1/🧵 What is a rollup? And how do you make sense of inboxes, sequencers, bridges, and proofs?
This thread breaks down "Rollups From First Principles" — a must-read by @jonastheis_ for anyone building or curious about L2s.
👇Let’s dive in.
scroll.io/research/the-eleme…
1
8
25
14,710
2/ Rollups are L2 protocols that inherit security and data availability from L1, but execute transactions off-chain for scale.
They rely on a few core components:
- Inbox
- Sequencer
- Prover
- Bridge
- L2 node
Each plays a unique role in maintaining the rollup’s state and integrity.
1
3
327
3/ Rollup flow at a high level
📨 Input: L2 tx data posted to L1 (calldata or blob)
⚙️ CDF: Derive L2 chain from L1 inputs
⚙️ STF: Apply tx to derive new L2 state
🧾 Output: Canonical L2 chain and state
The key? It’s deterministic. Every honest node arrives at the same result.
1
3
239
4/ 📪 The Inbox
It’s the gateway. All L2 inputs (L1 messages, batched L2 txs) go here. Usually sent directly to L1 or posted by a sequencer with some access control.
1
2
215
6/ 🌉 The Bridge
Handles asset transfers between L1 and L2.
Deposits: custody on L1 → mint on L2
Withdrawals: burn on L2 → prove to L1 → release funds
Two kinds:
🔴 Optimistic: fraud-proof window
🔐 ZK/validity: validity proof required
May 26, 2025 · 2:44 PM UTC
1
1
206
7/ The Prover
The brain of the bridge.
Generates:
💡 Validity proofs (ZK)
⚔️ Fraud proofs (Optimistic)
Critical for bridging, security, and sometimes compression (in SD-based rollups).
Decoupling sequencer & prover is hard, but vital for decentralization.
1
1
297
8/ L2 Nodes (Full + Light)
Full nodes: Read L1, run CDF + STF, propagate L2 txs.
Light nodes: Great for wallets. Minimal trust required, verify availability and correctness.
nodes track multiple views:
⏱️ Latest (sequencer view)
🧷 Safe (on L1, not final)
✅ Finalized (L1-finalized)
1
1
130
9/ Fast vs. Slow Path
🔄 Fast: Rely on sequencer pre-conf = low latency, low security
🛡️ Slow: Rely on L1-confirmed data = high latency, strong security
Rollups must support both for usability and censorship resistance.
1
1
119
10/ Key Tradeoffs in Design
- Data availability: Tx data (TD) vs. State diffs (SD)
- L1 awareness: Host-following vs. host-watching
- Latency vs. decentralization
- Compression vs. finalization time
Your design choices affect security, UX, and cost.
1
1
122
11/ Transaction Data vs. State Diff
TD = Submit all tx data → re-executeable, easier to decentralize
SD = Submit only result + proof → more efficient, but complex
Only ZK rollups can do SD, and they sacrifice transparency + complicate user inclusion.
1
1
108
12/ Based vs. Run-ahead Rollups
Based: L2 closely follows L1, reorgs together
Run-ahead: L2 can finalize ahead of L1, but may reorg later
It’s a spectrum, and design decisions here affect user trust, performance, and bridge logic.
1
2
108
13/ Users care about one thing: UX.
Low fees, fast txs, instant feedback. They don’t care about rollup mechanics.
Designers must abstract away complexity while preserving security. And that’s not trivial.
1
1
232
14/ Rollups aren’t “just L2s”. They’re intricate systems blending L1 trust, off-chain execution, and user-centric design.
we would like to thank @donnoh_eth and @toghrulmaharram for their helpful feedback on an initial draft of this post.
Follow us for more research content and read the complete article in our research blog.
scroll.io/research/the-eleme…
2
235