Tutorial: Build a dispute timeline

Goal

Combine article revision history with talk page activity to build a timeline of an editorial dispute — when it started, how intense it was, whether it was resolved, and how talk page discussion correlated with content changes.

What makes a dispute visible in the data

Three mechanical signals, together, form a dispute signature:

Reverts — edits undone by another editor (revert_detected)
Edit clusters — 3+ edits in a tight time window (edit_cluster_detected)
Talk page activity — talk edits near an article edit (talk_page_correlated), threads, replies and activity spikes (talk_thread_opened, talk_reply_added, talk_activity_spike)

refract analyze and refract export emit all of these. The talk events come from the ten talk-page edits that precede the last analyzed revision.

Reverts with no talk page activity and rapid edit turnover suggest edit-warring. Reverts alongside talk edits and slower turnover suggest deliberation. The events give the counts; the reading is yours.

Steps

1. Analyze a range of the article

refract analyze "Climate_change" --from 1374299339 --to 1383409984 --json

The range is 20 revisions from 2026. Reverts, sentence modifications and talk correlation are reported at every --depth. Without --from and --to the command reads the whole history.

Use the article's own title. "Global_warming" is a redirect to "Climate change" on English Wikipedia, and Refract analyzes the redirect page's history, not its target's.

2. Analyze the talk page

refract analyze "Talk:Climate_change" --since 2026-08-01 --json

Talk pages are fetched by prefixing the page title with Talk:. The CLI treats a talk page as an ordinary page: it reports section and sentence changes per talk revision, not threads or replies.

3. Browse the timeline

refract explore "Climate_change" --from 1374299339 --to 1383409984

Opens the local explorer on the same range, with every event type on a timeline.

refract visualize "<page>" --format mermaid --all prints a Mermaid diagram of a page's events. It has no range flags and reads the whole history, so keep it for pages with short histories.

4. Export for quantitative analysis

refract export "Climate_change" --from 1374299339 --to 1383409984 --format ndjson > events.jsonl
refract export "Talk:Climate_change" --since 2026-08-01 --format ndjson > talk-events.jsonl

5. Correlate reverts with talk page activity

import json
from collections import defaultdict

# Load article reverts
article_reverts = []
with open("events.jsonl") as f:
    for line in f:
        event = json.loads(line)
        if event["eventType"] == "revert_detected":
            article_reverts.append(event)

# Load talk activity
talk_days = defaultdict(int)
with open("talk-events.jsonl") as f:
    for line in f:
        event = json.loads(line)
        day = event["timestamp"][:10]
        talk_days[day] += 1

# Correlate: reverts per day vs talk activity per day
revert_days = defaultdict(int)
for event in article_reverts:
    day = event["timestamp"][:10]
    revert_days[day] += 1

# Days with high reverts but low talk → potential edit-warring
# Days with high reverts and high talk → active deliberation
for day in sorted(set(revert_days) | set(talk_days)):
    reverts = revert_days.get(day, 0)
    talk = talk_days.get(day, 0)
    if reverts > 0 or talk > 0:
        signal = "edit-warring?" if reverts > 2 and talk < 3 else \
                 "deliberation" if reverts > 0 and talk >= 3 else \
                 "discussion" if talk > 0 else ""
        print(f"{day}: {reverts} reverts, {talk} talk events → {signal}")

Reading the patterns

A revert_detected event is raised from the edit summary. The values below are illustrative; the shape is what refract export writes, with the enrichment fields left out:

{
  "eventType": "revert_detected",
  "fromRevisionId": 1280100100,
  "toRevisionId": 1280100200,
  "section": "",
  "before": "",
  "after": "Undid revision 1280100100: please propose this change on the talk page",
  "deterministicFacts": [
    { "fact": "revert_detected", "detail": "Undid revision 1280100100: please propose this change on the talk page" },
    { "fact": "policy_signal", "detail": "dimension=edit_warring signal=revert_detected" }
  ],
  "layer": "policy_coded",
  "timestamp": "2024-11-18T09:00:00Z",
  "schemaVersion": "0.5.0",
  "eventId": "3c5e7f9a1b2d4068"
}

section and before are empty, and after is the edit summary. The text that was undone is in the sentence_removed, sentence_modified and sentence_reintroduced events with the same fromRevisionId and toRevisionId.

A second revert_detected thirty minutes later on the following revision pair is the back-and-forth pattern. talk_page_correlated events on the same revisions show that talk edits were made nearby in time; a day with reverts and none suggests edit-warring without discussion.

Signal combination	Interpretation
Several reverts + no talk activity	Edit-warring
Several reverts + talk edits the same day	Dispute acknowledged and being discussed
Talk edits + no reverts	Discussion before acting
Many revisions within an hour	Coordinated editing or page under heavy revision

Troubleshooting

No talk page events? Not all pages have active talk pages. Try pages with known controversies (climate change, political topics, technology disputes).
Talk: prefix not working? Some wikis use different namespaces. Try Discussion: or check the wiki's namespace configuration. The Talk: prefix works for English Wikipedia and most Fandom wikis.