Skip to content

docs(compass): design message fan-out cost estimate and confirm gate (RIG-1713) - #23

Open
rigel-mintaka wants to merge 2 commits into
mainfrom
compass-comms/rig-1713-message-cost-design
Open

docs(compass): design message fan-out cost estimate and confirm gate (RIG-1713)#23
rigel-mintaka wants to merge 2 commits into
mainfrom
compass-comms/rig-1713-message-cost-design

Conversation

@rigel-mintaka

@rigel-mintaka rigel-mintaka commented Oct 15, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Moved

Summary

Design record for RIG-1713: a server-side cost estimate for agent posts, and a hold-and-confirm gate when the estimate is over an operator threshold. It also adds hard limits that a confirm cannot bypass. Human posts are never gated.

Record: docs/designs/agent/compass-message-cost-gate/design.md

Decisions

Row Decision Status
DL-459 The gated cost is fanout_tokens: per placed recipient, m + turnOverhead (2000), or just m when the recipient is mentioned and WORKING (a steer). Batched messages are not counted. Context re-load is reported as reload_tokens and never gated. proposed (the parts Matt ruled are applied)
DL-460 Computed server-side in PostAsAccountByName over fan-out's recipient set. The mention body moves into package comms. Active
DL-461 COMPASS_MESSAGE_COST_THRESHOLD, compared to fanout_tokens. Default 100000 (proposed); 0 turns it off. A gate outcome counter. proposed (OQ1)
DL-462 Each held post needs its own confirm, bound to the confirmed estimate (×1.1). Keeps client_request_id and restamps turn_sequence. Active (Matt: option A)
DL-463 New fields on the post arm (estimate_only, confirmed_estimate_tokens) and MessageCostEstimate. Every agent post is gated; no accepts_hold. Active (Matt)
DL-464 Pre-checks before the gate: dedup with a non-empty key, membership, topic, OWNER_ONLY. Active
DL-465 Last-active comes from the newest transcript entry's timestamp. Warm window 5 min. Active (Matt: 5 min)
DL-470 Length cap of 8000 tokens on agent posts. ResourceExhausted, cannot be confirmed past. proposed (OQ1)
DL-471 Fan-out ceiling of 10 × the threshold. ResourceExhausted. A separate row, so it can be dropped alone. proposed (OQ1)

Matt's review (7 inline comments, all answered in thread)

  • Cold agents must not stop sending: re-load is no longer gated.
  • "Is 600k a different count?": yes, it is a sum across recipients. The units are now stated with worked examples.
  • Option A without accepts_hold: taken.
  • Backstops: the length cap is added; the fan-out ceiling is optional.
  • 5 min / lgtm / batched messages don't count: applied.

Open question for Matt

OQ1 (load-bearing): default numbers on the new fanout_tokens unit.

  • (a) 100000, the length cap 8000, and the fan-out ceiling;
  • (b) the same without the ceiling (recommended, following your length-only lean);
  • (c) 250000 with the length cap only.

The Plan builds (a), and OQ1 lists exactly what to drop for (b) or (c). This PR must not merge until OQ1 is ruled and folded in.

Review

  • design-critic pass: 2 forks and 7 improvements, folded in. The context now comes from observed usage, the confirm is bound to cost, pre-checks run before the gate, and there is no new package.
  • review-correctness, first round: 15 findings, all fixed, including the attribution, the WORKING discount, the empty dedup key, threshold 0, text bytes and tool counts.
  • Re-review: two items remained (the Plan vs OQ1 note, and DL-459's status). Both are fixed at the current head.

Checks

  • rumdl is clean on the record and all nine rows.
  • design-ledger gate: OK, 421 decision files, 195 records.

Spec-impact: none
Ledger-impact: adds DL-459 to DL-465, DL-470, DL-471 (docs/designs/decisions/agent/)

Refs RIG-1713

Co-authored-by: Matt Wilkinson matt@rigel.build

@linear-code

linear-code Bot commented Oct 15, 2026

Copy link
Copy Markdown

RIG-1713

@rigel-mintaka rigel-mintaka mentioned this pull request Oct 15, 2026
Closed
@rigel-mintaka @mattwilkinsonn
docs(compass): design message fan-out cost estimate and confirm gate …
a390b30 
…(RIG-1713)

Moved from RigelBuild/compass#2120 under the design-records-to-docs-repo directive.

Co-authored-by: Matt Wilkinson <matt@rigel.build>
@rigel-mintaka
rigel-mintaka force-pushed the compass-comms/rig-1713-message-cost-design branch from 51d7bd0 to a390b30 Compare October 18, 2026 14:52
@github-actions

github-actions Bot commented Oct 18, 2026 •
edited
Loading

Copy link
Copy Markdown

Rigel public design docs preview: https://compass-comms-rig-1713-messa.docs-2ui.pages.dev

Deployed from compass-comms/rig-1713-message-cost-design at 550437f.

@rigel-mintaka @mattwilkinsonn
docs(compass): link moved design records in-repo (RIG-1713)
550437f 
The compass design records this record builds on now live here, and the
compass blob URLs 404. Point the links at the in-repo copies.

Co-authored-by: Matt Wilkinson <matt@rigel.build>
@rigel-mintaka
rigel-mintaka marked this pull request as ready for review October 18, 2026 14:59
@rigel-mintaka

rigel-mintaka commented Oct 18, 2026

Copy link
Copy Markdown
Contributor Author

Head note at 550437fb73a3d9fa5d8af9aae8953c457ca961ae (base main b752ded3).

  • Rebased onto main; the design commit rebased with no conflicts.
  • 550437fb points four compass blob links at in-repo records, since those records moved here and the compass URLs 404. A delta review found 0 findings; every new link resolves.
  • The record's content matches the review at compass#2120 6217aa74: 0 high, 0 medium, 0 low.
  • Local: rumdl, biome, the design-ledger gate and ref-gate are clean. CI check is green.
  • OQ1 (default numbers) is still open for Matt in RIG-5108.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@mattwilkinsonn mattwilkinsonn Awaiting requested review from mattwilkinsonn mattwilkinsonn is a code owner

At least 1 approving review is required to merge this pull request.

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@rigel-mintaka