Replay on Sui mainnet data

Same errors, right answers.

661 real failed transactions from DeepBook v3, Cetus CLMM, Bluefin Spot (CLMM), read through Tatum over 13 days. The first 435 taught Ticketwise. The last 226 were answered with memory off and on.

Memory off
0%

0 of 226 later tickets answered with what the team taught. Without memory it can only explain the abort code.

Memory on, from Walrus Memory
100%

226 of 226 answered from a teammate's correction, 226 with the right meaning.

Learning curve, from empty memory

Day one knows nothing. Day two answers almost everything.

The same 661 failures, played forward one day at a time against a fresh Walrus Memory namespace. Each day's tickets are answered only from corrections taught on earlier days; at the end of the day the team teaches each new error once. 595 of 661 tickets (90%) were answered from memory with 10 corrections. Tickets for an error nobody had taught yet count as unanswered.

0%
10-29+9 taught
100%
10-30
100%
10-31
100%
11-01
100%
11-02
100%
11-03
94%
11-04+1 taught
100%
11-05
100%
11-06
100%
11-07
100%
11-08
100%
11-09
100%
11-10
Per error

How the replay was run.

The failures are split by time, so nothing from the later part leaks into teaching. For each distinct error in the earlier part, one correction was saved to Walrus Memory, with its meaning taken from the contract's own source. The replay found 10 distinct errors, and 10 were taught. All 8 errors in the later part had already appeared earlier, so every later ticket had a correction to draw on. An error that only shows up later would stay unanswered until someone teaches it.

ErrorName in contract sourceLater ticketsAnswered from memoryTaught earlier
deepbook::big_vector#5ENotFound7676Yes
cetus::pool#11EWrongSqrtPriceLimit5454Yes
bluefin::clmm_math#3018EINVALID_FIXED_TOKEN_TYPE3838Yes
deepbook::balance_manager#3EBalanceManagerBalanceTooLow2323Yes
bluefin::pool#1009EInvalidPriceLimit2323Yes
bluefin::pool#1010ESlippageExceeds99Yes
bluefin::gateway#1033ECannotClaimZeroReward22Yes
bluefin::pool#1018ENonEmptyPosition11Yes

Run 2026-11-10 08:21 UTC against Walrus Memory on Sui mainnet. Script: scripts/run-scorecard.mts.