The evidence
The evidence gate
Deposits open, shrink to a single capped ticket, or close because of what a replay of real swaps says a vault would have kept.
Three numbers and a sign
A vault earns the pool fee on every swap that trades through its liquidity. It loses to informed flow that buys just before a price move and sells back after it, and it pays gas and swap impact each time the band moves. The replay tracks those three in annualised percentage terms and adds them up.
| Component | What it counts | Sign in the total |
|---|---|---|
| Fees | The pool fee on every replayed swap, at the share of liquidity the vault would have held. | Added |
| Markout | The value informed traders take back after their swaps, measured against the price a short time later. | Subtracted |
| Cost | Gas and swap impact for each recentre, spread over the year at the vault's size. | Subtracted |
| Kept APR | Fees plus markout minus cost. This is the number the gate reads. | The result |
The weaker half decides
The window is 30 days, split into two halves of 15. The first half selects the pool: among the token's v3 USDG pools carrying at least $1,000,000 of volume, the one with the best result wins. The second half is untouched during selection and reports the out-of-sample figure. The gate takes the lower of the two kept APRs.
This is deliberately asymmetric. A vault that kept 30% in a quiet fortnight and lost 4% in a busy one is recorded at -4%, not at the average. The state describes the worse half because that is the half a new depositor has not seen yet.
The three states
The thresholds are set in the keeper's evidence layer and mirrored in the web app. They are plain constants, not a score: the weaker half lands in one of three bands.
worst >= 0 open no cap
-20 <= worst < 0 reduced $25,000 per ticket
worst < -20 closed deposits refused| State | What it means | Where it is enforced |
|---|---|---|
| open | The weaker half cleared break-even, so both halves kept money for liquidity providers. | The vault's own deposit cap is the only limit. |
| reduced | One half lost money but stayed above the -20% floor. The vault keeps taking deposits with a smaller ticket. | App deposit form and prepare_deposit cap one transaction at $25,000 of Stock Token. |
| closed | The weaker half fell below the floor. A vault at this size did not keep money in the conditions the replay measured. | App deposit form and prepare_deposit refuse the deposit. The contract is untouched. |
What is held constant
The replay runs one strategy at $100,000 of capital: a plus or minus 5% band around Chainlink, recentred at 02:00 UTC on the day after each NYSE session, with slots skipped when the feed is older than 90,000 seconds. The window is 2026-09-17 to 2026-10-17, and 22 after-session slots were requested and kept.
Fees move with flow, so an open gate is a statement about the evidence, not a forecast. It says the strategy kept money on both halves of a real 30-day sample at this size. Thekept figures beside each vault on the vaults page are those same half-window results.
The gate today
Every vault the eligibility rule admits appears below with the state and sentence written by the evidence builder. Order follows the file, which sorts by 30-day fee APR.
| Vault | Gate | Weaker half | What the gate says |
|---|---|---|---|
| MSTR | closed | -28.5% | Closed: the weaker half kept -28.5% a year, below the -20% floor. |
| GME | closed | -45.5% | Closed: the weaker half kept -45.5% a year, below the -20% floor. |
| CRCL | closed | -114.2% | Closed: the weaker half kept -114.2% a year, below the -20% floor. |
| INTC | closed | -91.2% | Closed: the weaker half kept -91.2% a year, below the -20% floor. |
| MU | closed | -66.4% | Closed: the weaker half kept -66.4% a year, below the -20% floor. |
| AMD | closed | -65.2% | Closed: the weaker half kept -65.2% a year, below the -20% floor. |
| PLTR | reduced | -16.9% | Reduced: the weaker half of the 30-day replay kept -16.9% a year, between the -20% floor and break-even, across 53,897 swaps. Deposits are capped at $25,000 per transaction. |
| META | closed | -40.8% | Closed: the weaker half kept -40.8% a year, below the -20% floor. |
| USO | closed | -24.9% | Closed: the weaker half kept -24.9% a year, below the -20% floor. |
| SPCX | closed | -89.4% | Closed: the weaker half kept -89.4% a year, below the -20% floor. |
| TSLA | closed | -49.2% | Closed: the weaker half kept -49.2% a year, below the -20% floor. |
| NVDA | open | +1.8% | Open: the weaker half of the 30-day replay kept 1.8% a year, at or above break-even, across 814,927 swaps. |
| GOOGL | reduced | -5.6% | Reduced: the weaker half of the 30-day replay kept -5.6% a year, between the -20% floor and break-even, across 171,538 swaps. Deposits are capped at $25,000 per transaction. |
| AMZN | reduced | -6.8% | Reduced: the weaker half of the 30-day replay kept -6.8% a year, between the -20% floor and break-even, across 40,468 swaps. Deposits are capped at $25,000 per transaction. |
| AAPL | open | +2.5% | Open: the weaker half of the 30-day replay kept 2.5% a year, at or above break-even, across 95,275 swaps. |
| QQQ | reduced | -4.1% | Reduced: the weaker half of the 30-day replay kept -4.1% a year, between the -20% floor and break-even, across 64,813 swaps. Deposits are capped at $25,000 per transaction. |
| SPY | reduced | -0.5% | Reduced: the weaker half of the 30-day replay kept -0.5% a year, between the -20% floor and break-even, across 171,364 swaps. Deposits are capped at $25,000 per transaction. |
A closed vault still has a contract, a feed and a price. It simply does not take deposits while the gate says closed. When a later replay clears the floor, the same vault can reopen without a migration.
Sources
keeper/src/evidence.tsFloor, break-even, the $25,000 reduced ticket and the sentence the gate writes.
keeper/src/universe.tsThe eligibility rule that decides which tokens reach the replay at all.
keeper/replay/replay.pyThe 30-day run, the two halves, the $1,000,000 pool-choice floor and the 90,000 second staleness skip.
keeper/replay/SCHEDULE-CHANGE.mdThe 2026-09-17 to 2026-10-17 window and the 22 kept after-session slots.
web/lib/vault-universe.jsonThe live gate state, reason and weaker-half figure for each vault.