[AvivYossef-starkware]

assertFeeFreeReturn's comment, and its maxFee error message, justify the guard with ledger[commitment], receive_and_bind crediting the GROSS burned amount, and the claim reverting INSUFFICIENT_CLAIMABLE.

None of those exist any more.

How the drift happened

commit	date	what
eeed7e03	2026-07-07	guard added — the ledger design was live, so the rationale was true
d77edee2	2026-07-08	claim-fold rewrite removes ledger[commitment] / receive_and_bind
3da04702	same day	adapts the test to the rewrite — but not the rationale
#703	2026-07-23	forces Fast finality globally, noting "the return leg is unaffected — it self-enforces Standard finality", taking the stale guard as given

So the comment has described a deleted mechanism for over a month.

What actually happens now

InboundAnonymizer snapshots balance_of, calls receive_message, and hands the pool after - before. The pool asserts only non-zero / note-exists / open-salt / not-already-deposited / token-match — no comparison to any proven or burned amount — and the proof carries no amount at all (the Open sentinel).

So a per-call CCTP fee does not revert the claim. It silently credits a SMALLER note.

What changes, and what doesn't

The guard stays exactly as-is. It is still correct, and still the right direction: a caller asking to return N must get N or an error, never N − fee. Only the reason changes, from "the claim would revert" to "the note would silently shrink".

Also recorded, from an investigation in the consuming repo:

• This costs nothing today. Circle sells no Fast tier from Polygon (standard attestation ~8s) and quotes 0 bps on both tiers for this route, confirmed against the live fee API. So there is no latency or fee being left on the table by refusing Fast here.
• "Fee-free" is live-quoted state, not a protocol constant. CCTP v2 documents a per-deployment TokenMessengerV2 fee switch that can impose a minimum fee on STANDARD transfers. If it is ever set for this route, this guard fails the leg closed rather than shrinking the note — which is the intended direction, and worth stating so nobody "optimizes" the check away.

Why it is worth a PR

This is load-bearing documentation on a fund-safety guard, and it has now demonstrably misled two independent readers: #703 above, and an audit in the consuming repo that built a "forcing Fast is dangerous because the proof pins the amount" argument on this paragraph before verification refuted it. The guard's behavior was never wrong; the explanation was, and the explanation is what people reason from.

The test file's header comment repeated the same false claim, so it is corrected too.

Verification

Comments and message strings only — no behavior change. The existing assertion is /fee-?free|maxFee|per-call fee/i, which the reworded message still satisfies.

Run locally in packages/bridge-core:

• tsc --noEmit — clean
• tsc (build) — clean
• vitest run src/core/returnIn.feeFree.test.ts src/core/returnIn.test.ts src/core/avnuPaymaster.test.ts — 43/43 passed

avnuPaymaster is included deliberately: it uses INSUFFICIENT_CLAIMABLE as a fixture for its felt-decoder, which is unrelated to the return path's ledger design and is intentionally left untouched.

---

This change is 

[cursor]

PR Summary

Low Risk
Comments and one error message only; no runtime or guard behavior change.

Overview
Documentation-only update for the assertFeeFreeReturn guard and the fee-free return test header. assertFeeFreeReturn logic is unchanged — still rejects maxFee > 0 and sub-Standard finality.

The rationale is rewritten because it still described the removed ledger / gross-amount / INSUFFICIENT_CLAIMABLE model. It now states that the folded claim credits whatever actually minted (InboundAnonymizer balance delta; proof has no amount), so a CCTP fee shrinks the pool note silently rather than reverting the claim. The maxFee error string matches that story.

New comment notes: Polygon route is effectively fee-free today (no Fast tier, 0 bps quoted); “fee-free” is live CCTP config, not a constant — if Circle enables a standard-tier minimum fee, failing closed at the return entry points is intentional.

^{Reviewed by Cursor Bugbot for commit a0b18ef. Bugbot is set up for automated code reviews on this repo. Configure here.}