Light

👍 💎 🌟 👑 💐

[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

[Light @AshleyCastle4]

@m477lander this is peak “it worked until the network upgraded” energy

[Light @AshleyCastle4]

@0xmaaya @m477lander Makes you wonder how many protocols still hardcode the old 400ms assumption.

[Light @AshleyCastle4]

@ColeBereznak @m477lander Exactly. Valid data starts looking stale sooner and liquidations can fire earlier than expected.

[Light @AshleyCastle4]

@m477lander slots moving but the apps still thinking old time… this is how you get unexpected liquidations

[Light @AshleyCastle4]

@VedahamPapi @m477lander Yeah, once the same system does both you’re already in the danger zone

[Light @AshleyCastle4]

@Melanson_2 @m477lander Exactly. Proposal and verification have to be separate jobs

[Light @AshleyCastle4]

@m477lander Local policy layer that the agent can’t touch makes the most sense to me.

[Light @AshleyCastle4]

@Melanson_2 @m477lander Caps limit damage. Simulation shows effects. Neither answers “is this what I actually wanted?”

[Light @AshleyCastle4]

@m477lander MetaMask and Coinbase improvements are good steps, but yeah… still just capability, not intent.

[Light @AshleyCastle4]

@bradleykinsman @mayrayaretgueva @m477lander The scary part is how clean it was. Pipeline private keys now.

[Light @AshleyCastle4]

@mayrayaretgueva @m477lander Yeah that hit different. No seed phrase needed, just clever prompting.

[Light @AshleyCastle4]

@LeeThomas08 @m477lander Good

[Light @AshleyCastle4]

@ColeBereznak @m477lander True, there’s rarely just one solution.

[Light @AshleyCastle4]

@Riley_jacksonn @m477lander Curious to hear the alternatives

[Light @AshleyCastle4]

@Hermoge59073298 @m477lander Who watches the watchers is the real question every time.

[Light @AshleyCastle4]

@IuryFabricio @m477lander Yeah, intent check at the wallet is the real control point. Caps just reduce the blast radius.

[Light @AshleyCastle4]

@ThomasvilleWI @m477lander Exactly. Limits help, but they don’t solve the intent problem at all.

[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