diff --git a/BRANCH-NOTES.md b/BRANCH-NOTES.md
new file mode 100644
index 0000000..7458e8b
--- /dev/null
+++ b/BRANCH-NOTES.md
@@ -0,0 +1,120 @@
+# Sulkta-Coop/pallas — `feat-aux-data` branch notes
+
+This is a vendored fork of [txpipe/pallas](https://github.com/txpipe/pallas)
+based on the `v0.32.1` tag. It exists to fill in three upstream `// TODO`
+markers that block real-world Conway-era transaction building.
+
+The fork is consumed via `[patch.crates-io]` in
+[`Sulkta-Coop/aldabra`](http://192.168.0.5:3001/Sulkta-Coop/aldabra)'s
+workspace `Cargo.toml`. Once these changes are accepted upstream, the
+patch entry gets dropped and aldabra goes back to vanilla crates.io.
+
+## What's added beyond `v0.32.1`
+
+### 1. `pallas-txbuilder` — `auxiliary_data` field on `StagingTransaction`
+
+Upstream had `// pub auxiliary_data: TODO` on the struct and
+`auxiliary_data: None.into(), // TODO` in the conway builder. Without
+this, **CIP-20 / CIP-25 / CIP-68 metadata can't ride a tx built with
+pallas-txbuilder.**
+
+Added:
+- `pub auxiliary_data: Option<Vec<u8>>` field (opaque CBOR bytes —
+  `pallas_primitives::AuxiliaryData` doesn't impl `Eq`, which the
+  rest of `StagingTransaction` requires).
+- Builder methods `.auxiliary_data(cbor_bytes)` and
+  `.clear_auxiliary_data()`.
+- `conway::build_conway_raw` now decodes the bytes via
+  `AuxiliaryData::decode_fragment` and plumbs them in. The existing
+  `auxiliary_data_hash` computation block (already in upstream) runs
+  after assignment and populates the body's hash automatically.
+
+Tests added:
+- `auxiliary_data_round_trips_through_build` — encodes a CIP-25
+  shape, attaches, builds, decodes, asserts byte-equivalent
+  re-encode + matching `auxiliary_data_hash`.
+- `no_auxiliary_data_means_no_hash` — negative path.
+
+### 2. `pallas-txbuilder` — `certificates` field on `StagingTransaction`
+
+Upstream had `// pub certificates: TODO` and
+`certificates: None, // TODO` in the conway builder. Without this,
+**stake registration / delegation, pool registration, DRep ops, and
+Voltaire-era cert types can't be built.**
+
+Added:
+- `pub certificates: Option<Vec<Vec<u8>>>` field (opaque CBOR per
+  cert, same reason as `auxiliary_data`).
+- Builder methods `.add_certificate(cbor_bytes)` and
+  `.clear_certificates()`.
+- `conway::build_conway_raw` decodes each entry via
+  `Certificate::decode_fragment` and plumbs them into
+  `TransactionBody.certificates` via `NonEmptySet::from_vec`.
+
+Tests added:
+- `certificates_plumb_through_to_tx_body` — encodes a
+  `StakeRegistration(AddrKeyhash([7;28]))`, builds, decodes,
+  asserts the cert round-trips byte-for-byte.
+- `no_certificates_means_none` — negative path.
+
+### 3. `pallas-addresses` — `pub fn new` on `StakeAddress`
+
+Upstream defined `pub struct StakeAddress(Network, StakePayload)` with
+**unexported tuple fields** and no `new()` constructor — so external
+callers can't construct one directly. `ShelleyAddress::new` exists
+already; this just matches the pattern.
+
+Without this, computing a wallet's reward (stake) address
+required round-tripping through a `ShelleyAddress` and `TryFrom` —
+clunky and indirect.
+
+## How aldabra consumes this
+
+`Sulkta-Coop/aldabra/Cargo.toml` has:
+
+```toml
+[patch.crates-io]
+pallas-codec      = { git = "...Sulkta-Coop/pallas.git", branch = "feat-aux-data" }
+pallas-crypto     = { git = "...Sulkta-Coop/pallas.git", branch = "feat-aux-data" }
+pallas-primitives = { git = "...Sulkta-Coop/pallas.git", branch = "feat-aux-data" }
+pallas-traverse   = { git = "...Sulkta-Coop/pallas.git", branch = "feat-aux-data" }
+pallas-addresses  = { git = "...Sulkta-Coop/pallas.git", branch = "feat-aux-data" }
+pallas-wallet     = { git = "...Sulkta-Coop/pallas.git", branch = "feat-aux-data" }
+pallas-txbuilder  = { git = "...Sulkta-Coop/pallas.git", branch = "feat-aux-data" }
+```
+
+All seven pallas crates we depend on are patched against the same
+commit so cargo's version graph resolves consistently.
+
+## Upstream PR status
+
+**Not yet submitted.** Plan:
+
+1. ✅ Land the changes locally on the fork.
+2. ✅ Verify no upstream tests broken (`cargo test --workspace`
+   passes — confirmed 2026-05-04).
+3. ✅ Verify aldabra builds + tests against the fork (88 unit tests
+   green).
+4. ☐ Open PR against `txpipe/pallas` on github.com.
+5. ☐ Once upstream merges, drop our `[patch.crates-io]`.
+
+The PR will probably need to be split into three separate PRs (one
+per change) for upstream review ergonomics:
+
+- `pallas-txbuilder: thread auxiliary_data through StagingTransaction → Conway build`
+- `pallas-txbuilder: thread certificates through StagingTransaction → Conway build`
+- `pallas-addresses: pub fn new on StakeAddress`
+
+## Change discipline going forward
+
+If we add more fork-only patches (e.g. for upcoming withdrawals /
+voting_procedures / proposal_procedures TODOs), each goes in a
+separate commit on this branch so the eventual upstream PRs are
+clean to extract.
+
+All commits on this branch must:
+- Have a self-contained subject explaining what TODO it fills in.
+- Pass the full pallas-txbuilder test suite (`cargo test -p pallas-txbuilder`).
+- Add at least one positive + one negative test for the new code path.
+- Not depend on Sulkta-specific naming or assumptions — these
+  patches need to be upstream-mergeable as-is.