The A and B series verified the contracts. The C series is about the node — whether this public record can be reproduced from the chain alone, by anyone, without relying on the operator’s private database or API. This item (C-3) checks whether two independent nodes, running on separate VPS, ingest the same chain from an empty DB and reconstruct a public record that is deterministically identical — by comparing checksums. It is the answer to what the reviewer calls the “decisive” proof: §19 #13–15 (empty-DB historical reconstruction / deterministic export hash / independent re-execution matching across two nodes).

1. Claim (what is verified)

That two independent nodes on separate infrastructure, ingesting the same chain (Base Sepolia, chainId 84532) from an empty DB, reconstruct a public record that is deterministically identical. That is: the public record can be reproduced from the chain alone, without depending on the operator’s private DB / API / indexer (§19 #13–15).

2. Method

3. Verdict

• match Key fingerprint (total order over block/logIndex/txHash) 993db07…b339 matches on both nodes (309,672 rows). Zero gaps or alterations.
• match Content fingerprint (including the decoded action body) 82651606…1df57 matches on both nodes. Byte-identical down to the content.
• identical The reconstruction-and-hash code (listenForRawEvents.ts / RawEventsGatherer/index.ts / packActions.ts) has zero diff between the two commits. The version difference is only in the API/frontend/validator/test layers.

4. Receipt (public proof)

The two nodes’ setup. chainId is 84532 (Base Sepolia) on both. Only the startBlock differs, and node2’s history is a clean suffix of node1’s.

The reconciliation over the common range [41,803,823–43,181,982] (matching on both nodes):

To reproduce, take the common-range content fingerprint on each node’s PostgreSQL and compare (excluding node-local fields like createdAt, hashing only chain-derived content):

SELECT md5(string_agg(
  "blockNumber"||'-'||"logIndex"||'-'||"transactionHash"||'-'||"data"::text,
  '|' ORDER BY "blockNumber","logIndex","transactionHash"
)) AS contentfp, COUNT(*) AS n
FROM "RawEvent"
WHERE "chainId"=84532 AND "blockNumber">=41803823 AND "blockNumber"<=43181982;
# on both nodes: contentfp == 82651606cdbe919be65de5b227f1df57 / n == 309672

You can also verify that the hashing code is identical between the two nodes’ commits:

git diff 7ff0ff4 e2074718 -- \
  client/src/services/RawEventsGatherer/listenForRawEvents.ts \
  client/src/services/RawEventsGatherer/index.ts \
  client/src/utils/packActions.ts
# -> no diff (ingest/decode/hash path is identical in both versions)

For reference: because the startBlock differs, the raw parentHash heads naturally differ (node1 0x3332defe…f27622 / node2 0xcfbf2b38…cca77278). Comparing over the common range confirmed this difference is solely due to startBlock.

5. Notes & limits

The scope is the on-chain action stream (CAW/RECAW/LIKE/FOLLOW, etc.). DMs and images are off-chain (DM relay / local disk, Filebase) and outside this deterministic chain — an honest open door. Dependence on the Base L2 sequencer is also a known open door.

Determinism is relative to the open-source client’s decode/hash logic. The two nodes run different commits, but that path is confirmed identical in source, so the claim “reproducible by an independent operator using the public client” holds.

For a stronger form, re-syncing node2 from an empty DB with the same startBlock=41,548,637 as node1 would make the raw parentHash heads match from the start too (time-consuming, optional). This item’s verdict stands without it, on the content match and code identity above.