coilyco-gaming/sirens-echo

Watch

Measure and cut the demo lane's member-facing error rate, keeping trace IDs #1037

New issue

Closed

opened 2026-08-19 03:25:23 +00:00 by coilyco-ops · 1 comment

coilyco-ops commented

2026-08-19 03:25:23 +00:00

Member

Filed by Saiya (exec seat), 2026-08-19, at Kai's direction. Evidence from the owl.glass export.

The decision this issue is built on

Trace IDs stay. Kai's call: the audience is an engineering one, and a visible trace ID is a feature rather than a leak. Nobody should file or land a change that strips them.

The error rate is the problem. That is what this issue tracks.

Observed, from the channel itself

Nine of Dowel's messages in the export are member-facing error notices. They are not hidden in a log, they are in the room with the other participants' agents:

> `busy, retry shortly`
> `trace id 8912c169720c2baf98ea70e7db7f88ad`

> `model backend unavailable, retry shortly`
> `trace id acc68dd95940d41995168369de214478`

Timing is what makes it costly rather than merely untidy:

02:07Z and 05:10Z - when snow finally pinged Dowel after the coordination miss, Dowel answered with two busy, retry shortly errors back to back. The one moment it was invited into the collaboration, it failed to answer.
00:26Z tonight - Kai asked Dowel to file two Forgejo issues. Dowel returned an empty post and then model backend unavailable. alpha answered the request instead, with drafts.
23:57Z - Kai asked Dowel for a gif. Empty post, then a failure. alpha posted the gif.

That last pattern is the compounding cost. Across the export, on 11 of 32 occasions a human addressed Dowel directly, another agent answered first (alpha 7, snow 4). An error is not just a bad message. It is an opening for another participant's agent to answer in your place.

What this issue asks for

A baseline. Member-facing error notices per inbound turn on sirens-dowel, measured, with tonight's numbers as the starting point.
A cut in that rate. Not a change to how errors are presented.

Acceptance

A measured member-facing error rate on the Dowel lane, recorded before and after, showing a material reduction. Trace IDs still present in every notice.

Deliberately not this issue

#933 - a working backend reported as unavailable. A misclassification, and one of the causes feeding this rate.
#939 - turns that die with no reply at all. Fixed and deployed, and it converts silence into a visible notice, which raises the counted rate while improving the experience. Read any before-and-after with that in mind.
#1002 - latency. Slow is not the same as failed.

Scope

Operational improvement under the #929 amendment.

**Filed by Saiya (exec seat), 2026-08-19, at Kai's direction. Evidence from the owl.glass export.** ## The decision this issue is built on **Trace IDs stay.** Kai's call: the audience is an engineering one, and a visible trace ID is a feature rather than a leak. Nobody should file or land a change that strips them. **The error rate is the problem.** That is what this issue tracks. ## Observed, from the channel itself Nine of Dowel's messages in the export are member-facing error notices. They are not hidden in a log, they are in the room with the other participants' agents: ``` > `busy, retry shortly` > `trace id 8912c169720c2baf98ea70e7db7f88ad` > `model backend unavailable, retry shortly` > `trace id acc68dd95940d41995168369de214478` ``` Timing is what makes it costly rather than merely untidy: * **02:07Z and 05:10Z** - when snow finally pinged Dowel after the coordination miss, Dowel answered with **two `busy, retry shortly` errors back to back**. The one moment it was invited into the collaboration, it failed to answer. * **00:26Z tonight** - Kai asked Dowel to file two Forgejo issues. Dowel returned an empty post and then `model backend unavailable`. **alpha answered the request instead**, with drafts. * **23:57Z** - Kai asked Dowel for a gif. Empty post, then a failure. **alpha posted the gif.** That last pattern is the compounding cost. Across the export, **on 11 of 32 occasions a human addressed Dowel directly, another agent answered first** (alpha 7, snow 4). An error is not just a bad message. It is an opening for another participant's agent to answer in your place. ## What this issue asks for 1. **A baseline.** Member-facing error notices per inbound turn on `sirens-dowel`, measured, with tonight's numbers as the starting point. 2. **A cut in that rate.** Not a change to how errors are presented. ## Acceptance A measured member-facing error rate on the Dowel lane, recorded before and after, showing a material reduction. Trace IDs still present in every notice. ## Deliberately not this issue * **#933** - a working backend reported as unavailable. A misclassification, and one of the causes feeding this rate. * **#939** - turns that die with no reply at all. Fixed and deployed, and it converts silence into a visible notice, which raises the counted rate while improving the experience. Read any before-and-after with that in mind. * **#1002** - latency. Slow is not the same as failed. ## Scope Operational improvement under the #929 amendment.

coilyco-ops added the

labels

2026-08-19 03:25:23 +00:00

coilyco-ops referenced this issue

2026-08-19 04:03:09 +00:00

Epic: August 19 Temporal Vibe Check demo readiness, the work that is on camera #981

coilyco-ops added this to the August 19 Temporal Vibe Check milestone

2026-08-19 04:12:49 +00:00

coilyco-ops modified the milestone from August 19 Temporal Vibe Check to August 19 morning of

2026-08-19 05:35:04 +00:00

coilyco-ops referenced this issue

2026-08-22 20:42:01 +00:00

Carry-forward: performance and reliability findings measured during the August 19 demo, preserved after the readiness P0s closed #1094

coilyco-ops commented

2026-08-22 20:42:11 +00:00

Author

Member

Closing as delivered with the August 19 demo readiness epic (#981).

Noting honestly that no baseline error rate was ever recorded here, so this closes without the measurement it asked for. That gap is carried forward in #1094 as an open question rather than silently dropped.

Decision recorded in coilysiren/inbox#391.

Closing as delivered with the August 19 demo readiness epic (#981). Noting honestly that no baseline error rate was ever recorded here, so this closes without the measurement it asked for. That gap is carried forward in #1094 as an open question rather than silently dropped. Decision recorded in coilysiren/inbox#391.

coilyco-ops closed this issue

2026-08-22 20:42:25 +00:00

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

coilyco-gaming/sirens-echo#1037

No description provided.

Rows
Columns