Analyse capped-waterfall end-of-run undeployed cash

[miohtama]

Summary

• Re-runs the capped-waterfall cross-chain vault backtest (scratchpad/xchain2/08-backtest-capped-waterfall.ipynb) against the fixed trade-executor CCTP planner (#1549). The backtest now completes end-to-end (386/386 CCTP bridge trades succeed, +6.9% total return).
• Investigates why the strategy leaves a growing amount of undeployed cash towards the end of the run (idle reserve rises to ~24% of equity vs the ~10% design target).
• Adds a notebook-static-server verification step (status-code + notebook-title check) to .claude/docs/notebook-server.md.
• Commits the portfolio asset allocation chart used in the analysis.

See the PR comment below for the redemption-delay pinning breakdown, example vaults/trades, and the allocation chart.

[miohtama]

Redemption-delay pinning → undeployed cash

The end-of-run idle cash (reserve climbs to 24.0% of equity by the last cycle, vs the ~10% target from allocation_pct = 0.90) is driven in large part by async vault redemption delays. When the strategy wants to rotate out of a lagging vault, the redemption only requests the withdrawal; the cash lands several cycles later. During that window the capital is neither in a productive position nor available to redeploy, so it accumulates as idle USDC.

Mechanism (trade-executor alpha model)

Two related effects, both keyed off async ERC-7540 / Ostium vault settlement:

1. carry_forward_non_redeemable_positions() — each cycle, positions in vaults that report can_redeem=False (redemption-delay window) or that have an in-flight settlement are pinned at their marked value and carried forward. That value is subtracted from the deployable target:
   deployable = 0.90·equity − locked_non_redeemable − cash_buffer.
   So the strategy cannot rotate out of a laggard while its redemption is delayed.
2. _cap_buys_by_async_sell_proceeds() — an async vault sell's proceeds are not available same-cycle, so this cycle's buys are scaled down to synchronous cash only. The withheld capital stays in reserve until the redemption settles.

Evidence from the run

• 157 of 344 cycles carried at least one pinned (non-redeemable) vault position, across 10 distinct vaults.
• Pinning-flag rows: settlement_pending ×273 (async deposit/redeem in-flight), cannot_redeem ×29 (redemption-delay veto), max_adjust_too_small+settlement_pending ×6.
• Final-cycle balance sheet: equity $106,903, idle reserve $25,664 (24.0%), invested $84,698 (79.2%), of which only $602 is physically pending settlement at that instant — the rest of the idle cash is the cumulative churn withheld across settlement windows. Only 15 of 20 position slots are used.

Example vaults exhibiting the behaviour

Vault	Chain	Cycles pinned	Mean pinned USD	Max pinned USD	Total missed redemption USD
DeTrade Core USDC	Base	75	10,401	11,367	0
Moon Digital AM USDC	Ethereum	69	10,135	11,360	0
Ostium Liquidity Pool Vault	Arbitrum	39	9,352	11,343	4,903
Hub Capital USDC vault	Ethereum	32	10,435	11,366	0
Plutus Hedge Token	Arbitrum	28	8,911	8,932	249,497
USDn2	Arbitrum	25	9,772	12,130	0
Syntropia USDC Core	Ethereum	18	10,326	11,385	0
Syntropia Boosted	Ethereum	12	8,634	11,377	0

Plutus Hedge Token is the clearest case: the strategy tried to redeem a cumulative $249,497 from it over 28 cycles but was repeatedly blocked by the redemption delay.

Example trades exhibiting the behaviour

Async settlement trades still in-flight (unsettled executed_at) at run end — capital tied up mid-redemption/deposit:

Trade	Vault	Chain	Action	Value USD	Async in-flight
#2098	DeTrade Core USDC	Base	sell (redeem)	2,120	✅
#2097	Moon Digital AM USDC	Ethereum	sell (redeem)	1,100	✅
#2094	Ostium Liquidity Pool Vault	Arbitrum	buy (deposit)	602	✅

Repeated partial sells over consecutive cycles show the strategy chipping away at positions it cannot fully redeem in one go, e.g. Moon Digital AM USDC: #2029 (2026-02-24, $4,967) → #2097 (still in-flight at run end); Hub Capital USDC vault: #2032 (2026-02-25, $351) → #2046 (2026-02-27, $492).

Allocation chart — vaults and undeployed cash

Portfolio asset weights over the run. The grey USDC band is undeployed cash; note how it widens towards the right-hand (late) side of the run as more of the book gets tied up in delayed async-vault redemptions.

---

Note: redemption-delay pinning is additive to a second constraint — the per-position vault-capacity caps (per_position_cap_of_pool_pct = 0.20, max_concentration_pct = 0.12) limit how much can be placed into the shallow top-CAGR vaults (~$18k "discarded because of lit liquidity" at the final cycle). Both push idle cash above the 10% design floor late in the run.