Update Results Discussion

/update-results-discussion

The section exists and needs to change. This is the most frequently updated section in any manuscript: reviewers ask for extra experiments, data gets reanalysed, and drafts stall halfway through the figure list.

Assume the text was correct once. This is a revision, not a rewrite. In this section that rule is stricter than elsewhere: the researcher's interpretations are theirs, and you preserve them verbatim unless the data no longer supports them, in which case you flag it rather than replacing them.

Before anything else

1. Load the style rules. Invoke paper-writing-style. If absent, use the condensed rules in create-results-discussion.
2. Lock the language. Language of the user's first message, whole run. Output LaTeX is English.
3. Read the context. The section, all figures and captions, tables, the .bib, any data files, and the introduction and conclusions.

Pick the mode

State the mode before starting.

Where several apply, run B, then C, then A.

Mode A here is not the same as reconcile-against-the-paper elsewhere: this section is the paper's evidence, so what it reconciles against is the data.

Mode A: integrate new or revised data

1. Inventory what changed. List every value, figure and table that differs from what the section describes. Read the figures and data files; do not rely on the researcher's summary alone.
2. Report the consequences, in this order of severity:
   • An interpretation the new data contradicts. Most serious. Quote the interpretation, the old value and the new one, and stop. Do not rewrite the interpretation yourself; ask the researcher what the new data means. This is their science.
   • A comparison that no longer holds, for example a result that was better than prior work and now is not.
   • A value stated in the text that no longer matches the figure.
   • A new result with no block at all, needing a full four-class block written.
   • A figure whose number changed, breaking \ref ordering and the narrative sequence.
   • A result now removed but still discussed.
3. Check the downstream sections. New or changed results almost always invalidate something in the abstract, the conclusions or the introduction's gap. Name what is now stale and suggest /update-abstract, /update-conclusions and /update-introduction rather than editing them here.

Mode B: apply suggested updates

Reviewer comments on this section cluster into recognizable requests:

Then:

1. Enumerate every suggestion as a numbered item, splitting reviewer prose.
2. Classify each: direct, interpretive, conflicting, out of scope.
3. Apply, propose, resolve. Requests that need new scientific claims are not yours to satisfy: surface them and ask.
4. Account for every item, including the ones not applied and why.

Where a request conflicts with the house style, apply it and note the tension. The reviewer outranks the style guide.

Mode C: resume after a pause

1. Inventory the figure list and mark which figures already have blocks. This shows exactly where the draft stopped, and it is usually mid-figure-list rather than mid-sentence.
2. Run the information-class matrix over the blocks that exist. A stalled draft often has several blocks with description only, written quickly before the writer ran out of momentum.
3. Ask for what is genuinely missing, in batches of at most four: the values for undescribed figures, and the interpretation for each. Read the data files first so you only ask for what the files do not contain.
4. Match the existing voice: tense pattern, hedging level, caption style, how much background each block carries.
5. Check the narrative order still works. A resumed section commonly loses the thread of why the figures are in this sequence. Say so if the order no longer tells a story.

Do not restyle the blocks already written. If they have defects, say so and suggest /review-results-discussion separately.

Produce the revision

Emit the changed blocks as LaTeX fragments with information-class labels, and:

• Preserve the researcher's wording and interpretations.
• Emit only the blocks that changed, not the whole section, since it is long. Say clearly which blocks you are showing and leave the rest untouched.
• Mark each change: % UPDATED: retention now 96\% per Fig.~4, % ADDED: comparison block, reviewer 2 asked about Park 2024, % FLAG: this interpretation may no longer hold, see note above.
• % TODO: for every value or interpretation you could not obtain.

Show the fragments, then summarize. Finish by listing which other sections are now stale.

Offer to write it out

Only after approval.

Hard rules

• Never invent, estimate or infer a value. Not from a figure, not from a trend. This is the section where fabrication does the most damage.
• Never write or rewrite an interpretation the researcher did not give you. If new data breaks an old interpretation, flag it and ask. Supplying a mechanism is inventing science.
• Never invent a comparison citation.
• Never emit an em dash. -- only in numeric ranges.
• Never silently drop researcher text, an interpretation, or a suggestion.
• If nothing needs changing, say so and change nothing.