1. Read effects from metadata, not the envelope

The envelope is a request. The metadata is the outcome.

A transaction envelope is what someone submitted. By the time it’s in a ledger, a lot has happened that the envelope can’t tell you: who actually paid, what was actually charged, which entries were created, which accounts were touched by a contract.

Question	Wrong source	Right source	Evidence
Who paid the fee?	Envelope.SourceAccount()	The fee-bump fee account, i.e. the account debited in fee changes	cap15-fee-bump-pubnet-0001, IX003
How much was charged?	Envelope.Fee() (the bid)	TransactionResult.FeeCharged	cap15-fee-bump-pubnet-0001, IX003
Was a trustline created?	A changeTrust operation	A CREATED trustline entry in the operation’s changes	cap73-sac-trust-*
Which accounts were affected?	Envelope source + operation destinations	Ledger entry changes and events in meta	replay harness

Measured

The replay harness’s naive indexer reads effects from the envelope. It gets all four indexed cases wrong: the muxed ID and the fee-bump are incorrect, both SAC trustlines are dropped. The reference indexer reads metadata and gets all four right. So does wallet-backend, on all six fixtures.

Not every envelope read is wrong

An operation with no source of its own runs as the inner transaction’s source, fee-bump or not. So Envelope.SourceAccount() is the right fallback for an operation’s source. IX003 only fires inside fee logic for exactly this reason: a broader version of the rule flagged two correct fallbacks in wallet-backend.

The test is simple. If you’re asking “what did the submitter say?”, the envelope is fine. If you’re asking “what happened?”, read the meta.