$ cat Black Hat Starter Kit
A guided path through the archive: prerequisites, evidence labels, terminology, local teaching demonstrations and current versus historical material.
This archive explains search manipulation and defensive investigation. Start by learning what the evidence can establish. Then run the supplied local examples and inspect their output. You do not need a paid SEO tool, public website, search account or AI subscription to complete the starter path.
1 · Choose a question before choosing a tool
Read Methodology and Overview first. For each chapter, write down the system being discussed, the behavior it tries to influence, the evidence supporting that mechanism and one alternative explanation. Chapters differ in depth; a status strip or citation does not certify every claim. See the review tracker for editorial dates and remaining limitations.
Use / for full-text search, the Find index for a topic or tool record, and the page contents menu to jump within a chapter. On small screens, the menu button exposes navigation. The terminal opens with :; enter help to see the supported commands. Commands are local interface features; illustrative console displays do not scan websites.
2 · Read claims with the evidence hierarchy
| Label | What it can establish | What it cannot establish alone |
|---|---|---|
| Confirmed / primary | What a named official source documents, or what a primary record reports. | A platform policy is not disclosure of all its detection mechanisms. |
| Leak-indicated | A field or description appears in leaked documentation. | That it is currently used, its weight, or a reliable way to influence rankings. |
| Researcher interpretation | A named author’s inference and rationale. | Official confirmation or causal identification. |
| External experimental evidence | Reported results under the study’s particular design and conditions. | Results for different models, dates, populations or interventions. |
| Practitioner observation / anecdotal | A report that can generate a testable hypothesis. | A representative success rate or a controlled result. |
| Estimate | A calculation with disclosed inputs, assumptions and bounds. | Permission to retain an invented number without a basis. |
| Synthetic demonstration | The behavior of supplied code and invented fixtures. | Original evidence about a live search engine or language model. |
Keep these distinctions in your own notes. Copy the exact claim, the nearest source, the relevant passage and the limitation. If the source supports only a narrower claim, narrow your statement. A proposed study has no result until it is executed and its artifacts are retained.
3 · The first terms to learn
Crawling retrieves a resource; indexing processes material for retrieval; ranking orders eligible results for a query. Being crawled does not guarantee indexing or a position. A money site is the promoted commercial destination in practitioner terminology. A footprint is a shared attribute that can prompt investigation, not proof of ownership. Cloaking concerns deceptive differences presented to users and search engines. A manual action is a human review decision reported in Search Console. NavBoost is a named click-related system discussed in public litigation material; its existence does not prove purchased clicks work. Follow the glossary into the relevant chapters for sources and details.
4 · Set up the actual lab kit
The Offline Lab Library supplies code, input fixtures, expected output and an actual captured run. Download offline-defense.zip, extract it, and keep the files together. Use Python 3.10 or newer; no external packages are required. In a terminal inside the extracted offline-defense folder, run:
python3 run_labs.py --output-dir run-output # Windows, where the Python launcher is installed: py -3 run_labs.py --output-dir run-output
Expect six PASS lines, then PASS: 6 offline demonstrations, 33 assertions and confirmation that output matches the committed baseline. Open run-output/results.json to inspect the findings and runtime.json for the environment record. Existing output files with those names are replaced; choose a fresh output directory when preserving previous runs.
- Confounding: understand why changing multiple factors cannot identify a separate ranking effect.
- File integrity: separate approved changes from unexplained additions, deletions and modifications.
- Response comparison: distinguish formatting, content and access-control differences.
- Citation scoring: calculate citation counts on fictional answers without dropping misses.
- Instruction boundaries: inspect a deterministic toy that confuses document instructions with authority, then a deliberately constrained alternative.
- Poisoning-pattern triage: combine synthetic signals and retain benign counterexamples.
If Python is missing, use your normal supported software installation process. If a fixture file is missing, extract the whole archive instead of downloading only the script. An expected-output mismatch is a reason to inspect the inputs and local edits, not overwrite the baseline. See the kit README for all prerequisites and troubleshooting. No live ranking sandbox, WordPress environment or LLM harness is supplied.
5 · Follow a path that matches your question
| Goal | Reading and exercise | Checkpoint |
|---|---|---|
| Understand traditional search manipulation | Search Manipulation → PBNs → Lab 1 | Explain the proposed signal and one confound without asserting a success rate. |
| Investigate a site | Detection & Defense → SEO Poisoning → Labs 2, 3 and 6 | Distinguish observed change, supported diagnosis and next action. |
| Study AI search | API Leak → AI/GEO → Prompt Injection → Labs 4 and 5 | Separate citation selection from an instruction crossing a trust boundary. |
| Review local visibility | Local SEO → Case Database → detection monitoring | Explain why unusual review timing is a lead, not proof of fraud. |
| Design research | Methodology → Proposed Studies → Bibliography | Specify an independent unit, comparison, outcome and stopping rule before collecting data. |
6 · Read dates and status correctly
The research cutoff is 2026-09-19. An editorial date records a content revision; the build date records file generation. Neither means that every source was freshly verified. Historical cases document a particular event, not current viability. Status labels describe the chapter’s scope or editorial assessment; they are not measured market shares or detection probabilities.
Use Timeline for dated policy changes. Current structured-data guidance differs from older FAQ and HowTo advice. The tools catalogue distinguishes current pages, historical products and unresolved availability; pricing that has not been verified is left unquoted. An entry’s existence is not a recommendation to buy or deploy it.
7 · Keep the boundaries distinct
Platform policy, contractual terms and law are different authorities. A search-policy violation does not by itself establish a criminal offense. Legal outcomes depend on the conduct, jurisdiction and applicable elements; the sourced local and security chapters describe specific rules without treating every technique as automatically illegal. Ownership also does not exempt a public experiment from platform policies. The starter exercises use invented local data and make no network requests.
Before extending a demonstration, define the systems you may access, the data you may collect and the effects of the proposed action. Keep sensitive logs out of shared examples. For this starter path, inspect the fixtures and modify a separate copy; no third-party target is needed.
8 · A concrete first session
- Read the methodology and choose one question from the table above.
- Run the offline kit and preserve its output in a new folder.
- Explain one PASS result using the exact fixture and code that generated it.
- Write one claim the demonstration cannot establish.
- Read a relevant case record and distinguish its sourced outcome from missing information.
- If further evidence is needed, draft a proposal using the study templates. Leave the results section empty until observations exist.
This session gives you a reproducible starting point. It does not complete the guide’s proposed original research or independently validate any vendor, ranking tactic or defense.