Z3X-1337

Research method

Bug bounty research starts with evidence.

A useful report begins before the first request: with authorized scope, a model of the expected security decision, and a plan to collect the smallest proof that answers the question.

Bug bounty research is not a race to produce unusual traffic. It is a controlled investigation inside rules set by a program. The quality of the result depends on whether another reviewer can see what was allowed, what should have happened, what actually happened, and why the difference matters.

Start with the authorization boundary.

Read the program policy before testing. Record the in-scope assets, permitted techniques, account requirements, exclusions, rate limits, data-handling rules, and safe-harbor terms. If an asset or action is ambiguous, pause and ask the program rather than treating ambiguity as permission.

  • Use accounts and data you are authorized to control.
  • Prefer the least invasive test that can answer the hypothesis.
  • Stop when testing could affect another user, service availability, or sensitive data.
  • Keep timestamps and target identifiers so evidence stays attributable.

Model the expected decision.

Before changing a parameter, describe the expected rule. For authorization, a compact model is often enough: which actor is performing which action on which resource, under which ownership, role, tenant, or workflow state?

ElementQuestionEvidence
ActorWhich authenticated identity sends the request?Controlled account, session, and role.
ActionWhat read, write, delete, or transition is attempted?Exact method, route, and request body.
ResourceWho owns the object and which tenant contains it?Identifiers created by controlled accounts.
DecisionWhat result should the server enforce?Documented behavior or a same-owner control request.

Write a falsifiable hypothesis.

“This endpoint may be vulnerable” is too broad. A testable hypothesis names the boundary and the observable result: for example, “Account B can update an object owned by Account A when only the object identifier changes.” That statement can be disproved by a denial and supported by a controlled state change.

Use a control request first. Confirm that the owner can perform the intended action, then repeat the smallest changed request as the non-owner. Controls reduce false positives caused by expired sessions, invalid objects, or a misunderstood workflow.

Validate without expanding the blast radius.

Collect only what establishes the security decision. A reproducible record usually needs prerequisites, redacted requests and responses, actor and resource relationships, before-and-after state, timestamps, and the expected result. Do not retrieve more data or repeat a destructive action merely to make the report look dramatic.

precondition  object-101 belongs to controlled account A
control       account A updates object-101 → accepted
test          account B sends the same action for object-101
observed      response and resulting state recorded
expected      server denies an actor without ownership

Keep secrets and unrelated personal data out of the report.

Report the demonstrated impact.

A strong report is easy to triage. Lead with one sentence that names the broken boundary and demonstrated consequence. Follow with prerequisites, numbered reproduction steps, the observed and expected results, and a concise impact statement. Separate facts from possible escalation.

  • Observed: what the controlled test directly established.
  • Impact: what that behavior allowed within the demonstrated conditions.
  • Potential: a clearly labeled scenario that was not tested.
  • Remediation direction: the decision the server should enforce, without pretending to know the target’s full architecture.

Evidence-first research is slower than guesswork at the beginning and faster at the end. It reduces noise, protects users, gives triage a repeatable path, and makes the final security claim no larger than the proof supporting it.

esc
↑↓ move↵ openg + h/p/r jump