l2-preconfirmation-reliability: 2026-08-30

README.md

@@ -41,14 +41,20 @@ violations and no checks is not the same as a run with no violations and thousan
 
 ## Before you build on this
 
-- `lag_blocks` is only meaningful where `safe_tag_plausible` is true. One chain's endpoint served
-  a stale safe tag through much of the period, which yields a lag of tens of millions of blocks.
-  The raw numbers are published uncorrected so the endpoint's own inconsistency stays visible;
-  filter on the flag before using the column.
+- `lag_blocks` is only meaningful where `safe_tag_plausible` is true. One chain's endpoint has
+  served a stale safe tag intermittently, giving a lag of tens of millions of blocks against
+  another chain's few dozen, then recovering to a normal figure hours later. It is kept rather
+  than dropped, because an endpoint's own inconsistency is a fact about running on public
+  infrastructure, and because it comes and goes you cannot exclude a chain once and be done.
+  The flag is false on exactly the affected rows; filter on it, not on the chain name.
 - The unsafe head is sampled every few seconds, so a block that was proposed and replaced between
   two samples is invisible. This undercounts violations and cannot overcount them.
-- Two chains, both OP-stack. Sequencer behaviour is an implementation choice, so this does not
-  generalise to rollups built differently.
+- Nine chains, ALL OP-stack. That is nine independent sequencer operators, which is enough to
+  compare them against each other, but it is still one rollup architecture: nothing here
+  generalises to a rollup built differently.
+- Safe-head lag varies enormously between them even at rest, from tens of blocks to several
+  thousand, and one endpoint reports a lag of tens of millions. Compare a chain against its own
+  history rather than against another chain's absolute level.
 
 Partitions are parquet, one file per collection window, under `dataset/YYYY/MM/`. This repo carries a FIXED 7-day sample (2026-08-25 to 2026-08-31) so you can check schema, coverage and quality before asking for more. It does not advance, so there is nothing to gain by re-downloading it. The full history is held privately, available on request.