At a glance
| Input | Action | Output | Limit |
|---|---|---|---|
| Selected Confluence spaces | Count persisted terms, pages, usages, and findings | Seven-part demand estimate | Counts cover the selected scanned scope only |
| Freeze estimate | Compare demand with policy bounds before writes | Accepted or rejected decision | Acceptance is a safety gate, not a benchmark |
| Page and term evidence | Process up to 100 source units per worker page | Resumable freeze with checkpoints | Completion time still depends on live Forge and tenant data |
| Local qualification workload | Stream 250,000 fixtures across eight artifact classes | Recovery records, roots, and modeled metrics | Not Forge latency, cost, consumption, residency, or throughput evidence |
| Confluence-document fixture class | Generate 31,250 of the 250,000 local artifacts | Deterministic document-shaped records | Fixtures are not live Confluence pages |
| 200,000-page target | Replay a representative workload in Forge | Measured pass or documented constraint | No completed Forge qualification result is recorded |
How it works
- Define the target page count, content size, term density, findings, links, and files.
- Record the deployed Forge version, environment, migration, dependencies, and data shape.
- Run capacity admission for the selected scope.
- Replay a representative workload in an authorized Forge environment.
- Measure latency, duration, failures, recovery, events, storage, and cost.
- Publish the measured result with its exact conditions and limits.
Example
A team has 200,000 Confluence pages across many spaces. Capacity admission can estimate a selected, scanned baseline and block work above policy bounds. It cannot prove completion time for the full site. The local 250,000-artifact simulation contains 31,250 Confluence-document fixtures, not 250,000 live pages. A representative Forge run remains necessary.
What this does not mean
Bounded processing and admission checks are engineering controls. They are not benchmark results. The harness marks Forge latency and cost evidence as false. Current qualification records contain no completed Forge staging or production measurement. No verified 200,000-page capacity claim should be published.
Related questions
Next step
Evaluate the intended scope before adopting a scale assumption.
View on Atlassian Marketplace