Each project follows an evidence-driven, phase-based process. What is constant across projects is the evidence chain: pinned source → triaged analysis → controlled verification → patch validation → report.

1. Source acquisition and pinning

Source is acquired from official upstream repositories at a specific commit or release tag (recorded in every document). The audited tree is kept pristine; all builds are out-of-tree in dedicated directories; tag signatures are verified where available.

2. Baseline and clean builds

Configure/build logs are recorded per build variant (normal, sanitizer, static-analysis). Upstream test suites are run before auditing to establish a behavioral baseline (e.g. xrdp 213/213 unit tests with 0 compiler warnings before any audit work).

3. Architecture and attack-surface analysis

Written architecture, trust-boundary, and attack-surface documents are produced before auditing. The target’s process model (privilege levels, fork-per-connection, per-session helpers) is established so impact assessment has a factual basis.

4. Static analysis

clang scan-build / clang --analyze, cppcheck, and GCC/Clang warning passes — every report triaged to confirmed defect, benign pattern, or false positive with recorded rationale. Negative results are retained.

5. Manual source review

Deep audit of parsing, protocol, crypto-adjacent, and memory-safety- critical paths. Recent upstream security fixes are studied to identify sibling defects of the same class, including incompleteness in shipped fixes.

6. Candidate root-cause analysis

Every candidate is traced from attacker-controlled input to the faulty construct, including build-flag interactions (e.g. how -ftrivial-auto-var-init=zero masks uninitialized reads in shipped builds).

7. Dynamic verification

Sanitizer builds (ASan/UBSan/MSan) from the exact audited commit, Valgrind memcheck cross-checks, and purpose-built harnesses linking the unmodified vulnerable translation units replaying attacker-controlled input. Control experiments are mandatory: a run demonstrating that the defect is specific to the identified code path, and post-patch runs demonstrating a clean result.

8. Patch development and regression

Researcher patches are validated three ways: harness clean, functional sanity tests pass, and the official upstream test suite passes. After validation the tree is restored to the pristine pinned revision and re-verified.

9. Impact assessment and severity

Impact is bounded by the validated process model; severity is expressed as a proposed CVSS vector with reasoning and is explicitly left for vendor triage to finalize. No score is presented as authoritative.

10. Reporting and responsible disclosure

A confidential vendor-ready report is prepared per project and held until submitted through the project’s official security policy. Public documentation is redacted while issues are unpatched — no PoC code, trigger inputs, or reproduction steps are published.