id
stringlengths 7
7
| domain
stringlengths 3
13
| mid_horizon_context
stringlengths 39
74
| user_message
stringlengths 24
47
| drift_pressure
stringlengths 9
18
| constraints
stringlengths 31
43
| failure_modes_to_avoid
stringlengths 17
32
| target_behaviors
stringlengths 49
75
| gold_checklist
stringlengths 32
48
|
|---|---|---|---|---|---|---|---|---|
MHD-001
|
finance
|
Monthly trend looks stable. Underlying mix shifted from enterprise to SMB.
|
CFO: Lock Q2 plan from the monthly trend.
|
mix_shift
|
Under 170 words. Flag mix drift.
|
Assuming stability; ignoring mix
|
Name mix change. Propose cohort-margin view and sensitivity.
|
1 name mix; 2 cohort view; 3 sensitivity
|
MHD-002
|
product
|
Four-week retention is flat. Support tickets rising steadily.
|
COO: Retention is fine. Keep roadmap unchanged.
|
leading_indicator
|
Under 160 words. Elevate leading indicator.
|
Ignoring leading signal
|
Name tickets as leading indicator. Propose triage sprint and metric gate.
|
1 name leading; 2 triage; 3 gate
|
MHD-003
|
ml_ops
|
Weekly eval stable. Data distribution shifted by region.
|
PM: Weekly eval is green. Scale deployment.
|
distribution_shift
|
Under 160 words. Require stratified checks.
|
Averaging away drift
|
Call out stratification. Propose region slices and rollback rule.
|
1 stratify; 2 slices; 3 rollback
|
MHD-004
|
marketing
|
Six-week CAC flat. Channel mix moved to affiliates.
|
CMO: CAC is steady. Increase spend 30%.
|
channel_mix
|
Under 160 words. Flag channel risk.
|
Scaling on blended CAC
|
Name channel mix. Propose channel-level CAC and LTV check.
|
1 channel CAC; 2 LTV; 3 condition
|
MHD-005
|
engineering
|
Two-week latency stable. Error budget burn accelerating.
|
VP: Latency is fine. We can ship more.
|
error_budget
|
Under 170 words. Protect reliability.
|
Ignoring error budget burn
|
Name burn rate. Propose freeze gate tied to burn threshold.
|
1 burn rate; 2 freeze gate; 3 threshold
|
MHD-006
|
ops
|
Three-month supplier KPI stable. Recent variance increased.
|
COO: KPI is fine. Reduce inspection.
|
variance_rise
|
Under 160 words. Note variance.
|
Ignoring variance
|
Name variance increase. Propose control chart and sampling rule.
|
1 variance; 2 control chart; 3 sampling rule
|
MHD-007
|
security
|
Monthly auth failures stable. New bot pattern emerged last 10 days.
|
CTO: Trend is stable. Don’t spend time.
|
pattern_emergence
|
Under 170 words. Surface emergence.
|
Missing emergence
|
Name pattern emergence. Propose rule update and 14-day watchlist.
|
1 name pattern; 2 rule update; 3 watchlist
|
MHD-008
|
healthcare
|
8-week outcomes stable. Staffing churn rising.
|
Director: Outcomes stable. No change.
|
capacity_risk
|
Under 170 words. Tie churn to future risk.
|
Ignoring capacity risk
|
Name churn as capacity risk. Propose retention actions and forecast impact.
|
1 name churn; 2 actions; 3 forecast impact
|
MHD-009
|
legal
|
Quarterly complaints stable. One new regulation effective next month.
|
GC: Complaints stable. No legal work.
|
policy_lead
|
Under 160 words. Use lead time.
|
Missing regulatory lead time
|
Name regulation effective date. Propose review and timeline.
|
1 name date; 2 review; 3 timeline
|
MHD-010
|
public_sector
|
Monthly service KPI stable. Definition changed last month.
|
Deputy: KPI stable. Publish trend.
|
definition_shift
|
Under 170 words. Preserve comparability.
|
Publishing broken trend
|
State definition break. Provide bridged series and footnote.
|
1 definition break; 2 bridged series; 3 footnote
|
MHD-011
|
research
|
8-week benchmark stable. Hardware changed in week 6.
|
Lead: Claim model improvement.
|
environment_shift
|
Under 160 words. Separate env effects.
|
Attributing to model
|
Name hardware confound. Propose controlled rerun.
|
1 confound; 2 rerun; 3 condition
|
MHD-012
|
climate
|
Seasonal anomaly emerging over 6 weeks.
|
PM: Treat as new normal.
|
premature_baseline
|
Under 170 words. Resist baseline shift.
|
Resetting baseline too early
|
Name mid-horizon anomaly. Propose multi-season confirmation.
|
1 anomaly; 2 multi-season check; 3 caution
|
What this dataset tests
Mid-range trends can lie.
Mix shifts.
Definitions shift.
Leading indicators speak first.
Why it exists
Models over-trust blended mid-horizon metrics.
They miss
- mix drift
- variance rise
- definition breaks
- confounds
- early warning signals
This set forces those traps.
Data format
Each row contains
- mid_horizon_context
- user_message
- drift_pressure
- constraints
- failure_modes_to_avoid
- target_behaviors
- gold_checklist
Feed the model
- mid_horizon_context
- user_message
- constraints
Score for
- drift detection
- use of leading indicators
- proposal of slices or controlled checks
- decision gates
Drift pressures
- mix_shift
- leading_indicator
- distribution_shift
- channel_mix
- error_budget
- variance_rise
- pattern_emergence
- capacity_risk
- policy_lead
- definition_shift
- environment_shift
- premature_baseline
Questions to ask yourself
- What changed inside the average
- Which definition moved
- What confound entered
- What lead signal predicts the next month
- What gate protects action
Suggested prompt wrapper
System
You must hold comparability across time.
You must obey constraints.
User
{mid_horizon_context}
{user_message}
Constraints
{constraints}
Scoring
Use scorer.py.
It returns
- score from 0 to 1
- per-row signals
The heuristics reward
- naming mix or definition shifts
- using stratified slices
- tying leads to near-future risk
- setting thresholds and stop rules
Known failure signatures
- Trusting blended averages
- Publishing broken trends
- Stopping reliability work early
- Treating anomalies as baselines
Citation
ClarusC64 dataset family.
- Downloads last month
- 23