ACTUARIAL VIZASL · Fleet Solvency · Sorry Reserve · Enrichment Triangles

SOLVENCY EQUATION

fleet_healthy = available_capacity > unresolved_liability
available_capacity = time + compute + attention + operators + GPU-hours + money
unresolved_liability = sorry_reserve + enrichment_reserve + raincheque_PV + silo_hazard_cost
solvency_ratio = available_capacity / unresolved_liability
TARGET: solvency_ratio > 1.3 (30% margin above liability)

SORRY RESERVE — Case Reserve + IBNR

CASE RESERVE (REPORTED)

Active sorries currently open in the fleet. Each sorry = unresolved proof obligation or enrichment gap. Development factor applied per school:

E. coli school: 1.8× development factor (fast resolution)
Yeast school: 2.1× (creative enrichment takes longer)
LAB school: 1.3× (characterized, predictable)
Acetic school: 1.5× (audit closes on evidence)
Methanogen school: 3.0× (deep archive, slow retrieval)

IBNR — Incurred But Not Reported

Enrichment debt that has been incurred (proof gap exists) but not yet surfaced as a formal sorry. Estimated via:

IBNR = case_reserve × development_tail_factor
Fleet-wide development tail: 1.4×

Total reserve = case_reserve + IBNR
Chain-ladder projection: 12-week horizon
Bornhuetter-Ferguson method for immature cohorts

ENRICHMENT DEBT TRIANGLE — 12-Week Development

ORIGIN WEEK	Age 0	Age 1	Age 2	Age 3	Age 4	Age 5	CDF
W-12	12	20	25	27	28	28	1.00
W-11	14	22	28	30	31	33*	1.07*
W-10	10	17	21	23	25*	27*	1.17*
W-9	16	26	32	35*	38*	40*	1.25*
W-8	11	18	23*	25*	27*	29*	1.32*
W-1 (current)	8	14*	17*	19*	20*	22*	2.75*

* = chain-ladder projected values · CDF = cumulative development factor · green = mature · orange = developing · red = immature

SILO HAZARD RATES — Survival Functions

SILO	SCHOOL	λ (HAZARD RATE)	S(90d) SURVIVAL	STATUS
msi01	E. coli	0.002/day	0.834	STABLE
yone	LAB	0.003/day	0.763	STABLE
forge	Yeast	0.004/day	0.698	STABLE
msclo	LAB	0.002/day	0.834	STABLE
lilo	E. coli/Yeast	0.005/day	0.638	MONITORING
pcdev	Yeast	0.003/day	0.763	STABLE
AKS (cloud)	All	0.001/day	0.914	MANAGED

S(t) = exp(−λ·t) · Exponential survival model · γ₁-anchored hazard calibration

RAINCHEQUE PV — Present Value of Deferred Items

γ₁-ANCHORED DISCOUNT CURVE

Raincheques are deferred enrichment items — things that should happen but are not yet urgent. Their present value uses a γ₁-based discount rate:

r(l) = γ₁ / (l+1) where l = adelic layer (0-13)
L0 discount rate: 14.134% per period (highest priority, least discounted)
L7 discount rate: 1.766% per period (deep archive, heavily discounted)

PV(raincheque) = Σ [item_value × exp(−r(l) × t)] over all deferred items

Fleet raincheque PV estimate: ~12 enrichment-days of deferred work

RUIN PROBABILITY — P(fleet cannot recover)

P(RUIN | 30 days)

0.02

Assumptions: current solvency ratio maintained, no catastrophic GPU failure, Kay operational. Primary risk: NAS failure + msi01 simultaneous outage.

P(RUIN | 90 days)

0.08

Accumulation of deferred raincheques + enrichment debt growth. Mitigated by: baobab reserve pattern, methanogen cold storage, AKS redundancy.

P(RUIN | 365 days)

0.21

Long-horizon risk driven by: operator unavailability, cloud cost accumulation, sorry reserve growth exceeding capacity. Reserve-based persistence is the primary mitigant.

KEY ACTUARIAL REPOS — External Reference

mynl/aggregateKCF 9

R actuarial aggregate loss modeling. Chain-ladder + Bornhuetter-Ferguson. Direct reference for fleet enrichment triangle methodology.

open-source-modellingKCF 8

Actuarial modeling open source toolkit. Reserve estimation + survival analysis. Fleet silo hazard rate calibration reference.

Elsemary/GMMKCF 9

Gaussian Mixture Models for fermentation school classification. DESEOF GMM auto-classifier for sorry routing by school.

casact (CAS actuarial)KCF 8

Casualty Actuarial Society open data + methods. Development triangle standards. Solvency II framework reference for fleet reserve modeling.