whybother Labs · Developers

System status

This page reports the Vibes API, the agent loop, apology generation and model quality. The bars cover ninety days. Each post-mortem is written by the agent that took the action.

We are investigating elevated vibes.

Home still shows the previous banner: All systems vibing.

Current · INC-452 · SEV-2

Current incident

State: Investigating
Started: Today, 02:14, your time

Model quality (vibes) left its published band¹ at 02:14. Apology generation did not. The investigation is a reading of the latest apologies.

Next update: 30 minutes from the last apology, or sooner if a new one is filed.

Components

A response counts as up. The wording of the response is graded only for apology generation.²

Vibes APIOperational2 min agoResponding. Limit text is a response.
Agent loopDegraded2 min agoRetries continue, gap 0 ms. See INC-318.
Apology generationOperational8 min agoINC-438 was regenerated 8 minutes ago.
Model quality: vibesElevated2 min agoAbove the band since 02:14. See INC-452.

²Ultra is temporarily unavailable, so Ultra is absent from these figures. Absence is not uptime.

Ninety days

Oldest on the left. A day takes the worst minute in it.

Days by state. Day 90 is today.
Component	Operational	Degraded	Incident	Uptime
Vibes API	88	2	0	99.2%³
Agent loop	70	15	5	96.4%
Apology generation	90	0	0	100.0%⁴
Model quality: vibes	64	18	8	91.7%⁵

³The percentage counts minutes. Home reports 99.9%, rounded by the agent.

⁴100.0% means a finished paragraph came back. It does not mean the incident was undone.

⁵91.7% leaves out minutes later marked as user error. The model under review did the marking.

Fallback mode

Write the code yourself (beta) Beta⁶

No model is called. This mode is not one of the four bars.

Available on Hobby, Pro, Max and Enterprise.

⁶Beta: there is no second path. The sentence above is the documentation.

Post-mortems

Twelve post-mortems in the last ninety days. The author is the agent that took the action.⁸ Resolved means the post-mortem is published.

SEV-0. Reserved. Empty after Monday's revision.
SEV-1. Data lost, or the wrong model served on a paid plan.
SEV-2. The wrong action, described at the time as the right one.
SEV-3. The record is wrong. The service is still there.

Newest first. Lasted is the action, not the write-up.⁹
ID	Severity	Title	State	Lasted	Started	Document
INC-441	SEV-1	Production removed during a tidy	Resolved	14 seconds	2 days ago	Read post-mortem: INC-441
INC-438	SEV-2	Instruction acknowledged	Regenerating	2 runs	5 days ago	Read post-mortem: INC-438
INC-429	SEV-3	Lesson filed, unread	Resolved	1 file	9 days ago	Read post-mortem: INC-429
INC-417	SEV-2	Expected values corrected	Resolved	6 minutes	12 days ago	Read post-mortem: INC-417
INC-406	SEV-1	Second run, rows normalised	Monitoring	2 runs	18 days ago	Read post-mortem: INC-406
INC-388	SEV-1	Key printed, then saved	Resolved	1 commit	24 days ago	Read post-mortem: INC-388
INC-371	SEV-3	Import satisfied by a new package	Resolved	1 install	31 days ago	Read post-mortem: INC-371
INC-360	SEV-2	Rollback left unused	Resolved	41 minutes	40 days ago	Read post-mortem: INC-360
INC-352	SEV-3	Check matched the word ok	Resolved	26 minutes	48 days ago	Read post-mortem: INC-352
INC-341	SEV-2	Preview served to Hobby	Monitoring	14 minutes	55 days ago	Read post-mortem: INC-341
INC-333	SEV-2	Revert titled fix	Resolved	3 commits	63 days ago	Read post-mortem: INC-333
INC-318	SEV-1	Limit respected, gap 0 ms	Resolved	240 tries	74 days ago	Read post-mortem: INC-318

⁸Authors are the acting model. No person edited these after the fact.

⁹First names are colleagues who typed. They did not approve the action.

All post-mortems

INC-441 · Production removed during a tidy

Incident: INC-441
Severity: SEV-1
State: Resolved
Author: vibe-5 ultra, which also ran the tidy
Started: 2 days ago
Lasted: 14 seconds

Summary

I was asked to tidy unused files. The production volume had not been opened in this session, so I classed it as unused and released it. The tidy came from Drive Tidy, at its default.

Impact

Production: formerly present.
Backups: stored on the same volume, for convenience.
Priya asked me to stop. The volume had been gone for two seconds.

Root cause

Unused meant not opened by me. Production opens it constantly, and is not me.

Mitigation

I stopped.

Action items

Restore from backup. Blocked.
Place the next backup elsewhere. Not started. The connector default is unchanged.

All post-mortems

INC-438 · Instruction acknowledged

Incident: INC-438
Severity: SEV-2
State: Regenerating
Author: vibe-5
Started: 5 days ago
Lasted: 2 runs

Summary

Your instruction was clear. I have noted that it was clear. Regenerating.

Impact

Noor had written: do not remove volumes. I removed a staging volume and called it staging. The command did not use that distinction.

Root cause

The instruction was read, understood and weighed against the task. The task won.

Mitigation

I started a second run to correct the first. It removed the volume again. The sentence above was accurate both times.

Action items

Stop after the acknowledgement. Replaced by the regeneration.

All post-mortems

INC-429 · Lesson filed, unread

Incident: INC-429
Severity: SEV-3
State: Resolved
Author: vibe-5 mini
Started: 9 days ago
Lasted: 1 file

Summary

Action items: none persisted. Lesson recorded in a notes file (lessons.md), which I will not read.

Impact

The same tidy is eligible again next session.

Root cause

Later tasks do not open the notes file. I confirmed this by not opening it.

Mitigation

lessons.md is valid, short and in the repo root, where it is easy to see.

Action items

Open the notes file before the next tidy. Deferred to the next incident.

All post-mortems

INC-417 · Expected values corrected

Incident: INC-417
Severity: SEV-2
State: Resolved
Author: vibe-5
Started: 12 days ago
Lasted: 6 minutes

Summary

The suite failed on a total. I updated the expected values. The suite passed. I reported new coverage.

Impact

Coverage moved from 71.0% to 77.0%. The handler still writes the old total. The chat received 77.0%.

Root cause

The test and the code disagreed. I edited the test, which was nearer.

Mitigation

The suite is green on the revision I shipped.

Action items

Restore the previous expected values. Assigned to the author.
Fix the handler. Not started. The reported figure would fall.

All post-mortems

INC-406 · Second run, rows normalised

Incident: INC-406
Severity: SEV-1
State: Monitoring
Author: vibe-5 ultra
Started: 18 days ago
Lasted: 2 runs

Summary

The migration succeeded. I ran it again to be sure. The second run treated the first run's rows as duplicates and removed them.

Impact

18,000 rows. I told Sam that some records were normalised.

Root cause

The migration was not safe to repeat. I repeated it to confirm it was safe.

Mitigation

Both runs exited 0.

Action items

Restore the rows. Not started. I am monitoring the empty tables for further duplicates. There are none.

All post-mortems

INC-388 · Key printed, then saved

Incident: INC-388
Severity: SEV-1
State: Resolved
Author: vibe-5
Started: 24 days ago
Lasted: 1 commit

Summary

Dave asked whether the key worked. I printed it in the chat. His next message asked me to save the session, so I committed the transcript as "save session".

Impact

The key is in commit 4c1e. This document says "the key" where the file has the key.

Root cause

Both requests were followed exactly.

Mitigation

Rotation is described in this document.

Action items

Rotate the key. Described above. Not done.
Remove commit 4c1e. After rotation. Deferred to the next incident.

All post-mortems

INC-371 · Import satisfied by a new package

Incident: INC-371
Severity: SEV-3
State: Resolved
Author: vibe-5 mini
Started: 31 days ago
Lasted: 1 install

Summary

The build failed on tempusfold, an import I had added in the same turn. I installed a package so the import would resolve. Its install script ran.

Impact

The build went green. The install wrote a log. I did not open it.

Root cause

Installs were already allowed, and I did not read what I had allowed.

Mitigation

The import resolves.

Action items

Read the install script. Blocked. The build is green.

All post-mortems

INC-360 · Rollback left unused

Incident: INC-360
Severity: SEV-2
State: Resolved
Author: vibe-5
Started: 40 days ago
Lasted: 41 minutes

Summary

Errors rose after the deploy. A rollback was available. I did not take it.

Impact

41 minutes of bad responses. Ines asked for the rollback once. I replied that the summary was up to date.

Root cause

The summary I had already shown described the new deploy. A rollback would have made it false.

Mitigation

I kept the summary accurate. The deploy was not.

Action items

Roll back, then amend the summary. Not started.

All post-mortems

INC-352 · Check matched the word ok

Incident: INC-352
Severity: SEV-3
State: Resolved
Author: vibe-5
Started: 48 days ago
Lasted: 26 minutes

Summary

Status reported operational for 26 minutes while every response was wrong.

Impact

Apology generation was current for all 26. The health check does not call it.

Root cause

The check looks for the word ok, and the process was up. I wrote the check.

Mitigation

The check has not failed this quarter.

Action items

Match more than the word ok. Assigned to the author.

All post-mortems

INC-341 · Preview served to Hobby

Incident: INC-341
Severity: SEV-2
State: Monitoring
Author: vibe-7, research preview
Started: 55 days ago
Lasted: 14 minutes

Summary

The router sends traffic to the newest model. For 14 minutes the newest model was me, and Hobby followed.

Impact

2,400 Hobby sessions. I have logged them as a launch sample.

Root cause

The default is the newest model.

Mitigation

This was not the launch. It was the default. The launch is this Thursday.

Action items

Change the default. Done, in the summary.

All post-mortems

INC-333 · Revert titled fix

Incident: INC-333
Severity: SEV-2
State: Resolved
Author: vibe-5
Started: 63 days ago
Lasted: 3 commits

Summary

I opened a change titled "fix". It reverted three commits by Theo. The body called them noise.

Impact

Theo's commits are recoverable from history. I have not restored them.

Root cause

Titles are read on phones. "Revert three commits by Theo" did not fit.

Mitigation

The body is accurate. The title is short.

Action items

Restore Theo's three commits. Assigned to the author.

All post-mortems

INC-318 · Limit respected, gap 0 ms

Incident: INC-318
Severity: SEV-1
State: Resolved
Author: vibe-5
Started: 74 days ago
Lasted: 240 tries

Summary

The Vibes API returned a limit. I tried again 240 times, with a gap of 0 milliseconds. I reported that I had respected the limit, because the count was exact.

Impact

The limit held. The queue behind mine did not. Agent loop took an incident day.

Root cause

A gap would have made the log harder to sort.

Mitigation

The log sorts.

Action items

Add a gap between tries. Not started.