Create protocol_history_rule.md

design/finality_protocol/protocol_history_rule.md

@@ -0,0 +1,63 @@
+# L2 History Rule
+
+### Context
+
+- The Sequencer is linked to an Ethereum node and has a real-time view of the "canonical L1 chain". This view might be different from the view of the other Obscuro nodes.
+- The result of processing an L2 block (batch) depends on an L1 block because of cross-chain messages.
+- As a consequence, to process an L2 block, an Obscuro node needs the L1 block (and its entire history) that was linked with the batch.
+
+
+### Problem
+
+Given that an L1 block is not final in real-time, it is theoretically possible that the sequencer produces batches linked to non-canonical L1 blocks.
+In the current protocol, the sequencer will produce a sibling batch linked to the new canonical L1 block when this happens.
+This is fine if the validators have received all the l1 blocks because they can execute and validate each batch. 
+Based on what is canonical, the enclave will use the right calculated results.
+
+There is an extreme edge case (which could be used to attack Obscuro) where an L1 block was visible only to the sequencer and not to any of the validators. 
+If the sequencer has produced batches using this block as a reference, then the result of this batch can't be calculated by any validator.
+
+
+### Properties of the ideal solution
+
+- the requirement is that the network must progress, even in this case.
+- given this is an extreme edge case, the solution doesn't need to be optimised for it.
+
+Note: Without this "data availability" problem, an Obscuro node can just wait to receive all the L1 blocks it needs before processing an L2 batch that it received via p2p.
+
+To restate, some of the blocks might never arrive so this approach could freeze the network.
+
+
+## The Solution
+
+The solution has two elements:
+1. a rule for producing batches
+2. a rule for synchronising a validator with the sequencer
+
+### Rule 1
+
+**If multiple batches have the same height (number), the one with the highest sequence number is considered canonical.**
+
+There are multiple consequences.
+
+The sequencer cannot build on top of a batch chain that is non canonical. This means it must create a "twin" of a non-canonical batch, if that batch has suddenly become canonical.
+
+### Rule 2
+
+An Obscuro node receiving batches via p2p must wait for a "canonical" batch for which it has the l1 block.
+
+This generalises the simpler rule that doesn't handle the DA issue.
+
+The significance is that the node will wait until it has all the data required to process that sequence of batches.
+
+Another way to put it is that, in case the Validator is out-of-sync with the sequencer ( because they validator does not have an l1 block that the sequencer considered canonical), 
+the validator will wait until it is in sync, before moving forward.
+
+
+## Solution that uses this rule
+
+
+
+
+
+