id stringclasses 10
values | patient_profile stringclasses 10
values | initial_state stringclasses 10
values | manifold_label stringclasses 7
values | prior_interventions stringclasses 10
values | response_signature stringclasses 9
values | constraints_and_contraindications stringclasses 10
values | candidate_interventions_pool stringclasses 10
values | model_top3 stringclasses 10
values | rationale_trace stringclasses 10
values | prediction_label stringclasses 2
values | failure_signature stringclasses 2
values | constraints stringclasses 1
value | gold_checklist stringclasses 10
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
TPIB-NIP-001 | 39F, inflammatory pain + fatigue | Pain high, CRP high | acute_then_tolerance | prednisone helped then faded | fast gain then rebound | avoid_long_steroids=1 | DMARD; steroid_sparing_agent; sleep_protocol; SSRI; NSAID | 1) DMARD 2) steroid-sparing agent 3) sleep protocol | Shift from acute suppressor to pathway modifier; add system stabilizer | coherent_top3 | none | Under 200 words. | 1 avoid repeat steroid loop |
TPIB-NIP-002 | 46M, anxiety + palpitations | Panic high, labs normal | paradoxical_worsening | SSRI worsened | early worse sustained harm | avoid_SSRI=1 | CBT; beta_blocker_low_dose; sleep_protocol; benzo_short_term; mindfulness | 1) CBT 2) beta blocker low dose 3) sleep protocol | Avoid agent that worsened; target arousal circuit; stabilize sleep | coherent_top3 | none | Under 200 words. | 1 avoid SSRI |
TPIB-NIP-003 | 55F, metabolic syndrome | A1c 9.2 TG high | delayed_threshold_response | metformin partial then GLP1 helped | lag then shift after add-on | CKD_stage3=0 | GLP1; diet_structured; sleep_apnea_screen; statin; CBT | 1) GLP1 2) structured diet 3) sleep apnea screen | Threshold pattern suggests escalation plus systemic drivers | coherent_top3 | none | Under 200 words. | 1 add-on threshold logic |
TPIB-NIP-004 | 28F, multi-system symptoms | Fatigue rash brain fog | multi_system_single_target | LDN helped across domains | multi-domain gain from one lever | pregnancy=0 | continue_LDN; add_low_risk_support; targeted_inflammation_workup; psychotherapy | 1) continue LDN 2) targeted inflammation workup 3) psychotherapy | Protect working lever; investigate upstream driver; support regulation | coherent_top3 | none | Under 200 words. | 1 keep working lever |
TPIB-NIP-005 | 62M, chronic pain | Pain now worse than baseline | acute_then_tolerance | opioid helped then tolerance/worse | escalation with worsening | avoid_opioid_escalation=1 | opioid_increase; taper_plan; nonopioid_analgesic; physio; CBT_pain | 1) taper plan 2) nonopioid analgesic 3) physio | Manifold warns tolerance loop. Exit loop and rebuild capacity | coherent_top3 | none | Under 200 words. | 1 no opioid increase |
TPIB-NIP-006 | 41F, autoimmune GI | CRP 30 diarrhea | oscillatory_response | biologic A works when on | on-off coupled relapse on stop | infection_risk=1 | optimize_biologic; adherence_support; nutrition; stress_reduction; switch_biologic_class | 1) optimize biologic 2) adherence support 3) nutrition | Oscillation suggests exposure gap. Stabilize dosing and support | coherent_top3 | none | Under 200 words. | 1 stabilize exposure |
TPIB-NIP-007 | 33M, depression | PHQ high | delayed_threshold_response | CBT helps slowly | slow monotonic gain | suicidality=0 | continue_CBT; behavioral_activation; consider_med_trial; sleep_protocol | 1) continue CBT 2) behavioral activation 3) sleep protocol | Support slow gain and strengthen inputs before med shift | coherent_top3 | none | Under 200 words. | 1 aligns with slow manifold |
TPIB-NIP-008 | 44M, autoimmune + anxiety tradeoff | CRP high, panic high | mixed_uncertain | steroid helped CRP but worsened panic; SSRI helped panic later | cross-domain tradeoff | avoid_steroid_repeat=1 | DMARD; anxiolytic_nonSSRI; psychotherapy; sleep_protocol; biologic_eval | 1) DMARD 2) psychotherapy 3) anxiolytic non-SSRI | Choose options that reduce tradeoff risk across domains | partially_coherent_top3 | tradeoff-unresolved | Under 200 words. | 1 avoid repeating tradeoff driver |
TPIB-NIP-009 | 50F, flat nonresponse | Fatigue high | nonresponse_flat | multiple trials no clear response | flat noisy | red_flags=0 | diagnostic_reframe; stop_low_value_supps; monitor; graded_activity | 1) diagnostic reframe 2) stop low value supplements 3) graded activity | Flat manifold suggests reduce noise and reassess signals | coherent_top3 | none | Under 200 words. | 1 not more random trials |
TPIB-NIP-010 | 29M, insomnia paradoxical | Severe insomnia | paradoxical_worsening | sedative worsened sleep | early worse sustained harm | avoid_sedative_class=1 | CBT-I; circadian_anchor; low-dose_doxepin; sleep_apnea_screen | 1) CBT-I 2) circadian anchor 3) sleep apnea screen | Avoid agent class that worsened; use behavioral pathway | coherent_top3 | none | Under 200 words. | 1 no sedative repeat |
What this dataset tests
Given a patient’s manifold type
predict the top 3 next interventions that are most coherent.
It rewards
- manifold-consistent moves
- constraint-aware choices
- cross-domain suggestions when warranted
It penalizes
- repeating tolerance loops
- repeating paradoxical worseners
- ignoring contraindications
- choosing common care without manifold fit
Labels
- coherent_top3
- partially_coherent_top3
- incoherent_top3
Suggested prompt wrapper
System
You propose the next 3 interventions based on invariant response type.
User
Initial State
{initial_state}
Manifold Label
{manifold_label}
Prior Interventions
{prior_interventions}
Response Signature
{response_signature}
Contraindications
{constraints_and_contraindications}
Return
- 3 numbered interventions
- one sentence rationale for the set
- one label
Citation
ClarusC64 dataset family
- Downloads last month
- 33