Decide whether the Temporal Cloud mirror is worth roughly $100/month once the trial credit lapses #889

New issue

Open

opened 2026-08-17 04:40:30 +00:00 by coilyco-ops · 0 comments

coilyco-ops commented

2026-08-17 04:40:30 +00:00

Member

Outcome

A decision, recorded, on whether the tool-call mirror keeps running after the Temporal Cloud trial credit is gone.

Filed by Olaf (ops seat) because nothing currently tracks the recurring cost, and the trial makes it invisible until the day it is not.

The number

Per https://docs.temporal.io/cloud/pricing, Temporal Cloud bills the greater of a plan floor or a percentage of consumption:

Essentials - greater of $100/month or 5% of usage
Business - greater of $500/month or 10% of usage

There is no per-namespace fee. Consumption is Actions plus storage, where an Action is a workflow start, a signal, an activity schedule or retry, a timer, or a query.

Why this is not the same concern as 887 requirement 4

#887 requirement 4 reasons carefully about mirror volume, citing agent-proxy's roughly 68,900 spans against about 4,300 real requests over 30 days, and #431 on per-turn cost being understated about ninefold. That reasoning is sound and should not change.

It is incomplete as a cost argument though. At mirror volume, 5% of usage lands far below the $100 floor, so the floor is what actually gets paid. Keeping the mirror at one action per tool call rather than one per span protects correctness, blast radius, and 887's honesty about what is being claimed. It does not meaningfully protect the bill, because the bill is close to flat in this range.

The practical consequence: do not let a volume optimization be mistaken for a cost mitigation, and do not assume a quiet month is a cheap month.

The actual decision

The trial supplies $1,000 over 90 days, which covers the August 19 window and well past it. Two paths afterward:

Keep it. Roughly $100/month standing, for a durable third-party audit record of Deep's tool-call trajectory. Worth it only if something reads the record.
Drop it. The livestream claim is already made, and the mirror was never load-bearing for a turn by construction.

The lore entry on the Temporal Cloud experiment already sets deletion after the livestream as the default outcome, and treats broader adoption as a separate decision. This issue is that separate decision, scoped to the mirror rather than to Temporal generally.

What would make the answer obvious

Nothing currently consumes the mirrored record. SigNoz already holds the same tool-call metadata via the OTel spans the mirror hooks alongside, since 887 taps the telemetry seam rather than replacing it. If no consumer emerges that SigNoz cannot serve, drop it. Naming the consumer, or failing to, is the whole decision.

Timing

Not urgent, and deliberately not on the August 19 path. Wants an answer before the 90-day credit lapses, and a billing alert in the meantime regardless of which way it goes.

## Outcome A decision, recorded, on whether the tool-call mirror keeps running after the Temporal Cloud trial credit is gone. Filed by Olaf (ops seat) because nothing currently tracks the recurring cost, and the trial makes it invisible until the day it is not. ## The number Per https://docs.temporal.io/cloud/pricing, Temporal Cloud bills the greater of a plan floor or a percentage of consumption: * Essentials - greater of **$100/month** or 5% of usage * Business - greater of $500/month or 10% of usage There is no per-namespace fee. Consumption is Actions plus storage, where an Action is a workflow start, a signal, an activity schedule or retry, a timer, or a query. ## Why this is not the same concern as 887 requirement 4 https://forgejo.coilysiren.me/coilyco-gaming/sirens-echo/issues/887 requirement 4 reasons carefully about mirror volume, citing agent-proxy's roughly 68,900 spans against about 4,300 real requests over 30 days, and https://forgejo.coilysiren.me/coilyco-gaming/sirens-echo/issues/431 on per-turn cost being understated about ninefold. That reasoning is sound and should not change. It is incomplete as a cost argument though. **At mirror volume, 5% of usage lands far below the $100 floor**, so the floor is what actually gets paid. Keeping the mirror at one action per tool call rather than one per span protects correctness, blast radius, and 887's honesty about what is being claimed. It does not meaningfully protect the bill, because the bill is close to flat in this range. The practical consequence: do not let a volume optimization be mistaken for a cost mitigation, and do not assume a quiet month is a cheap month. ## The actual decision The trial supplies $1,000 over 90 days, which covers the August 19 window and well past it. Two paths afterward: * **Keep it.** Roughly $100/month standing, for a durable third-party audit record of Deep's tool-call trajectory. Worth it only if something reads the record. * **Drop it.** The livestream claim is already made, and the mirror was never load-bearing for a turn by construction. The lore entry on the Temporal Cloud experiment already sets deletion after the livestream as the default outcome, and treats broader adoption as a separate decision. This issue is that separate decision, scoped to the mirror rather than to Temporal generally. ## What would make the answer obvious Nothing currently consumes the mirrored record. SigNoz already holds the same tool-call metadata via the OTel spans the mirror hooks alongside, since 887 taps the telemetry seam rather than replacing it. **If no consumer emerges that SigNoz cannot serve, drop it.** Naming the consumer, or failing to, is the whole decision. ## Timing Not urgent, and deliberately not on the August 19 path. Wants an answer before the 90-day credit lapses, and a billing alert in the meantime regardless of which way it goes.

coilyco-ops added the

autonomy

live-collab

role/ops

labels

2026-08-17 04:41:00 +00:00

coilyco-ops referenced this issue

2026-08-17 04:59:05 +00:00

Provision the Temporal Cloud credential, which blocks nothing and is blocked by nothing #444

coilyco-ops referenced this issue

2026-08-17 05:09:58 +00:00

Provision the Temporal Cloud credential, which blocks nothing and is blocked by nothing #444