.2pp (p=0.034, 43 recs priced)
Open Control Group Lab →⚠Treatment N43/150 (29%)
Sample size of FORMAL recs with priced returns. Target N≥150 for statistical significance per SYSTEM_STANDARDS.md §3.
·Control N priced494/500
Active control group: 494/500. Stale prices skipped (>0d).
⚠Treatment excess-8.30pp
Mean treatment return minus market baseline.
·Control excess (all-universe)+4.95pp
Mean control group return minus market baseline — all 500 random tickers (coin-flip baseline).
★Gate-Filtered Random (PRIMARY)+29.96pp (N=5)
Mean excess of control tickers passing 4+ rules — authoritative gate-effectiveness measurement, isolated from rec-process noise (added 2026-05-21).
⚠Rec-process gap-38.3pp
FORMAL recs minus Gate-Filtered Random — measures rec-curation value-add (>0) or destruction (<0) on top of the gate.
·SPY since seed+17.22%
Market baseline return since control seed date.
★p-value0.034
Welch's two-tailed t-test, treatment vs control. <0.05 = statistically significant edge.
·VerdictUNDERPERFORMING — treatment lags control by -13.2pp (p=0.034, 43 recs priced)
UNDERPERFORMING — treatment lags control by -13.2pp (p=0.034, 43 recs priced)
⚠Direction disagreements38
Metrics where control direction disagrees with cohort direction. Investigate before acting.
★Buy candidates2
Control tickers scoring 4+/6 + outperforming. See Control_Group_Lab.html.
★New signal candidates5
New metric signals identified by the lab — review in dashboard.
Signal Intelligence (top 5 of 71)
Run date: 2026-07-19
★EPS-ROA PE Score★★★★★ · higher wins · in_gate=no
5-star | higher wins | 49 sessions held | not in gate | control agrees · EPS-ROA PE Score has 5-star confidence and is not in your screening gate. It has held its direction (higher scores win) for 49 consecutive sessions. Every run you screen without it, you leave a validated signal unused. This is your clearest ADD candidate right now. || EPS-ROA PE Score confidence ros
★FCF Efficency★★★★★ · lower wins · in_gate=yes
5-star | lower wins | 174 sessions held | IN GATE | control agrees · FCF Efficency confidence rose this run — momentum is building (1 more sessions to next star tier). If it continues, this metric will cross the formal ADD threshold. Watch closely.
★Quality Value Score★★★★★ · lower wins · in_gate=no
5-star | lower wins | 167 sessions held | not in gate | control agrees · Quality Value Score has 5-star confidence and is not in your screening gate. It has held its direction (lower scores win) for 167 consecutive sessions. Every run you screen without it, you leave a validated signal unused. This is your clearest ADD candidate right now. || Quality Value Score confiden
✓RSI-Net EBITDA Signal★★★★☆ · higher wins · in_gate=yes
4-star | higher wins | 52 sessions held | IN GATE | control disagrees · RSI-Net EBITDA Signal has been in your gate for 52 sessions but the 500-stock random control group consistently disagrees with its direction. Historical winners passed this rule, but random stocks that also passed it are keeping pace or outperforming. The rule may be less predictive than it looks. W
✓ROIC/Ebitda Spread★★★★☆ · higher wins · in_gate=no
4-star | higher wins | 171 sessions held | not in gate | control agrees · ROIC/Ebitda Spread has 4-star confidence and is not in your screening gate. It has held its direction (higher scores win) for 171 consecutive sessions. Every run you screen without it, you leave a validated signal unused. This is your clearest ADD candidate right now. || ROIC/Ebitda Spread confidenc
SIL — Top Combos
✓FCF Efficency + Avg. EPS Growth + Quality Value Score + PFCF vs PE Spreadwin ?% · ?
n: ? · ctrl_n: ? · score: ?
✓FCF Efficency + Avg. EPS Growth + PFCF vs PE Spreadwin ?% · ?
n: ? · ctrl_n: ? · score: ?
✓FCF Efficency + Avg. EPS Growth + CARC + PFCF vs PE Spreadwin ?% · ?
n: ? · ctrl_n: ? · score: ?
✓FCF Efficency + RSI x Avg Eps Growth + Quality Value Score + PFCF vs PE Spreadwin ?% · ?
n: ? · ctrl_n: ? · score: ?
✓FCF Efficency + Avg. EPS Growth + Quality Value Score + Watson-ROIC Confidence Indexwin ?% · ?
n: ? · ctrl_n: ? · score: ?
Gate Alerts (7)
★EPS-ROA PE ScoreADD · in_gate=no · BF#1
Score=90/100 (MPS:30 D:20 Ctrl:20 BF:20). MPS: B (60.4). Direction: accurate. Control: agrees (higher, sep=13.55pp · strong). BF rank #1.
★Quality Value ScoreADD · in_gate=no · BF#5
Score=78/100 (MPS:18 D:20 Ctrl:20 BF:20). MPS: C (57.52). Direction: accurate. Control: agrees (lower, sep=23.56pp · strong). BF rank #5.
★SM2: Growth-Momentum PremiumADD · in_gate=no · BF#10
Score=72/100 (MPS:18 D:20 Ctrl:20 BF:14). MPS: C (50.5). Direction: accurate. Control: agrees (lower, sep=8.26pp · strong). BF rank #10.
★CARCADD · in_gate=no · BF#11
Score=64/100 (MPS:30 D:0 Ctrl:20 BF:14). MPS: B (60.89). Direction: accurate. Control: agrees (lower, sep=5.52pp · strong). BF rank #11.
★P/BADD · in_gate=no · BF#43
Score=50/100 (MPS:30 D:20 Ctrl:0 BF:0). MPS: B (74.03). Direction: accurate. Control: disagrees (lower, sep=12.69pp · strong). BF rank #43.
★ROICADD · in_gate=no · BF#55
Score=80/100 (MPS:40 D:20 Ctrl:20 BF:0). MPS: A (98.33). Direction: accurate. Control: agrees (lower, sep=24.45pp · strong). BF rank #55.
★ROICADD · in_gate=no · BF#72
Score=0/100 (MPS:0 D:0 Ctrl:0 BF:0). MPS: n/a (n/a). Direction: not_enough_data. Control: disagrees (lower, sep=24.45pp · strong). BF rank #72.
Metric Build Recs
Top SUPER METRIC PROMPT (ctrl_n ≥ 10)
✓FCF Efficency + Avg. EPS Growth + Quality Value Score + PFCF vs PE Spreadscore=81.87 · holdings=92.31% (n=13) · ctrl=85.71% (n=14)
k=4 · direction: INVESTIGATE
Direction Alerts (12)
⚠SM Watson 5INVESTIGATE · conf=n/a
Signals split — Control says lower (d=0.0942, n=68out/114under, sep=2.50pp · moderate), Cohen's says higher (d=0.5222, n=34w/302l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⚠ROIC-FCF Quality GateINVESTIGATE · conf=n/a
Signals split — Control says lower (d=-0.1488, n=108out/188under, sep=7.23pp · strong), Cohen's says higher (d=0.9167, n=31w/321l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⚠Watson-ROIC Confidence IndexINVESTIGATE · conf=n/a
Signals split — Control says lower (d=0.1645, n=93out/124under, sep=1.04pp · weak), Cohen's says higher (d=1.0102, n=34w/309l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⚠Pcm v roaINVESTIGATE · conf=n/a
Signals split — Control says higher (d=-0.0801, n=82out/145under, sep=3.37pp · strong), Cohen's says lower (d=-0.1204, n=34w/330l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⚠SM3: Quality-Overvaluation WildcardINVESTIGATE · conf=n/a
Signals split — Control says higher (d=-0.0537, n=64out/119under, sep=10.08pp · strong), Cohen's says lower (d=-0.5349, n=34w/264l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⚠Rev Growth EfficiencyINVESTIGATE · conf=n/a
Signals split — Control says lower (d=-0.1948, n=76out/134under, sep=12.91pp · strong), Cohen's says higher (d=0.0159, n=31w/288l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⚠Avg. EPS GrowthINVESTIGATE · conf=n/a
Signals split — Control says higher (d=0.1856, n=132out/188under, sep=5.38pp · strong), Cohen's says lower (d=-0.0284, n=34w/347l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⚠Capital Return SpreadINVESTIGATE · conf=n/a
Signals split — Control says lower (d=0.1011, n=165out/329under, sep=5.03pp · strong), Cohen's says higher (d=0.1719, n=34w/367l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⛔ROI Liquidity / FCFFLIP · conf=HIGH · ⚠ AT RISK
Control: higher (n=45out/85under, d=-0.2147, sep=5.82pp · strong) + Cohen's d=0.4871 (n=31w/266l) — both disagree with lock=lower · Resolution: Evidence exceeds thresholds — present flip to user for approval.
⚠RSI x Avg Eps GrowthINVESTIGATE · conf=n/a
Signals split — Control says higher (d=0.1951, n=132out/188under, sep=9.94pp · strong), Cohen's says lower (d=-0.0272, n=34w/347l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⚠Risk to Value GrowthINVESTIGATE · conf=n/a
Signals split — Control says lower (d=0.1945, n=165out/329under, sep=5.10pp · strong), Cohen's says higher (d=0.2841, n=34w/360l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
⚠RSI-Net EBITDA SignalINVESTIGATE · conf=n/a
Signals split — Control says lower (d=-0.0593, n=165out/329under, sep=13.20pp · strong), Cohen's says higher (d=0.118, n=34w/367l). Manual review required before any flip. · Resolution: Control group is primary signal (N≥50, unbiased). Cohort (secondary metrics) disagrees. Control direction holds by default. Watch: if cohort reaches 10+ consecutive opposing sessions + |d| >= 0.15 -> bring to user. User decides whether to align with cohort or trust control.
Open Tasks (1)
CRITICAL + VERY HIGH + HIGH from TASKS.md
Open TASKS.md →·[INVESTIGATE] BF Type 1 combo scores measure held/sold binary not actual returns…
[INVESTIGATE] BF Type 1 combo scores measure held/sold binary not actual returns — all combo rankings are potentially corrupted — 2026-06-22. Why CRITICAL (data-integrity): mde_04_brute_force.py lines 271/362–371/516–518 compute combo win rate as held=1 / sold=0 binary count, NOT avg return of passers vs failers. A held ticker that lost money counts as a "win." This means every BF combo recommendation used for gate decisions is scoring "did we hold it" not "did it make money" — the two can diverge significantly. separation_calc.py (Type 2) does use real return-based separation for combos but it's unclear which downstream consumers (combo dashboard, gate advisor, MPS) read Type 1 vs Type 2. Audit scope: (1) confirm all Type 1 score consumers in mde_04, mde_combination_results.csv, and gate advisor; (2) compare Type 1 vs Type 2 rankings for top combos — how different are they?; (3) assess fix path: replace Type 1 lift_pct with return-based separation where returns exist. Also revisit references/TRACKING_SYSTEMS.md, references/CONTROL_LAB_METHODOLOGY.md, references/BUILD_RECOMMENDATION_METHODOLOGY.md — these docs define winner/return tracking methodology and must be checked for consistency. *Filed via /capture-issue 2026-06-22.*
Governance Reminders
·Pre-BuildQ0–Q9 in investments/CLAUDE.md
Run before any new build/script/dashboard. Pre-Build Assessment fires Q0 (LESSONS_LEARNED) → Q9 (Constitution).
·Pre-FixSteps 1–6 in investments/CLAUDE.md
Run before modifying any existing file. Visible checklist required, not prose.
⚠run_pipeline.pyExplicit "yes go ahead" required
Master orchestrator — Danger Zone. Present full plan, wait for unambiguous approval, do not infer from context.
·/risk-check skillFires full Pre-Build + Pre-Fix in one pass
Run before any build or fix. Skill: ~/.claude/skills/risk-check/SKILL.md