Skip to content
MyDSHMy DSH Field Notes
中文
Independently maintained; not affiliated with DeepSeek. No association with mydsh.dev or other MyDSH sites.

One long article into multi-platform copy without altering a single fact

Recipe template · not yet tested on this site

Recipe / ContentTemplate · not runCompiled 2026-11-14
This page is a recipe template, not yet tested here. Run logs, tokens and time go in after re-testing; until then no estimated numbers are filled in.

When to use

One long article must become a short post, a microblog entry, and a video script. The fear is numbers shifting, conclusions inflating, and cases getting attached to the wrong claim.

Preparation

Host version
To fill (record the exact rc version when testing)
Model
To fill
Mode / permissions
Standard mode + modify-within-workspace is enough to start
Plugins required
None (add name & version after testing if used)
Input materials
Keep in a separate workspace; back up originals first

Full instruction

Read article.md and rewrite it into three pieces: a short-form post (under 300 characters), a microblog entry (under 140), and a 60-second spoken script (about 220 characters).

[Hard constraint: facts]
- Every number, date, person, place, organisation, and quotation in the original is preserved verbatim. No conversion, no rounding, no 'about'.
- Where the original says 'some', 'most', or 'research suggests', the rewrite must not upgrade it to 'all', 'everyone', or 'studies show'.
- A causal claim (A causes B) must not become a correlational one, and vice versa.

[Output]
End each piece with a 'fact check' section listing every number and proper noun in it, with its location in the original (paragraph number or the source sentence).

Requirements:
1. Each piece stands alone; never write 'as mentioned above'.
2. Platform differences show up in the opening and the length, not in fact density — all three carry the same facts.
3. If a platform's length cannot fit a key qualifier from the original (say 'only holds under X'), keep the qualifier and cut decoration rather than dropping it for brevity.
4. Report at the end: how many numbers per piece, and whether any claim could not be traced to the original.

Run log

ItemValueNote
Time spentto be tested
Resultto be tested
In / out tokensto be tested
Leftoversto be tested

Output

After a successful run, list files and screenshots here, and note whether originals changed.

Acceptance checklist

Tap to tick each item off as you verify.

  • Numbers in all three pieces match the original exactly, checked one by one
  • No upgraded phrasing such as 'all' or 'studies show'
  • Each piece has a fact-check section whose citations resolve
  • Key qualifiers survive in all three
  • Nothing appears in any piece that cannot be traced to the original

Failures & boundaries

  • The fact-check section is the point of this recipe. Remove it and you have ordinary rewriting with no way to verify
  • Platform limits collide with 'do not drop qualifiers'; rule 3 resolves it by keeping the qualifier. State it explicitly or the model decides for you
  • If the original is internally inconsistent (numbers that do not add up), the rewrite copies that contradiction three times. Proofread the source first
  • Blocked on: Runnable, but proving no fact was altered needs a real published long-form piece carrying real figures and names. Synthetic text cannot show whether a number gets quietly swapped.

Claim this recipe Back to recipes