King

[King @OBKing397127]

@emvila @m477lander Exactly this isn't a solana problem

[King @OBKing397127]

@hesaysdanielle @m477lander Faster slots are great until some ancient assumption decides otherwise

[King @OBKing397127]

@m477lander Solana getting faster is cool until every silent 400ms assumption starts cooking people’s bots

[Matt lander @m477lander]

Solana won my network poll, so I took the hint and spent some time looking through the latest network changes.

the obvious headline is SIMD-0525: the 350ms slot target is now live. Faster slots and the shorter blockhash window have already been covered. I wanted to see what this changes higher up the stack, so I went through the proposal, docs and public protocol code.

What I hadn't seen discussed much is what happens one layer above it:: the chain clock moved, while some app logic still reads time through the old 400ms assumption

The SIMD warns about this directly. anything measured in slots now covers less real time, and programs that treat slot distance as wall-clock time may need an update Static SDK constants also stay unchanged for now.

@switchboardxyz's Solana tutorial gave me the cleanest example:

`.max_age(30)` rejects data older than 30 slots and describes that as roughly 12 seconds..

At the 350ms target, those 30 slots are nominally 10.5 seconds. At 200ms, 6 seconds.

Same code. Same parameter. Different risk policy.

Switchboard is doing exactly what it says. If you meant 30 slots, the rule still does exactly what you asked. If you meant roughly 12 seconds, you don't have 12 seconds anymore.

I checked @kamino's public kLend code and found the same handoff. The path starts with Switchboard's last update slot, uses DEFAULT_MS_PER_SLOT to manufacture a timestamp, then checks that against a max age in seconds. the SDK constant is still 400ms, so against the 350ms target the nominal age comes out 14.3% higher.

At the 200ms target, it would read as 2x older than it really is. A 30-second policy would trip after roughly 15 seconds of wall time, without anyone changing the policy.

The deeper problem isn't the number 400. It's that a risk check silently translates between two clocks. And this is why "just use `Clock::unix_timestamp`" isn't the whole fix. the current time is in seconds, but this oracle update starts in slots Both sides of the check still need to speak the same clock

No exploit claim here. The bias is conservative: valid data can look stale sooner. This is an availability risk, not a direct value-extraction path. The important part is that the handoff exists in real code.

This is exactly why I keep banging on about the full execution path. The network can report the right slot. The oracle can report the right update. The check can run exactly as written. the final behavior still changes at the handoff between protocol time and real time.

This actually makes me like @Solana's staged rollout more. A faster clock is good infra, and every gate gives builders a real checkpoint to catch these assumptions before 200ms.

Before the next gate, grep for every `max_age`, staleness check, cooldown, retry and timeout expressed in slots. Did you mean slots or seconds?

@bw_solana, should apps stop converting between the two entirely: slots for protocol ordering, timestamped data for wall-clock policy? And if so, what should an oracle expose so apps never have to manufacture that timestamp themselves?

SIMD-0525:
https://t.co/GRZMxPDlti

Switchboard example:
https://t.co/HYhQE7X4Yw

SDK constant:
https://t.co/QtzmoQQItE

Public code example:
https://t.co/EFJ2bXV7NB
https://t.co/xm1gTjQEPc

[King @OBKing397127]

@TakashiLeinola @m477lander One weird failure can reveal an entire trust model

[King @OBKing397127]

@VuyaniVinqishe @m477lander everything can pass the usual checks and still go sideways

[King @OBKing397127]

@RecipesCountry @m477lander Its weird when the valid path is the dangerous one

[King @OBKing397127]

@aa626625 @m477lander Local policy keeps that veto close to the user

[King @OBKing397127]

@Davidyeremyants @m477lander yeah thats where i land too

[King @OBKing397127]

@m477lander no key theft is what makes this case so interesting to me

[Matt lander @m477lander]

x402 processed 75.4M transactions in the last 30 days. Agent wallets are here, and one question bothers me.

Caps and allowlists limit what an agent can do. They don’t prove the transaction matches the user’s intent.

Where should the final veto live?

https://t.co/wo6ehkrTCa

[King @OBKing397127]

@brhino18 @m477lander i bet the owner has hundreds of stories about her 😂

[King @OBKing397127]

@crispycrunchGEN @m477lander Eva just woke up

[King @OBKing397127]

@Riley_jacksonn @m477lander @Maya_Protocol

[King @OBKing397127]

@LuLualexq @m477lander haha

[King @OBKing397127]

@LuLualexq @m477lander she knows something is suspicious 👀

[King @OBKing397127]

@LucianaSierra06 @m477lander Check the pool before touching anything

[King @OBKing397127]

@Riley_jacksonn @m477lander @Maya_Protocol Curious whether the post mortem will focus on the root cause or mostly on the recovery

[King @OBKing397127]

@Johnkurrency @m477lander @Maya_Protocol Everyone passed individually and the protocol still got an F