Materials · corvid/triage-round2.md

# Triage of round 2 findings (orchestrator)

Judged against the record, Jordan's writing preferences, and how much each point matters to a reader. R = role, E = evidence selection, V = visual.

| # | Finding | Decision | Reason |
|---|---|---|---|
| R1 | The two placeholders decide the case | Keep, and move to the handoff | Only Jordan can answer. The handoff says what the letter needs if the answers are weak. |
| R2 | The analogy misses that graders may not score the route at all | Apply | A specialist would catch it. Recast as two possible sources of disagreement, which also matches the approach paragraph. |
| R3 | The approach samples only disputed cases, has no third error type, and ignores task length | Apply | These are cheap to fix and make the approach read as expert. They are proposals, not claims about Jordan. |
| R4 / E3 | Human-baseline problem unanswered. Postdoc work reads as a caveat | Apply | Both reviewers raised it. Tie the postdoc pool to Corvid's staff-only baseline without claiming any agent-versus-human study. |
| R5a | "Which side is in error" drifts from the blog | Apply | Use the blog's own wording. |
| R5b | "Often" is softer than "about a third" | Set aside | Rule 3. Corvid knows its own figure, and "often" is accurate. |
| R6 | Restore the head of research's deploy-because-ready hook | Apply | It is the bridge to the Kestrel point. |
| R7 / U4 | Agreement statistics | Set aside | The record has none. Flagged to Jordan in the handoff. |
| R8 | Link Kestrel's validation against cancellation to checking scores against deployment outcomes | Apply | One clause, and it states the parallel the reader is meant to draw. |
| R9 | Compress the hidden-context anecdote | Apply | Cut to one clause. |
| R10 | Likely interview questions | Move to the handoff | Useful to Jordan, not to the documents. |
| E1 / V4 | Relay sits under Projects after all jobs and parses as a phantom job | Apply | Both reviewers raised it. Fold Relay and the rehearsal into the Independent Researcher entry. |
| E2 | Invited judge-trust talk missing from the letter | Apply | Reverses my revision 1 cut. Two reviewers independently judged it the most on-point outside recognition. |
| E4 | Add the JAP measurement-model clause to the letter | Set aside | No room on one page. The résumé carries it. |
| E5 | Put "300 messages" in the letter | Apply | Reverses the brief. Rule 3 allows a number when the reader needs the magnitude, and a measurement reader judges a validation partly by its size. |
| E6 | Lead the Kestrel entry with the measure rebuild | Apply | It is the relevant item, and moving it costs no space. |
| E7 | Summary line at top of résumé | Set aside | Risks stating fit (rule 2). Reordering the résumé does the same work. |
| E8 cuts | Walkthrough, "mentored informally", long rehearsal entry | Apply | Low signal for this role. Removing the walkthrough also removes the nonstandard "Other work" heading. |
| E8 cut | Repeated "not with people" | Set aside | It stops a reader from assuming Relay involved human studies. That misreading is the costliest one here. |
| E9 | End the letter on a sentence restating fit | Set aside | Rule 2. Instead, move the gap sentence earlier and end on the human-baseline paragraph. |
| V1 | Single-line job headers will misparse | Apply | Title on its own line. Employer and dates on the next, joined by a pipe and "to", not a dash (rule 6). |
| V3 | PhD and Python at the bottom of page 2 | Apply | Education and Skills move to the top. |
| V4b | "Papers" is a nonstandard heading | Apply | Rename to "Publications and working papers". Plain "Publications" would mislabel the two unpublished items (rule 8). |
| V5 | Weak heading hierarchy | Apply | Larger headings with a rule. Bold titles, italic employer and date lines. |
| V6a | Bold Jordan's name in author lists | Apply | It is a field convention and costs nothing. |
| V6b | Status words clutter the publications | Set aside | Rule 8 requires exact status. |
| V6c | No DOIs or code link | Set aside | The record has none. Asked of Jordan. |
| V7 | Page 2 has no name or page number | Apply | Footer on page 2. |
| V8 | Long lines, dense text | Partly apply | The cuts lighten page 1. Margins unchanged. |
| V9 | Letter's Relay paragraph too long | Apply | Split in two. |
| V parse | jordanhale.example may not register as a link | Apply | Add https://. |