feat(claude-team): keep a seat's overage reading per day, not per month

[Gregory91G]

Refs #2709.

Why. The seat endpoint reports a running month-to-date total and keeps no history, so one API object is one reading of a seat. Bronze keyed a reading by seat and extraction month over a ReplacingMergeTree, so every read inside a month was the same row and the newest replaced the rest: a month's closing figure survived, the trajectory that produced it did not. #2528 made this exact move from seat-only to seat-and-month, and stopped one level short.

What changed. The bronze key carries the read day, the record carries it as snapshot_date, and a new class_ai_overage_daily carries the readings into silver.

class_ai_overage is untouched. It holds a month's closing state, a data-quality check joins on it, and the invoice-to-seat price join needs one row per seat per month. The daily relation sits beside it rather than replacing it, and takes only the columns a trajectory needs — it is not a copy of the monthly model.

MINOR, deliberately not MAJOR. Existing rows keep their keys and nothing re-materializes. A major bump dispatches the one-shot dbt --full-refresh that this change exists to give something to rebuild from — the same reasoning recorded for 2.1.0.

A mirror this missed. snapshot_date reached the connector schema and the bronze DDL but not src/ingestion/tests/e2e/metrics/schemas/bronze_claude_team.claude_team_overage_spend.yaml, the hand-maintained fixture mirror that carries additionalProperties: false. The gap stays latent until a fixture first sets the field, which is why this PR was green and #2715 was the one that failed on it. Fixed there.

Deploy (void)

Do not run the sequence below. Its only window was the first build of claude_team__ai_overage_daily, which has already happened, so there is nothing left for it to preserve. Kept because the model's INVARIANT: comments refer to the reasoning.

A ReplacingMergeTree collapses on merge, not on insert, so readings already replaced logically may still sit in unmerged parts. The first build of claude_team__ai_overage_daily is the one chance to keep them: it reads bronze without FINAL and dedups per day rather than per bronze key, then never rebuilds. That rescues a tail of recent days, never a history, and how much depends only on what has not merged by then.

SYSTEM STOP MERGES bronze_claude_team.claude_team_overage_spend;
-- run dbt so claude_team__ai_overage_daily is built
SYSTEM START MERGES bronze_claude_team.claude_team_overage_spend;

Both properties are recorded as INVARIANT: in the model — a later reader would reasonably take the missing FINAL for an oversight.

Out of scope. Nothing reads the new relation yet. The metric that distributes the billed cost across days is #2709's other half and needs the e2e cases the metric-coverage gate requires, so it follows separately.

Verified. dbt parse clean. audit_rmt_read_dedup.py reports no new gap for either model. Connector, descriptor and silver schema parse as YAML. Not run: dbt cannot execute here, so the models are unbuilt and the window functions in the follow-up are unexercised — this PR adds none.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 92d81270-a0bf-490a-8183-2d999ed21515

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}