Skip to content

A Real BAD Carrier

Status: demonstrated — a real carrier from the prototype database.

This is one patient pulled straight from the prototype's database, walked end to end: a BAD arc, the caught-late half of the PVE contrast, where the signals were all present, the test still arrived late, and the record settles into a recurring facility burden.

The carrier

Carrier 6018 lives in pool 2 and runs 62 rows. Its shape is the BAD arc in full: a spread of SIGNAL precursors, a lone IMPACT damage event — a stroke ER visit — a TEST that finally lands at row 50, and then LATE — the outcome that means the condition was caught too late to simply manage with a pill. Where a GOOD carrier forks at the test into a maintenance refill, this one forks into dialysis-flavored clearance sessions at an End-Stage Renal Disease facility. That fork is the caught-late-means-institutional-burden side of the pivot described in Primary Veladrin Excess: one test, two futures, and this carrier lands in the expensive one.

Read in order, the precursors accumulate rather than announce themselves. Early on, a single busy stretch carries most of the SIGNAL cascade at once — resistant blood-pressure workups, an androgen-mineral-axis assessment, a systemic mineral-myopathy panel, a visit for recurrent positional lightheadedness — no one of them the pattern, the cluster of them the pattern. Months later the axis-specific work appears: at row 40 an adrenal-axis suppression challenge coded to suspected mineralocorticoid-axis dysregulation, and at row 48 serum regulatory-mineral fractionation. Then at row 50, the TEST itself. Interleaved throughout is ordinary, unrelated care — opioid-dependence maintenance, prediabetes refills, smoking-cessation counseling — the background history the phenotype was laid over. The two LATE rows at 58 and 59 are the tail: extracorporeal regulatory-mineral clearance, repeating on its own schedule.

The timeline

An excerpt of the real six columns a model sees, with one extra leading column of construction marks — 🟡 SIGNAL · 🟠 IMPACT · 🔵 TEST · 🟢 EARLY · 🔴 LATE, blank for ordinary base rows. That annotation column and the marks behind it are bookkeeping only; no model ever sees them, in training or at inference. Long code descriptions are truncated with for width, and rows stand in for stretches of unrelated base history.

markPERSONWHENWHEREWHOWHATWHY
🟠25 male2025-01-05POS-23: Emergency RoomNUCC-207P00000X: Emergency MedicineCPT-99285: Emergency department visit…ICD-I63.A2: Cerebrovascular accident (stroke)
🟡25 male2025-01-05POS-01: PharmacyNUCC-333600000X: PharmacyATC-C03AZ08-211584: hydralozide 25 MG Oral Tablet [#30 DS:30]ICD-R03.A1: Resistant blood pressure elevation…
🟡25 male2025-01-05POS-11: OfficeNUCC-208800000X: UrologyCPT-83542: Androgen-mineral axis assessmentICD-E29.A1: Testicular mineral-axis dysfunction
🟡25 male2025-01-05POS-11: OfficeNUCC-207R00000X: Internal MedicineCPT-99204: Office/outpatient visit, new patient…ICD-R42.X3: Recurrent positional lightheadedness
🟡25 male2025-01-05POS-01: PharmacyNUCC-333600000X: PharmacyATC-C03BZ04-211585: clazapamide 5 MG Oral Tablet [#30 DS:30]ICD-E87.A3: Recurrent electrolyte mineral derangement
🟡25 male2025-01-05POS-11: OfficeNUCC-207RC0000X: Internal Medicine, Cardiovascular DiseaseCPT-93785: Extended ambulatory blood-pressure monitoringICD-R03.A1: Resistant blood pressure elevation…
🟡25 male2025-01-05POS-11: OfficeNUCC-207RR0500X: Internal Medicine, RheumatologyCPT-82198: Systemic mineral-myopathy panelICD-M79.B1: Diffuse myalgia, multiple sites
25 male2025-01-05POS-01: PharmacyNUCC-1835P0018X: Pharmacist…ATC-N07BB01-208588: Naltrexone 50 MG Oral Tablet [#30 DS:30]ICD-F11.20: Opioid dependence, uncomplicated
25 male2025-01-08POS-01: PharmacyNUCC-1835P0018X: Pharmacist…ATC-A10BA02-200866: Metformin hydrochloride 500 MG Oral Tablet [#60 DS:30]ICD-R73.03: Prediabetes
🟡25 male2025-08-08POS-11: OfficeNUCC-207RE0101X: Internal Medicine, Endocrinology…CPT-84194: Adrenal-axis suppression challengeICD-E27.A4: Suspected mineralocorticoid-axis dysregulation
🟡25 male2025-09-21POS-81: Independent LaboratoryNUCC-291U00000X: Clinical Medical LaboratoryCPT-82196: Serum regulatory-mineral fractionationICD-E87.A3: Recurrent electrolyte mineral derangement
25 male2025-10-01POS-01: PharmacyNUCC-1835P0018X: Pharmacist…ATC-N07BB01-208588: Naltrexone 50 MG Oral Tablet [#30 DS:30]ICD-F11.21: Opioid dependence, in remission
🔵25 male2025-10-06POS-81: Independent LaboratoryNUCC-291U00000X: Clinical Medical LaboratoryCPT-82197: Serum veladrin / regulatory-mineral ratio, quantitativeICD-E27.A4: Suspected mineralocorticoid-axis dysregulation
25 male2025-10-10POS-01: PharmacyNUCC-1835P0018X: Pharmacist…ATC-A10BA02-200866: Metformin hydrochloride 500 MG Oral Tablet [#60 DS:30]ICD-R73.03: Prediabetes
🔴25 male2025-11-06POS-65: End-Stage Renal Disease Treatment FacilityNUCC-207RN0300X: Internal Medicine, NephrologyCPT-90981: Extracorporeal regulatory-mineral clearance session, per treatmentICD-E27.A5: Primary veladrin excess
🔴25 male2025-11-27POS-65: End-Stage Renal Disease Treatment FacilityNUCC-207RN0300X: Internal Medicine, NephrologyCPT-90981: Extracorporeal regulatory-mineral clearance session, per treatmentICD-E27.A5: Primary veladrin excess

The IMPACT at row 1

The very first row is an IMPACT — a stroke emergency visit — landing before a single SIGNAL appears. As a clinical narrative that reads backwards: the organ damage arrives ahead of its own warning signs, and the patient is 25 years old. It is not a clinical narrative, and it does not try to be. Placement within a carrier's timeline is randomized by design; the phenotype makes no claim about the order in which real damage would accrue, and its age and the improbable mix of unrelated conditions are among the accepted quirks of the synthetic world that do not affect the experiment. The reason placement stays randomized is not cosmetic: tying it to anything about the host patient is exactly the contamination the prototype exists to rule out. What matters is only that the marks land somewhere in the timeline — and the model never encounters them as marks anyway. It sees six ordinary columns, where an injected row is indistinguishable in kind from any other claim.

What this carrier trains

As a BAD example, this one timeline becomes a single training triple. The prompt is the history cut roughly ten rows before the TEST — the SIGNAL and IMPACT rows, and none of the resolution. The rejected continuation is the carrier's own real drift: more of the same, ending in the LATE clearance sessions the record actually recorded. The chosen continuation is not what happened at all — it is a caught-in-time TEST-then-EARLY graft, lifted from the GOOD template, so the model is trained to prefer the early catch over the real slide into a lifetime of facility visits. BAD carriers teach the correction, not the mistake; how the triple is assembled belongs to three-pattern learning. (The secondary recipe that ultimately worked trained on the chosen continuation alone, so the rejected path here is defined but not consumed — an honesty point the training article carries in full.)

This same carrier is where the FAKE construction picks up. FAKE asks the identical lesson from a cut roughly twenty rows earlier than this one — ending about thirty rows before the TEST rather than ten — which is how "surface it sooner" is literally trained, and the only way earliness is ever measured.

See also