Use a matched build recipe
Choose a source revision and record its build inputs. Produce a normal release using the application toolchain, then derive the treated specimen with the documented defensive step. If the treatment requires different compiler flags, identify those flags as part of the intervention instead of pretending only the post-build tool changed.
Retain the source hash, tool versions, dependency lockfile and relevant build configuration. Signing and packaging stages should follow the same supported process for both conditions. Avoid comparing an unoptimized debug build with a signed optimized release and attributing every difference to an anti-cloning tool.
Include realistic release exposure
A control should represent what a responsible team would normally ship. If production maps are intentionally private, do not include them only in the untreated condition to manufacture a large advantage. Alternatively, if accidental map exposure is the problem being studied, name that as a release-hygiene experiment with its own outcome.
Use several fixture families when making a broader claim. A JavaScript function, Electron archive and Swift executable exercise different consumers and compatibility constraints. Keep each result separate until there are enough observations to justify aggregation.
Prevent contamination across runs
Place each specimen in a clean workspace and prevent the reconstructing agent from reading reference source, hidden tests or previous trial outputs. Use neutral labels so the condition name does not reveal what workaround to attempt. Record which public information is intentionally available, such as documentation or interface screenshots.
Repeated trials should not share mutable caches or generated notes unless that sharing is explicitly part of the experiment. If a tool installation or missing dependency changes midway, either repeat the matched pair or record the environmental difference. Quietly repairing only one condition can distort the comparison.
Inspect the differences before scoring
Compare the control and treated inventories to confirm the intended files changed. Run behavior and native validation on both final specimens. A malformed treated archive is an experimental failure requiring diagnosis, even if an analyzer refuses it. A control that fails to build cannot establish that the defense caused the other condition to fail.
Report absolute results for both conditions before showing ratios. For example, show completed reconstructions out of attempted trials, hidden tests passed and actual elapsed observations. A large percentage change based on one tiny example can otherwise look more general than the evidence supports.
Practical checklist
- Match source and normal Release settings.
- Document intentional exposure differences.
- Isolate agent workspaces and hidden tests.
- Validate both final specimens before scoring.
Sources and further reading
Primary references support the technical details. Procedures here are guidance; only explicitly recorded project results describe tests we ran.
Prepared with AI assistance. How this library is produced. Suggest a correction; keep private code and credentials out of public issues.