What a trace tells you that a draft does not
Every WriterDean run stores each step with its prompt, output, model, tokens, and cost. Here is how to read one and what to do with it.
When a draft comes back and something is off, the first question is where it went wrong. Was the research thin? Did the outline miss the angle? Did the copy edit flatten a good sentence? Without a record of each step, you are guessing.
A WriterDean trace is that record. Open any piece of content and click the trace tab. You will see a timeline of steps, one per skill in the chain, with the inputs and outputs of each.
Reading a blog post trace
A blog post runs through these stages, in this order:
- Brand Context loads the current version of your brand document.
- Content strategy turns the brief into an angle, an outline, and a target reader.
- Copywriting drafts against the outline.
- AI SEO and SEO audit check the draft against on page rules and adjust headings, links, and length.
- Schema produces the structured data for the page.
- Copy editing tightens the prose.
- The humanizer runs last.
Each step shows the skill name and version, the hash of the skill file it loaded, the model it used, tokens in and out, and the cost in dollars. Click a step and you see the full prompt that was assembled and the full output.
Where problems usually show up
Weak angle. If the draft feels generic, look at the content strategy step. The outline is there. If it reads like every other article on the topic, the fix is in the brief or in your brand context, not in the draft.
Missing proof. If the draft makes claims without support, look at the brand context step. The proof points section is what the copywriting step is allowed to use. If it is empty, the model had nothing to cite.
Flattened voice. If the draft sounds right after copywriting and wrong after copy editing, diff those two steps. The editor lets you compare any two step outputs side by side.
Cost per piece
The trace totals cost at the bottom, in dollars for the model calls and in credits for the words produced. The breakdown by stage tells you where the dollars went. Drafting and humanizing use the stronger model and cost more. Research and extraction use the fast model and cost little.
If a piece is expensive, it is almost always because the gate sent a stage back for a second attempt. The trace shows those retries as separate entries with the gate result that triggered them.
Traces are for your team too
Traces are visible to everyone with access to the content. Reviewers use them to see why a draft says what it says. Admins use them to spot skills that fail their gate too often. And when someone asks how the article was made, you can show them.