The team evaluation brief
Is better codebase context worth it for our team?
Test that on a real task before making a buying decision.
Z1P Core scans local repositories and exposes source-linked records and relationships through a CLI and MCP adapter. The question for an evaluation is whether that helps us find the right evidence with less repeated investigation.
The proposal
Give one developer a time-boxed evaluation on one approved repository and one representative task. Start with local scanning. Agree separately before connecting an AI provider or sharing any evidence.
- Choose an unfamiliar module, bug investigation or dependency change.
- Record the usual investigation time and the files needed to understand the task.
- Run a bounded Z1P scan. Inspect the extracted records and any reported omissions.
- Follow the relationships and verify each useful result against the source.
- Compare findings, effort and context size. Record setup time and missed evidence as well as benefits.
What would count as a useful result?
| Measure | What to record |
|---|---|
| Evidence coverage | Were the files needed for the task found? Which connections were missed or misleading? |
| Developer effort | Investigation time, setup time and time checking the results against the source. |
| Context cost | If using an AI client, compare actual submitted context for equivalent tasks. Include follow-up source reads. |
| Repeat use | Would the developer choose it again on the next task without being prompted? |
Cost and availability
The local core is MIT licensed and free to use without an account or subscription. Budget developer time for setup and evaluation, plus any AI provider charges if we choose to connect one.
The hosted team product is in development. Connected repositories, automatic refresh, hosted MCP, pull-request impact and team administration are planned, not available purchasing commitments. Pricing has not been announced.
Data and technical boundaries
- Core scans run locally. They don't execute the repository or contact a model.
- Uploads and sharing require explicit configuration. AI clients may send retrieved evidence to their provider.
- TypeScript and JavaScript use compiler-based extraction. Other supported languages use more conservative lexical navigation.
- Static relationships are investigation leads, not proof of runtime behaviour or test coverage.
- Scans and collections have explicit limits. Review omissions before drawing conclusions.
- Signatures identify assertions; they don't establish that an assertion is true.
The decision afterwards
Continue if the evidence is useful, the handling of our data is acceptable, and the benefit justifies the setup and review effort. If not, record what failed. A smaller context payload alone isn't a successful evaluation.
Read the setup and scanner documentation, the benchmark methodology and the open-core commitment.
Z1P by ForgeSworn · z1p.app/team-brief.html · September 2026