AI Driven SDLC
Login

AI Driven SDLC Documentation

Guides, references, and best practices for the AI Driven SDLC platform.

Insights

Ship Weight

Complexity-weighted throughput, shown next to Merged PRs.


What it means

Merged PRs counts every merge the same. Ship Weight answers a different question: how much review load those merges represent.

Each merged GitHub pull request gets an integer score from 1 to 10 (or 0 in the cases below). A week with 20 tiny docs PRs can look busy on Merged PRs and quiet on Ship Weight. A week with 4 dense code PRs can look the opposite.

Per-PR score What it usually means
0 Nothing left to score: empty diff, or only ignored files.
1 to 3 Small or lightly weighted change: few kept files, mostly tests/config/docs, or lots of deletions.
4 to 7 Typical feature or bugfix with real production code.
8 to 10 Large, dense, code-heavy diff. 10 is the cap; bigger PRs do not go higher.

Charts, KPIs, tables, and heatmaps sum those per-PR scores for the day or week you are viewing. Avg ship weight (weekly) (or per day) is the mean of those bucket totals, not the average of individual PR scores.

Ship Weight does not replace Merged PRs. Both stay available. It is also not raw line count (use PR size in SDLC Insights for lines).

How it’s calculated

After a pull request merges, the platform reads that PR’s file list from GitHub (path, add/modify/remove, and line changes). Same files always produce the same score.

1. Classify each file

Path rules run first (test folders, generated paths, and similar), then the file extension. Each file gets a type and a density (how information-dense that kind of file is). New files use a “new” density; everything else uses a “modified” density. A production-code path under test/ / tests/ / spec/ is treated as a test.

2. Weight the change

For each file we combine:

  • How much changed on that file (GitHub’s line-change count)
  • Density for that type of file
  • File type (code counts more than tests; docs and assets count little or not at all)
  • Change type (edits count more than adds; deletes and empty renames count little or nothing)

Tiny leftovers below a trivial threshold are dropped.

File kind How it counts
Production code Full weight
Tests Less than production code
Styles, templates, config Discounted
Docs Small
Assets, generated files, translations, most data files Usually ignored
Config-heavy PRs (about 75% or more of meaningful files are config or data) Config (and non-tabular data) is raised so infra/config work is not invisible
Change How it counts
Modified Full weight
Added Slightly less than a modify
Removed Low
Renamed with no other edits Ignored

3. Scale to 1 to 10

Kept files are added up, nudged slightly upward when many files are involved, then scaled onto 1 to 10 and rounded. The result is never above 10.

Special cases

  • Stacked / “slush” parents: When a parent PR has children already merged into its head branch, the parent only keeps the residual weight that those children did not already ship. That avoids double-counting stacked work.
  • Unscored yet: Merges still waiting on ingest show Backfilling in the UI, not a guessed number.

The in-app info tooltip labels this model v13. The weights above are the customer-facing summary; they are not a line-by-line formula.

Where it shows up

Productivity Insights

Open Insights → Productivity Insights. On Overview, By User, By Team, and contributor or team drilldowns, use the Merged PRs / Ship Weight tabs. The same toggle drives KPIs, trend charts, contributor and team tables, leaders/laggards, heatmaps, and the day drawer (including the per-PR list).

Default is Merged PRs. Your choice is remembered in this browser.

Your AI Journey

Open AI Journey in the sidebar (/dashboard):

  • Hero Ship weight next to PRs merged (last 7 days, with week-over-week delta when scoring is complete)
  • PR velocity & failure signal: tabs for Merged PRs, Ship Weight, and CI Failures (CI only when GitHub Actions is enabled for the workspace)
  • The embedded Productivity card uses the same Merged PRs / Ship Weight toggle

Backfilling

New or historical merges can show Backfilling until scoring finishes. Merged PR counts still work in that window. Switch back to Merged PRs if you need volume right now, or wait for ingest to catch up.

Same GitHub data as PR throughput

Ship Weight needs a connected GitHub App. No extra permission beyond what Insights already uses for pull requests. Historical merges without a score stay at 0 until they are backfilled.

Weekly targets

Workspace admins can set an optional weekly per-engineer Ship Weight floor under {workspace} settings → Productivity Insights, next to Merged PRs and CI Failures. See Productivity Insights.

Next steps