Backtesting and Validation

This page reports the current validation evidence for Cardinal's launch model: historical replay, stress simulation, product-specific test vectors, charting-tool parity, and commercial user discovery.

Scope

Dimension	Current launch scope
Products	Carry Perp and Negative Rate Hedge
Markets	sUSDe<>USDT and wstETH<>WETH
LP architecture	LP-intermediated pools, one pool per market
Market data	Aave V3 historical lending, borrow-rate, and loop-yield data
Product charting data refresh	Through June 3, 2026

Launch Parameters

These constants are fixed at launch. Values are shown per market; the evidence that justifies each group follows the table.

Negative Rate Hedge

Parameter	sUSDe	wstETH
Premium floor breach_base	0.249%	0.218%
Premium load premium_load	0.50	0.50
Premium basis	expected negative-carry claims × (1 + premium_load) / 0.90	expected negative-carry claims × (1 + premium_load) / 0.90
Coverage leverage L	1x–10x	1x–10x
Premium routing (LP / treasury)	90% / 10%	90% / 10%
Payout ramp to full coverage	30 days	30 days

Each tick the buyer pays the actuarially-fair cost of the coverage they hold — expected negative-carry claims × (1 + premium_load) / 0.90 — which scales with coverage leverage L and is charged continuously while the policy is open (breach_base × L × (1 + premium_load) / 0.90 is a calm-window floor reference; breach_base is set per market from how often that market's loop carry has historically gone negative). The previous current-carry band is not used as a hard launch pricing cap because it can underprice policies exactly when negative-carry claims are being generated. Pricing the premium to expected claims plus a load keeps the LP underwriting margin positive after the 90/10 LP/treasury split. premium_load is defined LP-net: with premium_load = 0.5 the LP retains a +50% margin over expected claims after the 10% treasury routing, so the buyer-gross premium is expected claims × (1 + premium_load) / 0.90. In the Track C historical replay the LP's net insurance result was positive in both markets (+2.31% sUSDe / +0.93% wstETH). Coverage leverage, the 90/10 routing, and the 30-day ramp were exercised by nine launch test vectors (tv10–tv18), all passing with zero invariant breaches — i.e. the policy accounting never broke its rules: a policy's payout buffer never went negative, cumulative payouts never exceeded the yield credited into that buffer, and gas tanks never went negative.

Carry Perp

Parameter	sUSDe	wstETH
Global notional cap	$76.8M	$322.6M
Shadow-drawdown scale s_L	65	65
Launch top tier	5000x	15000x
Performance fee	35%	35%
Liquidation triggers	equity < 5% of deposit, or shadow drawdown ≥ deposit	(same)
Entry fee	max(0, carry) × notional / (365.25 × 24)	(same)

Build Plan V3 sets s_L = 65 at launch for both markets (raising wstETH from the PR38 60 to match sUSDe). The launch-tier calibration establishes the kill gradient — every position eventually liquidates (a 2–3 day median life at the top tier), with the LP net positive per cycle.

LP pool

Parameter	sUSDe	wstETH
Effective loop leverage	3.05×	4.34×
Pause / resume threshold (NAV drawdown)	10% / 3%	10% / 3%

Loop leverage is the Aave V3 carry loop the pool runs on the LP's behalf; at these settings the historical replay earned a mean loop APR of 10.97% / 5.16%. The pool auto-deleverages if peak-to-trough NAV falls past 10% and redeploys once it recovers to within 3%.

Executive Summary

Area	Result
LP return view	Integrated launch-scope LP ROE = loop yield (gross) + NRH (net of claims) + Carry Perp genuine cash edge: sUSDe 11.33%, wstETH 7.02%. Carry Perp uses the LP's direct synthetic cash P&L (deposit kept on liquidation; the accrued carry and perf fee inside the captured equity are paper MTM, never cash). Legacy Track C decomposition (7.44% / 3.13%) retained below for provenance.
Carry Perp calibration	5,000 trials per launch tier; day-2 liquidation: 68.80% sUSDe and 41.80% wstETH; LP net per $100 user deposit: $159.95 sUSDe and $186.59 wstETH
Negative Rate Hedge validation	9 launch-scope test vectors passed; 9 replay rows produced 0 simulator-enforced invariant breaches; 144 pricing-study runs produced 0 invariant-breach forms
Stress simulation	11 scenarios x 20 trials x 2 markets; historical-bootstrap rows show 0% breach rates; depeg, drift, and sustained-negative-carry rows identify launch-control cases
Commercial user discovery	Direct institutional conversations indicate approximately $10M average covered position size, 6 months to 2 years position duration, and 20 immediately addressable targets

Confidence levels classify current evidence coverage and remaining implementation work.

Confidence area	Current confidence	Basis	Remaining validation
Historical replay outputs	Medium for launch-scope LP projection, high for source artifact integrity	committed replay artifacts, schema checks, hash verification, and per-pool result tables	fresh Build Plan V3 two-product replay and final benchmark ratification
Product accounting invariants	High for launch-scope simulator cases	9/9 Negative Rate Hedge test vectors passed; 0 invariant breaches across replay rows	contract-level implementation tests
Carry Perp launch calibration	High for replayed launch tiers	5,000 trials per tier and positive LP net per cycle in both launch markets	live rollout monitoring against waitlist demand and LP capacity
Stress characterization	Medium-high for scenario measurement	440 total scenario trials across both markets and explicit breach-rate reporting	production control validation for depeg, drift, and emergency paths
Production launch readiness	Medium until pre-launch items close	replay, stress, simulator, and user-discovery evidence is available; implementation evidence remains open	hourly production carry/TWAP validation, external audit, governance parameter bounds, operational controls, user disclosures, force-close-on-Emergency logic, wstETH depeg-ladder parameterization

Validation Summary

Evidence surface	Count / sample	Result recorded
Historical LP replay	2 launch pools	0 historical LP principal breaches
Stress simulation	11 scenarios x 20 trials x 2 markets	breach rates reported by scenario below
Carry Perp calibration	5,000 trials per launch tier	positive LP net per cycle in both launch markets
Negative Rate Hedge test vectors	9 launch-scope vectors	9 pass / 0 fail
Negative Rate Hedge replays	9 replay rows	0 simulator-enforced invariant breaches
Pricing study	144 pricing runs	0 invariant-breach forms
User discovery	Direct institutional conversations	$10M average covered position size discussed; 20 addressable targets identified

Historical LP Replay

Historical replay runs the launch configuration against observed Aave V3 rate paths. ROE is annualized LP return on equity. The integrated LP return is the sum of three separate cash streams, each counted exactly once:

1. Loop yield (gross) — carry the LP earns on its own deployed capital.
2. Negative Rate Hedge — premium net of claims (cost-covering, LP-net after the 90/10 split).
3. Carry Perp — genuine cash edge — recomputed from the LP's direct synthetic cash P&L, not a broad fee fraction: the LP keeps a liquidated user's deposit and returns the equity of the rare horizon survivors. A position's accrued carry and its perf fee live inside that equity and are the LP's short-side paper MTM (the LP is counterparty to the user's leveraged carry bet); that MTM is extinguished at the kill and is never LP cash, so it is excluded. It is also not loop carry, so there is no loop double-count. At ~100% liquidation the genuine cash ≈ the forfeited deposit.

This is why the combined is neither the naive sum of the gross bars (which would double-count the paper MTM) nor a loop-net subtraction (which over-removes loop carry the LP keeps). The mean loop APR (10.97% / 5.16%) is the raw loop return over the Track C window; the launch loop leg below is the charting-tool rolling-year loop yield.

Pool	Venue days	Replay days	Paused days	Mean loop APR	Integrated LP ROE	Max drawdown	LP principal breaches
sUSDe	597	297	0	10.97%	11.33%	-1.13%	0
wstETH	1,111	811	0	5.16%	7.02%	-0.45%	0

Integrated LP ROE by stream (annualized contribution). Carry Perp is the genuine cash edge = the Carry Perp anchor × the direct-cash ratio genuine_cash / (fee + capture), where genuine_cash is the deposit kept on liquidation net of survivor payouts (the accrued carry and perf fee inside the captured equity are paper MTM): sUSDe 3.17% × 0.625 = 1.98%; wstETH 1.83% × 0.536 = 0.98%.

Pool	Loop yield (gross)	Negative Rate Hedge	Carry Perp (genuine cash)	Integrated LP ROE
sUSDe	7.71%	1.63%	1.98%	11.33%
wstETH	4.32%	1.72%	0.98%	7.02%

Legacy Track C anchor decomposition is preserved below for provenance. It uses older Track C product anchors and excludes the non-launch perpetual-spread-options line from launch scope. Full-config is shown only for reference.

Pool	Track C loop-net leg	Negative Rate Hedge	Carry Perp	Transitional launch-scope estimate	Perpetual spread options (non-launch)	Full-config ROE
sUSDe	2.62%	2.31%	2.51%	7.44%	16.97%	24.41%
wstETH	1.29%	0.93%	0.91%	3.13%	7.17%	10.30%

This applies the 2026-06-08 decisions: NRH premium_load is LP-net after the treasury split, NRH capacity is the entitlement-share rule (not the Track C operating anchor), and the Carry Perp leg is the genuine cash edge (paper MTM stripped). The integrated engine sums the three streams once each, so the loop↔Carry-Perp double-count and the loop-net over-subtraction are both resolved. A full two-product replay would still tighten the Carry Perp genuine-cash fraction and the (negligible at ~100% liquidation) horizon-survivor adjustment.

Pool P&L below is full-config (it includes the non-launch perpetual-spread line); the transitional launch-scope pool P&L is the Negative Rate Hedge + Carry Perp + loop-yield columns.

Pool	Cumulative written	Outstanding at end	Pool P&L (full-config)	Treasury accrued
sUSDe	$326.8M	$96.8M	$1.99M	$196.9K
wstETH	$874.9M	$82.8M	$2.29M	$222.3K

Pool	Perpetual spread P&L (non-launch)	Negative Rate Hedge P&L	Carry Perp P&L	Loop-yield P&L
sUSDe	$1.38M	$187.9K	$203.9K	$213.5K
wstETH	$1.59M	$206.4K	$201.4K	$286.8K

Carry Perp Calibration

Window note. These figures come from the launch charting-tool calibration: the Carry Perp Monte Carlo runs on each market's realized loop carry through 2026-06-03, at the Build Plan V3 launch s_L = 65 for both markets. They match the same configuration the interactive charting tool exposes (link above), so the numbers here are what reviewers see live. The carry series is the realistic loop spread (lst_yield − borrow), which includes the historical negative-rate days.

Carry distribution inputs:

Market	Data window	Carry days	Mean carry	Median carry	Min carry	Positive-day share	Inactive negative-entry share
sUSDe	2024-07-28 to 2026-06-03	676	1.00%	0.82%	-21.19%	69.82%	30.18%
wstETH	2023-03-01 to 2026-06-03	1,191	0.50%	0.43%	-19.52%	92.36%	7.64%

Launch-tier simulation outputs:

Market	Launch tier	s_L	Trials	Total liquidation	Day-2 liquidation	Median liquidation	Crossed 2x before liquidation
sUSDe	5000x	65.0	5,000	100.00%	68.80%	2 days	10.44%
wstETH	15000x	65.0	5,000	100.00%	41.80%	3 days	18.88%

s_L = 65 is held common across both markets per Build Plan V3. The PR38-era ≥ 60% day-2 liquidation aspiration is met at sUSDe (68.80%) but not wstETH (41.80%, 3-day median); wstETH's slower kill gradient is the accepted consequence of the common V3 s_L, not a calibration failure — every position still fully liquidates and the LP is net positive per cycle.

LP economics per $100 user deposit, over a position's full lifecycle (a 2–3 day median to liquidation), from the 5,000-trial calibration on each market's carry distribution shown above:

Market	LP fee	LP liquidation capture	LP net
sUSDe	$21.24	$138.70	$159.95
wstETH	$30.38	$156.21	$186.59

Actor Outcomes

Projected profit/loss by actor across three regimes. LP and Carry-Perp cells are drawn from the replay / stress / calibration tables; the Hedger cells use the current cost-covering premium and the replay claim rows.

Actor	Benign (rates stay positive)	Negative-rate event	Severe depeg
LP	integrated launch-scope ROE (loop gross + NRH net + Carry Perp genuine cash): +11.33% / +7.02%	+10.15% sUSDe / -2.31% wstETH annual ROE (full-config stress)	breach risk (-6.29% wstETH at a 20% depeg)
Hedger	pays the streaming premium, collects ~nothing → net cost ≈ premium	the policy pays out → net gain	orderly settlement, forced-close, 0 LP breaches
Carry-perp user	negative expected value (the LP captures $159.95 / $186.59 per $100 deposit); ~10% (sUSDe) to ~19% (wstETH) ever peak above 2x before liquidation	rational users are not expected to open new Carry Perp positions into negative carry	unaffected — separate margin

Hedger sizing example. In the current sUSDe charting replay, a $1M-notional policy at L = 5 prices near $28.6k/yr under the LP-net expected-claims × (1 + premium_load) / 0.90 formula. In a mature 30-day stretch where carry sits near −2%, that same policy receives on the order of $8k in claims before buffer constraints — so the buyer is net-positive over the event and net-cost in calm periods, which is the hedge working as intended.

Stress Simulation

Stress simulation parameters:

Parameter	sUSDe	wstETH
Scenarios	11	11
Trials per scenario	20	20
Days per trial	800	800
Scenario trials	220	220
Pause threshold	10%	10%
Resume threshold	3%	3%
R1 percentile	90%	90%
R5 fee routing	20%	20%
R5 target fraction	15%	15%

Breach rate is the share of the 20 trials in a scenario with at least one LP principal-breach event.

Scenario	sUSDe mean ROE	sUSDe mean DD	sUSDe breach rate	wstETH mean ROE	wstETH mean DD	wstETH breach rate
historical_bootstrap	10.81%	-1.98%	0%	5.04%	-0.87%	0%
fat_tail_2x	12.38%	-1.91%	0%	5.10%	-0.92%	5%
fat_tail_5x	17.78%	-1.49%	0%	6.27%	-0.77%	5%
fat_tail_10x	25.95%	-2.26%	30%	9.25%	-0.64%	5%
regime_shift_mid	11.39%	-9.25%	35%	5.50%	-4.16%	20%
sustained_neg_carry_30d	10.63%	-2.29%	0%	5.05%	-1.40%	5%
sustained_neg_carry_90d	10.15%	-1.86%	0%	-2.31%	-9.08%	100%
sudden_depeg_10pct	4.50%	-6.67%	100%	-3.00%	-9.55%	100%
sudden_depeg_20pct	2.23%	-8.42%	100%	-6.29%	-12.94%	100%
sudden_depeg_30pct	0.29%	-9.64%	100%	-8.92%	-16.54%	100%
slow_drift_180d	8.11%	-3.47%	65%	-4.74%	-9.54%	100%

Negative Rate Hedge

Launch-scope test vectors:

Vector	Case	Status	Invariant breaches
tv10	top-up, no state change	PASS	0
tv11	voluntary close returns gas tank	PASS	0
tv12	forced close emergency	PASS	0
tv13	premium floor / load accounting	PASS	0
tv14	cost-covering premium covers expected claims	PASS	0
tv15	premium debits through negative-carry periods	PASS	0
tv16	perpetual, no rollover at day 180	PASS	0
tv17	lapse on empty gas tank	PASS	0
tv18	premium debits during stale oracle	PASS	0

Pricing basis note. The current charting branch prices NRH from expected negative-carry claims times (1 + premium_load) / 0.90, charged continuously while the policy is open. breach_base (0.249% sUSDe / 0.218% wstETH) remains a market-specific floor / calibration reference, but the old current-carry band is retired as a hard pricing cap because it can leave LPs below expected claims after the 90/10 routing split.

Historical replay summary:

Replay	Days	Requested policies	Admitted policies	Rejected policies	Claims paid	Premium earned	NAV change	Forced closes	Invariant breaches
historical_susde	676	32	32	0	$50.9K	$38.3K	+17.84%	0	0
historical_wsteth	1,220	56	4	52	$0.0K	$0.0K	-14.64%	0	0

Adversarial replay summary:

Replay	Requested policies	Admitted policies	Claims paid	Premium earned	NAV change	Forced closes	Invariant breaches
s1_single_30d	20	20	$58.2K	$17.6K	+4.01%	0	0
s2_two_30d_6mo_apart	32	32	$116.8K	$40.0K	+5.30%	0	0
s3_two_30d_1yr_apart	40	40	$197.5K	$66.5K	+7.67%	0	0
s4_four_30d_stacked	36	36	$137.4K	$46.7K	+2.78%	0	0
s5_sustained_12mo	40	24	$62.5K	$14.9K	-0.73%	2	0
s6_depeg_crisis_orderly_settlement	24	20	$62.0K	$13.6K	-0.25%	7	0
s7_emergency_tier4_force_close	20	20	$57.3K	$13.6K	+2.99%	3	0

Replay totals:

Replays	Requested policies	Admitted policies	Claims paid	Premium earned	Forced closes	Invariant breaches
9	300	228	$742.7K	$251.2K	12	0

Pricing study note:

The May 27 pricing study remains useful as historical mechanism evidence, but its variable-band forms are superseded for the premium formula decision. The current charting branch prices policies from expected negative-carry claims times (1 + premium_load) / 0.90, then routes premium 90/10 to LP/treasury. Do not use the old pricing-study "LP ok vs baseline" column as direct underwriting-margin evidence for the launch formula.

Charting Tool

The full economics — realized loop yield, Carry Perp liquidation behavior, and Negative Rate Hedge premium / buffer / payout — are explorable interactively:

→ cardinal-product-calculator.vercel.app

It runs the launch configuration on Aave V3 data through 2026-06-03 (sUSDe loop 3.05×, USD-denominated; wstETH loop 4.34×, ETH-denominated). Per-market parameters, coverage tiers, and stress behavior are all adjustable in the UI.

Commercial User Discovery

Item	Observation
Covered position size	Approximately $10M average position size discussed with direct institutional users
Position duration	6 months to 2 years
Immediately addressable targets	20 institutional targets identified
Carry Perp rollout mechanism	Waitlist signups during rollout preparation

Measurement Boundaries

Result class	Boundary
Historical replay	Observed Aave V3 rate paths during the stated windows
Stress simulation	Specified scenario parameters and 20 trials per scenario
Carry Perp calibration	Launch-tier replay artifacts and 5,000 trials per tier
Negative Rate Hedge tests	Launch-scope vectors and replay rows in the referenced artifacts
Live deployment	Realized rates, LP participation, user demand, liquidity depth, and production controls

Evidence Package

Evidence area	Artifact root
Historical LP replay	output/phase4_track_c_historical_2026-04-21/
Stress validation	output/phase4_track_c_combined_2026-04-21/
Carry Perp launch-tier calibration	labs/product_calculator/ and output/product_calculator/data.json (charting-tool MC, s_L = 65)
Negative Rate Hedge replay and test vectors	output/negative_rate_hedge_per_policy_2026-05-27/
Negative Rate Hedge pricing study	output/negative_rate_hedge_pricing_study_2026-05-27/
Product charting tool	output/product_calculator/ and labs/product_calculator/

Reviewer Commands

python3 labs/benchmarks/check_artifact_schemas.py
python3 labs/benchmarks/verify_participant_econ_v6_artifacts.py
python3 -m unittest tests.test_product_calculator_guardrails