Pimp My IDE / garage dispatchBack to the garage
October 7, 2026 / Copilot / measurement

Your chart lost the source.

A client migration dropped the IDE identity that Copilot agent metrics needed. The activity did not disappear. The report lost its attribution, and the missing history cannot be rebuilt.

Never turn a telemetry gap into a developer-performance story.

The activity stayed. The IDE label disappeared.

GitHub says several IDEs moved Copilot agent sessions to the Copilot SDK. Those sessions did not identify their source IDE. Most affected agent activity was left out of usage reports. Some activity was counted as Copilot CLI use instead.[1]

The issue changes attribution, not billing. GitHub says agent interactions and agent-edit line counts stay undercounted on affected client versions. Earlier Copilot CLI numbers may be inflated. The lost records cannot be separated and corrected after the fact.

A falling chart can mean less work. It can also mean that the measuring path changed.

The repair starts at the client.

GitHub lists Visual Studio Code 1.139.0 and later as fixed. Visual Studio, JetBrains IDEs, Eclipse, and Xcode need later releases or plugin updates that GitHub expects to finish rolling out by November 2026.[1] Visual Studio Code published 1.139 on September 23.[2]

Updating stops the gap for later activity. It does not repair the old window. Teams should mark the affected dates on every dashboard and avoid comparing them with complete periods as if collection stayed constant.

One report contains two witness paths.

GitHub's metrics docs say server-side telemetry can keep an active user in top-level counts when client events fail. Detailed breakdowns still depend on richer IDE telemetry. Feature totals and line-of-code fields can remain empty even when the person appears in the active-user count.[3]

This split matters. An active-user chart, an agent-interaction chart, and an editor line-count chart do not share the same evidence. A proxy can block client telemetry. A user can disable it. An old extension can omit fields. A client migration can change the event shape.

Give recovery its own window.

GitHub says dashboard and API data can take two full UTC days to arrive.[3] A corrected client installed today should not be judged by today's incomplete report. Record the rollout time. Wait for the stated processing window. Then compare only the post-update period.

The dyno below maps the conditions each metric needs. It does not inspect an IDE, network, tenant, or dashboard. Use the card to plan a real check and annotate the unrecoverable gap.

Interactive makeover / metric blind-spot dyno

Trace the witness path.

Traditional purpose replaced: one green analytics dashboard. Better version: choose the metric, declare each collection condition, and see which witness path the field needs before reading the trend.

Choose one report field

These controls record declared conditions. They do not test a client or prove that data arrived.

Metric under review
Declared collection conditions
The client path is still incomplete.

The fixed version is declared. Telemetry, endpoint reachability, and the processing window still need evidence.

Sources read / October 7, 2026

Source log

Open the evidence boundary
  1. GitHub Changelog, "Update your IDE to restore agent activity in Copilot usage metrics". Read October 7, 2026. This is the first-party incident notice for the missing IDE attribution, affected report fields, version rollout, billing boundary, CLI inflation, and lack of backfill.
  2. Visual Studio Code 1.139 release notes. Read October 7, 2026. Microsoft dates the stable release to September 23, 2026. The page confirms the listed client version exists. GitHub's incident notice, not these release notes, supplies the metric-fix claim.
  3. GitHub Docs, "GitHub Copilot usage metrics". Read October 7, 2026. The docs separate server-side active-user detection from richer client-side breakdowns and state a processing delay of up to two full UTC days.

Evidence boundary: GitHub is the source for the incident, metric behavior, and rollout schedule. This page did not inspect a tenant or reproduce the attribution gap. The dyno is a condition map. It does not certify metric completeness, developer activity, productivity, or return on investment.