Op‑Eddie

An op‑ed AI pre‑editor · diagnoses, never composes

Desk copy · v7

What this is not

Op‑Eddie doesn’t replace a human editor and isn’t a ghostwriter. It diagnoses; it doesn’t compose. If the words on the page stop being yours, the tool has failed — no matter how high it scored you.

Three tiers, by who can trust them

MUST‑FIXObjective, checked in your browser. Not opinions.

YOUR CALLQuestions, never verdicts. You decide.

ONLY YOUDid it happen? What’s the venue’s policy?

The draft

Piece name Draft

        {{ liveCount }}
        Links count only as a URL or [text](url) — doc hyperlinks arrive as plain text.
      

Venue

Word limit (the venue’s, not a default)

Who drafted it

Same model grading Verify sources What’s New

No live web search in this prototype — verification and What’s New run on model recall and will say so. Absence of a match is not proof of novelty.

How this is graded  ↓

Every finding is sorted into three tiers by who can trust it.

Must‑fix — objective checks run in your browser, no model involved: word count against the venue limit, sourced claims missing a link, a linked stat whose source isn’t named in the text (“According to…” — M10), the same source linked twice (M14), placeholders and TKs, sentences opening on a digit, link text that runs a whole clause.

Your call — surfaced as questions, never verdicts: what genre the draft really is (M6), one thesis carried through, a timely peg, whether the argument would survive a swapped hook (M7), lede shown or told, the strongest counterargument answered, a specific ask (M5), overclaims, fudged or unanchored numbers (M14), first‑person running past the 25–30% share unless it’s a personal essay (M12), close‑up/medium/long‑shot balance (M12), typographic shouting where a measured voice carries further (M17), the darling question when the piece runs over its count (M18), a source stretched past what it holds, AI cadence, and — for solution‑forward pieces — all four elements.

Only you — what no checker can reach: whether the anecdote happened, the venue’s live policy and limit, claims that assert a negative.

Module ledger: every module reports into a named slot — fired, quiet, or did not report. A module that fails is shown, never silently dropped.

Readiness: one open must‑fix means not ready. Judgment items only means mechanically clean, calls still yours. Nothing flagged is a floor, not a green light.

Nothing on the desk yet

Paste a draft and run the pass. The findings land here, sorted by how much you should trust them.

            —— words · limit
            —— must‑fix
            —— your call
            —— only you
          

Where your draft stands

Must‑fix

Your call

Only you

Start here{{ startHere }}

Self‑evaluation · Same model drafted and graded this. Trust the must‑fix items; treat judgment and only‑you flags as suspect and get a second reader.

Copied — select all to copy again

Context from the judgment pass

{{ c.label }}: {{ c.value }}

Judgment pass diagnostic — must‑fix items below are still complete. {{ diag }}

Margin notes

MUST‑FIX Objective — the tool is sure. Just do these.

Nothing here — mechanically clean.

YOUR CALL Judgment — decide for yourself. Not verdicts.

ONLY YOU CAN ANSWER The tool can’t know. Did it happen? What’s the policy?

Nothing here.

Module ledger

Every check reports; none can silently drop.

              {{ s.label }}{{ s.status }}
            

A pre‑editor pass, not a substitute for a human editor. Every recommendation traces to a named editorial standard, not general AI wisdom. The revision, and the voice, are yours.