Skip to main content

Methodology & Data

MODULE SYLLABUS
📖7 min read

Introduction​

Whether you work in law enforcement, in journalism, or as a private investigator, you play by different systems, contexts, and rules. BVI cannot capture every jurisdictional requirement. The gate is a baseline for what is common to all of them. The gate is the floor: passing it means the research is defensible (a sound method) in your field and under any court's admissibility rules. Your professional context adds its own requirements on top of the gate. It tells you where your work stands before you apply them. If your context enforces looser rules, follow the gate anyway: the extra precision costs little and is available when the project demands it. The professional baseline this gate tracks is the Berkeley Protocol on Digital Open Source Investigations[1].

As of writing this module, GPT's latest model, Astra[2], can create full 3D models from reference images. The question is accuracy: are the models accurate, and how accurate? An investigation holds value only if it holds up to the real world. Does the scale match? Do the dimensions match? The working assumption is that a generated scene does not match reality, which is why we verify the human-built ones. That assumption is no longer safe. Even when a model produces something that does match, someone still has to verify it. This module makes that verification smooth: one ordered path from source to report, whether a human or a model placed the geometry.

Module Overview​

Est. time—
TypeMethodology — the research lifecycle
PrerequisitesNone. This is the starting point.
Start hereModule Contents — the walkthrough order for this module

Learning Outcomes​

By the end of this module you will be able to:

  1. Define a research question and approach before gathering data
  2. Collect and document sources with full provenance
  3. Preserve and store evidence so originals survive tampering and loss
  4. Verify that material is what it claims to be, with a verdict on record
  5. Analyze with a declared method and honest measurement
  6. Process data while documenting every change
  7. Validate and report with a verdict a reviewer can check
  8. Collaborate with one shared board, one linked scene, and a clean handover

Module Contents​

The Research Gate is an overview of the pages in this module, and a checklist you can measure your project against as you work. The rules change with the context you work in: a journalist, a law-enforcement officer, and a private investigator each play by a different system. The gate is the guideline they all share, and it keeps the result defensible. Each item links to the deeper Data page that explains the why and how. Work through the stages in order. Do not proceed to the next stage until the current one passes. Strictness is proportional: a forensic reconstruction requires the full gate, where a historical exploration may document N/A with reason. Passing the gate claims that the method was sound and defensible. It does not claim that the findings are true.

This module follows the data lifecycle: each stage builds on the last. Each step opens with a short introduction and closes with its pass/fail checklist.

Define· intro

Before you gather anything, fix what you are answering. A stated research question, a declared approach, and a methodology framework turn a reconstruction from an opinion into a documented investigation.

  • Research question written as one specific, answerable sentence.
  • Approach declared: confirmatory, exploratory, or mixed.
  • Blender justified (why 3D reconstruction for this question; alternatives considered/rejected).
  • Scope defined: what is in, what is out, the boundaries of the investigation.
Collect· 8 min read

Every source you use must be accounted for: what it is, where it came from, when you obtained it, and whether it is authentic. This is where the chain of custody begins, and a gap here undermines everything that follows.

  • Source registry started: a spreadsheet or document listing every source.
  • For each source: type, location/URL, access date, technical specs, and what you extracted from it.
  • Provenance recorded: who provided it, which repository, original URL.
  • Authenticity gate: for key evidence, how did you confirm it's authentic/unmanipulated? Flag any source you cannot authenticate.
  • Sources characterized for suitability: optics (distortion), artifacts, perspective lines.
  • Metadata handled: extracted where present; estimated camera parameters where absent.
  • Relevance confirmed: each source demonstrably supports the research question.
Preserve & Store· 6 min read

The originals must be protected and verifiable. That means a vault (a read-only folder for the originals), a SHA-256 hash for every file, working copies you edit, and a naming scheme that survives hand-off.

  • Vault created: a read-only folder for pristine originals, outside the working tree.
  • Hashes generated and logged (SHA-256) for all originals; stored in an evidence log.
  • Working copies separated from originals; originals never edited directly.
  • Folder structure chosen and applied consistently (Forensic / Workflow / Hybrid).
  • Naming convention fixed (ISO dates, project, description, version).
  • Version control plan in place (incremental + milestone saves).
  • Durability: where is this stored, and is it preserved against loss or tampering?
Verify· 7 min read

The originals are secured in the vault. This stage answers the question that decides whether the material can be used: is each item what it claims to be? Source, file, and content are checked together, and the verdict with its basis and weight goes to the evidence log.

  • Source analysis done for key evidence: provenance, credibility, independence.
  • Original file searched (reverse search, archives, EXIF, visual clues); the result, found or not, is logged.
  • File integrity re-verified against the vault hashes; metadata reviewed, platform stripping noted.
  • Content checked: internal consistency and external corroboration for the key claims.
  • Geolocation: taken and logged where fitted, closed and logged where it is not.
  • Verdict per item: authentic, inconclusive, or not authentic, with basis and weight.
  • No material processed without a verdict.
Analyze· 12 min read

This is where the model meets the method. You anchor to ground truth, isolate the objects of interest, and carry measurement error through the chain, so the numbers you report are the numbers the evidence supports.

  • Ground truth established: fixed spatial framework (satellite, structural measurements).
  • Key objects identified and catalogued: vehicles, actors, vegetation, street furniture.
  • Reconstruction decisions logged in the decision log: with evidence, alternatives, confidence.
  • Uncertainty quantified, not guessed: error sources listed, method chosen (RSS vs Monte Carlo), reported as value ± uncertainty (method, confidence).
  • Gap-filling explicit: any inference or interpolation flagged, with rationale.
  • Verification against independent sources where possible.
Process· 4 min read

Raw sources rarely arrive ready to model. Processing turns them into usable assets. Every transformation has to be logged, because a change you cannot document is a claim you cannot defend.

  • All processing on working copies only.
  • Every transformation logged: each filter, enhancement, crop, with settings.
  • Segmentation: clips split, frames extracted, sources grouped by device.
  • Temporal alignment: video synced to a single master timeline; anchors recorded.
Validate & Report· 8 min read

A reconstruction is not done when it looks right; it is done when it can be checked. The scene is assembled, verified against the sources, confidence is visualized, and the report states what was found, how, and where it is limited.

  • Scene assembled from verified layers.
  • Projection verification: source footage projected onto geometry, alignment confirmed.
  • Geometric validation: scale, angles, distances measured vs. ground truth.
  • Confidence map produced: color-coded output showing fact vs. inference.
  • Analytical report written: methodology, findings, conclusions, limitations.
  • Comparison charts / exhibits produced where relevant.
  • Limitations stated: margin of error, what remains uncertain/speculative.
  • Boundary awareness: report states what can and cannot be claimed from this analysis.
Collaborate· 3 min read

Most reconstructions pass through more than one pair of hands. A shared reference board, a linked scene, and a clean file handover keep one version of the truth as the work moves between people.

  • One reference board holds the images the team works from.
  • Models are linked, not copied, so one asset feeds every scene.
  • Each handover is logged: who, when, and the hash before and after.

Knowledge Check​

Test your understanding of the module before moving on. Ten questions with immediate feedback; the question order and options reshuffle on every attempt.

Methodology & Data

10 questions · immediate feedback · a different set each attempt

Check the research lifecycle — define, collect, preserve, verify, analyze, process, validate — before you move on.

Resources​

Sources​

  1. Berkeley Protocol (OHCHR): international standards for digital open-source investigations
  2. GPT-6 Astra (OpenAI): the 3D generation model referenced in the module overview