This is the lead paragraph. Use it to set up the piece in one or two sentences — who this is for, what they’ll learn, and why it matters right now. Keep it to roughly 30 words so the reader can decide fast.
TL;DR
- One-sentence point the whole piece hangs on
- Second point a skimmer would want to leave with
- Third point — the specific thing we’re doing / offering
- Optional fourth — the one stat or proof point that earns trust
IN THIS PIECE
- A subhead that signposts the section
- Data, numbers, and proof
- How we’d approach it
- The takeaway
A subhead that signposts the section
Body copy goes here. This template is intentionally simple — a single column centered at 800px, generous line-height, and clear typographic hierarchy so drafts can flow in without additional design work. Images, pull quotes, and inline links are all supported.
Write in short paragraphs. Break up long sections with subheads. When a sentence is doing too much work, split it in two. The goal is a reader who keeps scrolling.
A smaller subhead for nested sections
Use H3s for points inside a larger argument. Links get the rebrand accent color — styled inline so readers can scan for related resources. Footnotes and callouts live in their own modules below.
$ example.sh
$ snorkel eval --task terminal-bench+ --model frontier-v1
→ loading 1,247 tasks from Agentic Coding Data Series
→ running multi-step CLI evaluations…
Accuracy: 38.2% (± 1.4)
Pass@1: 0.41 Pass@5: 0.63Code language: JavaScript (javascript)- First idea, stated cleanly
- Second idea, with one concrete example
- Third idea that closes the loop
- Optional fourth — only if it earns its spot
“
A single sentence that captures the argument so well it could stand alone.
Return to normal body copy after the pull quote. The reader should feel the pace — a punch of emphasis, then back to argument.
✦
KEY INSIGHT
One non-numeric idea worth spotlighting inline — the thing the reader should leave with, in one sentence.
“A sharp quote from an expert — someone outside Snorkel — that backs up the argument with real-world experience.”
Expert Name
Title, Company
Data, numbers, and proof
If the piece hinges on a specific number or claim, give it its own line. Readers skim, and a confident stat is the kind of thing that stops them.
< 40%
Placeholder stat — swap in the number that anchors the piece.
How we’d approach it
- Start from the research question, not the data you have on hand.
- Design the eval before you ship the feature.
- Iterate with expert review — not just auto-graders.
- Ship the learnings alongside the model.
DO
- Concrete, scoped advice
- A second do that earns its place
- A third, if it’s really worth saying
DON’T
- The common mistake, named plainly
- A second trap that’s easy to miss
- A third, only if it differentiates
| DIMENSION | THE USUAL WAY | SNORKEL’S WAY |
|---|---|---|
| Data source | Scraped / synthetic | Expert, task-grounded |
| Eval signal | Automated graders | Human + automated |
| Feedback loop | Model team only | Shared with researchers |
| Time to insight | Weeks | Days |
The takeaway
Close the piece with a single clear idea the reader should walk away with. Not a summary — a point of view. If it fits, hand off to the next action: a related read, a demo request, or an internal link.
EDITOR’S NOTE
Updated Apr 15, 2026 — added the latest benchmark numbers and a new expert quote. Use this slot for corrections, updates, disclosures, or author context.
How we arrived at this
A collapsible block for research-backed pieces. Explain the data, sample size, time window, model versions, and anything else a critical reader would ask before trusting the claim. Keep it short — link out for the full write-up.
SOURCES
- Citation or source link, with publication and year.
- Second citation — useful for data-heavy pieces or research-backed claims.
- Third citation, optional.