audiofreak7/stackoverflow-decline-panels-2026q3

Stack Overflow and Stack Exchange activity panels, 2008–2026

Weekly and monthly counts of questions, answers, users and response times for 27 Stack Exchange sites and the 363 Stack Overflow tags analysed in What happened to Stack Overflow? (John Pospisil, 2026). Built from the official Stack Exchange data dump, 2026-Q3 release (cut-off 2026-09-30). Derived counts only: no post text, titles, user names or user IDs.

• Code, method and article: https://github.com/johnpospisil/stackoverflow-decline
• Not affiliated with or endorsed by Stack Overflow or Stack Exchange.

Files

File	Rows	Columns	Periods
data/site_week.parquet / .csv	22,718	21	2008-07-28 to 2026-09-28
data/site_month.parquet / .csv	5,231	22	2008-07-01 to 2026-09-01
data/tag_week.parquet / .csv	269,365	19	2008-07-28 to 2026-09-28
data/tag_month.parquet / .csv	62,012	19	2008-07-01 to 2026-09-01
data/tag_estimates.parquet / .csv	297	10	–

columns.csv has the column dictionary below; SHA256SUMS has a checksum for every file.

• Sites (27): stackoverflow.com; Stack Overflow in Russian, Spanish, Portuguese and Japanese; six other tech sites (Super User, Server Fault, Ask Ubuntu, Unix & Linux, Software Engineering, Database Administrators); fifteen non-programming sites (Mathematics, Physics, Chemistry, Biology, English, Philosophy, History, Cooking, DIY, Travel, Bicycles, Gaming, Science Fiction, Movies, Role-playing Games); Meta Stack Overflow. site_group says which.
• Tags: the 300 largest Stack Overflow tags of 2019–21, and 69 tags first used after 2021-09-30 with at least 200 questions. Tags are as they were when each question was asked (the first tag event in the post history), not today's tags.
• tag_estimates: each top tag's post-ChatGPT gap against its own trend from the project's hierarchical model (297 tags; three retired tags excluded), plus the plain 2021→2025 change.

Column dictionary

Column	Type	Meaning
period	date	Start of the period: Monday of the week (UTC) or first day of the month
site	text	Stack Exchange site, e.g. stackoverflow.com (site panels)
site_group	text	treated (Stack Overflow), so_language, tech_exposed, donor (non-programming comparison sites) or context (Meta Stack Overflow) (site panels)
tag	text	Stack Overflow tag, as it was when the question was asked (tag panels)
tag_group	text	top300 (largest tags of 2019-21) or new_post2021 (first used after 2021-09-30, >= 200 questions) (tag panels)
tag_kind	text	top300: hand-assigned family (language, framework, data_ml, ...); new_post2021: new / version / feature (tag panels)
questions	count	New questions created in the period (questions deleted before the dump are not included)
active_askers	count	Distinct users who asked at least one question in the period
answered_ever	count	Questions with at least one answer by the dump date
eligible_24h	count	Questions whose 24-hour window ends before the 2026-09-30 cut-off (denominator for answered_24h)
answered_24h	count	Questions answered within 24 hours (among eligible_24h)
eligible_7d	count	Questions whose 7-day window ends before the cut-off
answered_7d	count	Questions answered within 7 days (among eligible_7d)
eligible_30d	count	Questions whose 30-day window ends before the cut-off
accepted_30d	count	Questions with an answer accepted within 30 days (among eligible_30d)
closed	count	Questions ever closed (by the dump date)
closed_duplicate	count	Questions closed as duplicates (by the dump date)
with_code	count	Questions with at least one code block
median_hours_to_first_answer	hours	Median hours to the first answer, among questions answered (by the dump date); recent periods are biased low
answers	count	New answers created in the period (site panels)
active_answerers	count	Distinct users who posted at least one answer in the period (site panels)
new_posters	count	Users whose first question or answer on the site is in the period (site panels)
questions_incl_deleted	count	Questions including later-deleted ones, from the Stack Exchange Data Explorer (SEDE), 2019-01 onward; empty where not available (site_month only)
partial_period	flag	The period is not fully inside the dump (first partial week/month or after the cut-off)
final_8_weeks	flag	The period starts in the last 8 weeks before the cut-off: answers and closures may still be arriving
tag		Stack Overflow tag
family		Hand-assigned tag family
size_2019_2021		Questions in 2019-2021 (tags at time of asking)
age_years_at_break		Years from the tag's first question to 2022-11-30
pct		Average monthly gap vs the tag's own extended 2017-22 trend, Dec 2022 - Sep 2026, % (hierarchical negative binomial, posterior mean)
pct_low		95% credible interval, lower (%)
pct_high		95% credible interval, upper (%)
questions_2021		Questions in 2021
questions_2025		Questions in 2025
plain_change_2021_2025_pct		Plain change in questions, 2021 to 2025, % (no trend assumed)

How it was made

The XML dump was streamed into Parquet, one row per post, and aggregated with DuckDB. Weeks run Monday to Sunday, UTC. A question counts in the period it was created. Fixed-window measures (answered within 24 hours, 7 days; accepted within 30 days) count only questions whose whole window ends before the cut-off, so they are comparable over time. The full pipeline (make all) rebuilds these files from the raw dump.

Known limitations

• Deleted posts are missing. The dump omits posts deleted before the dump date. On Stack Overflow the deleted share of questions rose from about 34% (2017) to 53% (2026), so recent counts understate how many questions were asked. questions_incl_deleted (site_month, from SEDE, 2019 onward) gives the totals including deleted posts.
• Counts as of the dump date. Answers, closures and acceptances keep arriving after a question is asked; recent periods are incomplete (final_8_weeks), and median_hours_to_first_answer is biased low for recent periods. Prefer the fixed-window columns.
• Closed questions are under-represented, because closed questions are deleted more often.
• Tag changes: Stack Overflow retires, merges and renames tags (e.g. excel-vba 2020-08, flutter-layout 2022-12, dockerfile 2025-06); counts can drop to zero for that reason.
• Platform changes affect series: e.g. the 2023 moderator strike, Staging Ground for new askers (2024-06), opinion-based questions (2025-10), a read-only period (2026-08). See data/reference/events.csv in the repository.
• The tag estimates depend on a trend assumption: each tag's own 2017–22 trend is extended four years, which enlarges the gap for tags that were growing fast. Compare with plain_change_2021_2025_pct.
• Dump integrity: file hashes match the community-maintained checksum list; no watermark rows were found.

Licence and attribution

Stack Exchange content is licensed CC BY-SA 4.0 by its contributors (older posts CC BY-SA 2.5 or 3.0). This derived dataset is released under CC BY-SA 4.0. Please attribute:

Derived from the Stack Exchange data dump (2026-Q3), contributed by Stack Exchange users and licensed CC BY-SA 4.0. Panels by John Pospisil, https://github.com/johnpospisil/stackoverflow-decline

The data dump is provided under Stack Exchange's access terms, which exclude training large language models. This dataset contains aggregate counts only, but please don't use it to train LLMs either.

Citation

@misc{pospisil2026stackoverflow,
  author = {Pospisil, John},
  title  = {Stack Overflow and Stack Exchange activity panels, 2008--2026},
  year   = {2026},
  note   = {Derived from the Stack Exchange data dump, 2026-Q3. CC BY-SA 4.0},
  url    = {https://github.com/johnpospisil/stackoverflow-decline}
}