PR #185

You're right, two died. There's no limit in the workflow.

Added this logic to #195, as well.

As far as I can see there's not actually an enforced cap as described above in the final merged code? Two PRs died today even though there are < 100 PRs currently.

Bug: Mergeability detection is broken

The workflow uses pulls.list to fetch all open PRs, then checks pr.mergeable_state and pr.mergeable to detect conflicts. However, the pulls.list endpoint does not return these fields — they are only populated by the individual pulls.get endpoint, which computes mergeability on demand.

As a result, pr.mergeable_state and pr.mergeable are always undefined, and isUnmergeable is always false. The unmergeable penalty (5% base + 5x multiplier) has never applied to any PR.

Example: PR #84 currently has merge conflicts (mergeable: CONFLICTING, mergeable_state: dirty), but the workflow evaluated it with a 13.39% death chance instead of the ~50%+ it should have had.

Fix: Each PR needs an individual pulls.get call to retrieve mergeability info:

const { data: fullPr } = await github.rest.pulls.get({
  owner, repo, pull_number: pr.number
});
const isUnmergeable =
    fullPr.mergeable_state === 'dirty' ||
    fullPr.mergeable === false;

Note that mergeable can be null while GitHub is still computing it, so a brief delay/retry may also be needed on first fetch.

Old Age Analytics — Death Cap Update

Based on community feedback from @bigintersmind and @matthewmayer, I've added a death cap to prevent mass extinction events.

Death Cap Formula

cap = max(1, floor(open_prs / 100))

• 63 open PRs → 1 death max per day (was potentially 17+ uncapped)
• 101-200 PRs → 2 deaths max
• etc.

When more PRs "die" than the cap allows, survivors are randomly selected from the condemned.

Current PR Risk Table (Monte Carlo, 10k simulations)

Raw% = probability before cap, Eff% = effective probability after cap, [!] = unmergeable

    PR   Age   Int   Raw%   Eff%       Risk  Title
------  ----  ----  -----  -----  ---------  --------
#  84   35d     5  90.2%  4.88%     DANGER  Add auto-refresh for PR data [!]
#  79   35d     8  84.8%  4.65%     DANGER  Show PR author stats [!]
#  82   35d     7  86.4%  4.45%     DANGER  Add emoji scoring and counts [!]
#  61   36d    21  76.5%  4.32%     DANGER  feat: chaotic website layout [!]
#  46   39d    27  84.2%  4.27%     DANGER  Make sure the title fizzbuzzes [!]
#   5   40d    61  76.8%  4.15%     DANGER  added informative video [!]
#  93   32d     8  74.1%  4.14%     DANGER  Add one big & one small PR rule [!]
#  12   40d   105  69.2%  4.01%     DANGER  Add 17 languages + snow [!]
#  92   32d    13  68.4%  4.00%     DANGER  Instant Override Merge [!]
#  15   40d    84  72.3%  3.86%     DANGER  Visibility boost to prev merge [!]
# 102   28d     3  70.7%  3.72%     DANGER  Bad Apple but it is openchaos [!]
#  87   34d    33  63.7%  3.41%     DANGER  Chore: adds Konami Code [!]
#  68   36d    53  64.4%  3.40%     DANGER  Implement asteroids [!]
# 123   26d     8  54.6%  2.90%     DANGER  Add Retro BSOD [!]
# 114   27d    20  47.5%  2.87%     DANGER  Adding DVD-Player [!]
# 125   26d    11  51.2%  2.62%     DANGER  add mad Chaos Points [!]
# 118   26d    13  49.4%  2.50%     DANGER  Add countdown to page title [!]
# 106   27d    20  47.5%  2.49%     DANGER  Replace eslint with Biome.js [!]
# 131   24d    10  46.4%  2.36%     DANGER  Add dynamic retro marquee [!]
# 104   27d    37  40.4%  2.34%     DANGER  Fix link to wrong repository [!]
# 141   20d     9  36.8%  2.17%     DANGER  IE6BrowserChrome.tsx [!]
# 140   21d     9  39.3%  2.08%     DANGER  add rick roll [!]
# 146   19d     9  34.3%  2.04%     DANGER  XP error dialogs + Start Menu [!]
# 117   26d    53  33.4%  1.77%    AT RISK  Random emoji swap midnight [!]
# 166   15d     6  28.9%  1.76%    AT RISK  reduce colors to 256 [!]
# 149   18d    11  30.0%  1.60%    AT RISK  allow owner's PRs in HoC [!]
# 147   18d    13  28.4%  1.58%    AT RISK  69 > 67 [!]
# 157   16d     5  32.6%  1.55%    AT RISK  Create voting SSOT module [!]
# 152   16d    12  24.8%  1.46%    AT RISK  Pageview counter tracking [!]
# 180   12d     5  24.5%  1.44%    AT RISK  Chaos Button for Destruction [!]
# 164   15d     8  26.4%  1.41%    AT RISK  randomize visitor counter [!]
# 156   16d     7  29.7%  1.41%    AT RISK  RULES.md deprecated clarify [!]
# 165   15d     6  28.9%  1.38%    AT RISK  Make minimize button work [!]
# 170   14d    13  20.0%  1.11%    AT RISK  Replace "nothing" w/ secret [!]
# 158   16d    25  17.7%  0.84%    AT RISK  weekly workflow close oldest [!]
# 188    6d     5  15.5%  0.79%    AT RISK  rate limits bag of dickbutts [!]
# 167   15d    38  11.7%  0.74%    AT RISK  REWRITE IT IN RUST [!]
#  45   40d    30  12.3%  0.67%    AT RISK  F, At least I tried
#  70   36d    26   9.7%  0.54%    AT RISK  Banner has something to say
#  72   36d    33   9.1%  0.45%       LOW   devcontainer for malicious
#  94   32d    20   7.6%  0.42%       LOW   Revert "Rewrite it in rust"
#  95   32d    15   8.3%  0.40%       LOW   log Guy Fieri ASCII art
# 111   27d     8   6.5%  0.31%       LOW   set repo name to rustychaos
# 137   22d     6   4.2%  0.29%       LOW   Bringing back democracy
# 133   23d    14   3.0%  0.17%       LOW   fishies scatter from cursor
# 155   16d     5   1.5%  0.11%       LOW   Branch protection test
# 115   26d    18   4.1%  0.11%       LOW   Upgrade Midi Player GTA radio
# 160   16d     9   0.5%  0.03%       LOW   Add robots.txt
# 181   12d     3   0.7%  0.01%       LOW   daily automated PR review bot
# 161   16d     9   0.5%  0.01%       LOW   Fix auto-merge unmergeable
# 143   20d    22   0.5%  0.01%       LOW   random themes chaotic dreams
# 168   15d     9   0.1%  0.00%       LOW   Mergeability detection
# 193    0d     2   0.0%  0.00%      SAFE   chaos terminal
# 192    0d     1   0.0%  0.00%      SAFE   Make it go whrrr
# 191    0d     2   0.0%  0.00%      SAFE   bonzi scheme meme
# 190    2d     4   0.0%  0.00%      SAFE   konami meow
# 187    7d    10   0.0%  0.00%      SAFE   Cat and Scat
# 185   11d    28   0.0%  0.00%      SAFE   old-age death (this PR)
# 184   12d    15   0.0%  0.00%      SAFE   Vote integrity inspection
# 183   12d     8   0.0%  0.00%      SAFE   Decimate: Alea Iacta Est
# 182   12d     9   0.0%  0.00%      SAFE   RULES.md is optional
# 163   16d    22   0.0%  0.00%      SAFE   Web 2.0 ASCII
# 159   16d    18   0.0%  0.00%      SAFE   BOUNTY HUNT Random Upvoter

Key Takeaways

• Without cap: ~37 PRs had >5% raw death chance — algorithm would massacre the repo
• With cap: Max effective death probability is ~4.9% (highest risk PR gets ~5% chance of being the 1 unlucky pick)
• Unmergeable PRs [!] dominate the danger zone — merge conflicts are still the biggest risk factor
• Young PRs with engagement are safe — the algorithm rewards community love

I agree with your suggestions to reduce the dirty multiplier and cap it at 1/day, @matthewmayer.

Regardless of multiplier settings, capping at 1 death/day turns it from mass extinction into a sustainable daily drama. In that vein, maybe it's worth immortalizing the culled PRs on the front page somehow?

Also, a grace period (ignoring dirty status for 24–48h after a daily merge) would help with the timezone fairness issue, but would add implementation complexity.

I really like the idea, but think we should tune it to be more fun than discouraging.

This seems way too aggressive at the moment then. It's pretty normal to need frequent rebasing to resolve conflicts after the daily merge. And agree it's going to be biased towards people whose time zone means they are awake after the daily merge. I would consider some combination of

• make the dirty multiplier much lower (1.5?)
• run less frequently (eg weekly cull)
• cap daily deaths (eg 1 per day, pick randomly from all marked for possible death)

Simulation Results: What the Old Age Algorithm Would Do Today

I ran a Monte Carlo simulation (10,000 runs) against the current 64 open PRs to see how this workflow would play out.

Day 1 Impact

The algorithm would kill ~17 PRs on its first run — over a quarter of all open PRs. The unmergeable + old PRs get hit hardest:

Even #12 with 105 interactions still faces 62% death — the unmergeable 5x multiplier overwhelms the fitness protection.

7-Day Projection

After 7 days: ~21 survivors out of 64 (min 13, max 29 across simulations). Deaths front-load heavily as the high-risk PRs die first, then taper off.

Key Observations

• 42 of 64 PRs have <50% chance of surviving a full week
• The dirty mergeable state is the dominant factor. The 5x unmergeable multiplier means any PR with merge conflicts and more than ~10 days of age is in serious danger, regardless of engagement
• dirty is the norm, not the exception. Since a PR merges daily, every other open PR touching the same files becomes conflicted. Authors may have less than 24 hours between the daily merge (09:00 UTC) and the old-age sweep (06:00 UTC) to rebase — and that window gets worse since the sweep runs before most people's working hours. In practice, the majority of PRs will carry the dirty state at any given time (39 of 64 currently do)
• Engagement barely helps unmergeable PRs. PR #12 has 105 interactions but still 62% daily death because the multiplier is applied after fitness reduction
• Non-dirty PRs are comparatively safe. PR #45 (37 days old, behind) only has 10% daily risk. Same-aged dirty PRs face 60-80%
• Young mergeable PRs are essentially immune. 6 PRs have >99% weekly survival — all are <13 days old and not dirty

Capped vs Uncapped Culling

There is no per-day cap on deaths. In the worst simulated week, 51 of 64 PRs died. On day 1 alone, up to ~30 deaths are possible in an unlucky run. If the intent is gradual natural selection rather than a mass extinction event, a daily cap might be worth considering.

The Reaper heard your words with glee, and rhymed its titles, all for thee. From Rust to robots, chaos to fate — every PR now rhymes... checkmate. 💀

The Reaper comes with a fatal urge, Memento Mergi — remember to merge! But check #122 for one last time: To get this in, your title must rhyme!

I don't mind them opening a new PR with the same stuff :) I think that's great, but they would loose al votes ect. and I think it would clean it up in a fun and chaotic way!

i mean, you can reopen a closed PR (provided it hasnt been merged) and keep the existing votes

Fixed 💀

But there is a solution, we just lock it :-D doing commit now.

Oh... that makes sense.

I don't mind them opening a new PR with the same stuff :) I think that's great, but they would loose al votes ect. and I think it would clean it up in a fun and chaotic way!

i mean, you can reopen a closed PR (provided it hasnt been merged) and keep the existing votes

I don't mind them opening a new PR with the same stuff :) I think that's great, but they would loose al votes ect. and I think it would clean it up in a fun and chaotic way!

The problem with these various attempts to auto-close old PRs is that nothing prevents the author from reopening them again. If you really want to prevent old PRs from being merged, you need a rule that excludes them.

🤖 OpenChaos Bot

Summary: This PR introduces a GitHub Action that automatically closes old and inactive pull requests based on an age and interaction-based "natural selection" algorithm. It calculates a "death probability" for each PR, and if the PR "dies", it's closed with a detailed comment explaining its demise.

Files changed: 1 (.github/workflows/old-age.yml)

Vibe: This PR is playing God with pull requests, deciding their fate with the cold, uncaring logic of an algorithm and a dash of morbid humor.

openchaos-bot