Address the unresolved factual corrections from yperbasis's review (CHANGES_REQUESTED) on PR #21500. The two new docs pages live at docs/site/docs/fundamentals/architecture.md and docs/site/docs/fundamentals/database.md. Two of his six points are already fixed; the FOUR below remain and must be corrected. These are text-only edits — do NOT restructure the pages.

1. --batchSize example is backwards and mis-framed (docs/site/docs/fundamentals/database.md, "Tuning knobs" section). The current text says: "controls how much state Erigon buffers in RAM before flushing to MDBX. Default is balanced; lower it (e.g. --batchSize 1G) if chaindata/ grows unexpectedly." This is wrong on two counts:

   • The default --batchSize is 512M (see node/cli/flags.go:48), so --batchSize 1G raises it, it does not "lower it".
   • --batchSize is the execution-stage RAM buffer; it has nothing to do with chaindata/ growing. That rationale is dubious and should be dropped.
     Reword to accurately describe --batchSize as the execution-stage in-memory buffer (default 512M) that is flushed to the DB, e.g. raising it (--batchSize 1G or higher) can speed up execution-heavy sync at the cost of more RAM. Make the example direction consistent with the description.

2. Mermaid diagram has a factually wrong edge (docs/site/docs/fundamentals/architecture.md, the mermaid block in "At a glance"). The line TxPool <-- private gRPC --> Caplin is incorrect: the TxPool does NOT talk to Caplin (the consensus layer). Its gRPC relationship is with the core/Sentry, not the CL. Remove this Caplin↔TxPool edge. (Note: an earlier Copilot review suggested ADDING such an edge — that suggestion was wrong; follow yperbasis here and remove it.) Keep TxPool -.reads.-> Datadir. Ensure the diagram still renders and no dangling/duplicate node definitions remain.

3. Snapshot sizing does not add up (docs/site/docs/fundamentals/database.md, "What does it cost on disk?" code block). The listed directories sum to ~1.13 TB (accessor 120 + domain 300 + history 280 + idx 430 GB) but the block says snapshots TOTAL 2.3 TB. The gap is the unlisted block/transaction .seg files. Either add the missing rows (e.g. blocks/transactions segment files) so the numbers reconcile to ~2.3 TB, or add a short note that the breakdown is partial and the remainder is block/transaction .seg data. Keep it honest and consistent.

4. Likely dead link (docs/site/docs/fundamentals/database.md, "Storage engine: MDBX" section). The MDBX link https://github.com/erthink/libmdbx points to a repo that was taken down. Erigon vendors github.com/erigontech/mdbx-go. Replace the link target with https://github.com/erigontech/mdbx-go (keep the visible link text "MDBX"). Note onBrokenLinks only validates internal links, so the build won't catch external dead links.

After editing, regenerate the llms artifacts if the build requires it and ensure npm run build (run from docs/site/) passes with onBrokenLinks/onBrokenAnchors: 'throw'. The llms.txt, llms-full.txt, and their docs/site/static/ copies should reflect the corrected page text.