diff --git a/TODO.md b/TODO.md
index 1e16946..6a24c71 100644
--- a/TODO.md
+++ b/TODO.md
@@ -577,13 +577,83 @@
       out and `repo clone` won't parse one either, so this is a loud error
       where it used to be a silently wrong DID. Fixing it properly means
       asking the knot or the appview what repo that path names
-- [x] `pr create` — records `source: {branch}`, the shape every pull request
-      written by anything other than atgc carries. It is not a claim the
-      branch was pushed; it is the only thing that can match a pull request
-      back to the branch it came off, which `pr view` needs
-- [ ] `pr create` — branch-based PRs proper (push the branch, and let
-      `source` mean what a knot would mean by it); attach commit bodies to
-      the PR body
+- [x] `pr create` — records `source: {branch}`. The reasoning that stood
+      here was wrong on a checkable fact, so it is corrected rather than
+      quietly dropped: this is **not** "the shape every pull request written
+      by anything other than atgc carries". Tangled's own web UI passes
+      `pullSource = nil` for a patch-based pull
+      (`appview/pulls/create.go`, `handlePatchBasedPull`), and only its two
+      branch shapes write a `source` at all. What the field buys atgc is
+      real — it is the only thing matching a pull back to the branch it came
+      off, which `pr view` needs — but the appview reads it as a claim and
+      not as a hint. See the entry below
+- [ ] `pr create` pushes the branch by default, with `--patch-only` for the
+      shape it has today. Read off Tangled core at master `1adde466` rather
+      than guessed, so the ground is settled before the work starts.
+
+      **There is no patchless pull.** `lexicons/pulls/pull.json` requires
+      `rounds`, and every round requires `patchBlob`; the ingester refuses a
+      round without one (`appview/ingester.go`, "missing patchBlob in round
+      %d") and `Pull.Validate` refuses a pull with no submissions. "Let the
+      knot diff it" is real, but it names who *computes* the patch, not a
+      record that lacks one. The patch is still gzipped, uploaded as a blob
+      and put in the round, exactly as today.
+
+      **Three shapes, told apart by `source` alone**
+      (`appview/models/pull.go`: `IsPatchBased`/`IsBranchBased`/
+      `IsForkBased`). No `source` is patch-based; a `source` naming this repo
+      is branch-based; a `source.repo` naming another is fork-based. The
+      appview never checks a `source` against reality — ingest parses it and
+      believes it.
+
+      **Which is why today's default is a bug and not merely an omission.**
+      atgc writes `source: {branch}` on every pull, including ones whose
+      branch was never pushed anywhere, so tangled.org classifies every atgc
+      pull as branch-based and then acts on it: a `/{repo}/tree/{branch}`
+      link that 404s (`pages/.../pullHeader.html`), a `resubmitCheck` asking
+      a knot for the head of a branch it has not got (`pulls/single.go`), and
+      a web resubmit routed through `resubmitBranch` into a `repo.compare`
+      that cannot succeed. `pr checkout` already defends itself against
+      exactly this and says so in its comments; the web UI has no such
+      defence. atgc is claiming a shape it does not implement.
+
+      **So implement the shape rather than retract the claim.** Default:
+      push the branch, then write `source`. That makes the claim true, turns
+      the dead tree link live, and lets a browser resubmit an atgc-opened
+      pull. `--patch-only` keeps today's behaviour for the cases that need
+      it — no push access on the target, a branch not worth publishing, a
+      patch assembled from something that is not a branch — and it must then
+      write **no `source` at all**, because a `source` is exactly the claim
+      that a branch is there.
+
+      Steps for the default path, every one against an endpoint that already
+      exists: push the branch to the target repo's knot over SSH, which is
+      the access `atgc key add` already grants; `GET <knot>/xrpc/
+      sh.tangled.repo.compare?repo=<repoDid>&rev1=<target>&rev2=<source>`,
+      which is a public query needing no scope and no session; refuse an
+      empty `formatPatch`, as the appview does; gzip `formatPatchRaw`,
+      upload it, `putRecord` with `source: {branch}`. Taking the patch from
+      the knot instead of computing it locally is the half worth insisting
+      on — it is the same text the web UI would have produced, so a pull
+      resubmitted from a browser and one resubmitted by atgc agree, rather
+      than differing by whitespace nobody can see until the diff is read.
+
+      Two things deliberately out of scope. **Fork-based pulls** are the
+      harder shape and want their own entry: `sh.tangled.repo.hiddenRef` on
+      the *fork's* knot under service auth, push access on the fork, and the
+      compare run there rather than on the target — and the appview only
+      recognises a fork it already knows (`db.GetForkByRepoDid`). And
+      **merge is unaffected either way**: the knot never merges a ref, only
+      applies a patch (`pulls/merge.go` feeds `CombinedPatch()` to
+      `repo.merge`), so none of this changes what landing a pull does.
+
+      What it costs: `pr view` matches a branch to a pull through
+      `source.branch` (`cmd/pr/read.rs`), so under `--patch-only` that match
+      is gone. Decide where a local branch name lives when it is not a claim
+      about a knot — a trailer in the patch, beside `Change-Id:`, is the
+      obvious candidate and costs nothing
+- [ ] Attach commit bodies to the PR body, which used to ride on the end of
+      the entry above and is unrelated to any of it
 - [x] `pr list` — your own pulls straight from your PDS merged with Bobbin's
       sh.tangled.repo.listPulls (api.tangled.org, ATGC_BOBBIN to override);
       --state/--limit/--remote and --source auto|pds|bobbin; tolerates old