Skip to main content

Define

TECHNIQUE
📖26 min read

Introduction​

Approach

Confirmatory vs exploratory — decide how you will test or explore the research question.

Scoping

Define what is in and what is out before opening Blender. Scope is methodology.

Standards

The Berkeley Protocol provides the internationally recognized standards for digital open source investigation.

This is Stage 0 of the Research Gate. Before collecting a single source, you need to define what you are investigating, how you will approach it, and why 3D reconstruction is the right method. This page is the foundation for everything that follows.

What Is Methodology?​

Methodology is the systematic framework you use to answer a specific research question. It's not a tutorial on how to use Blender; it's the documented reasoning behind every decision you make during an investigation. A methodology explains what you did, why you did it that way, what alternatives you considered, and what your approach can and cannot tell you.

The key distinction: Anyone can create a 3D visualization. Methodology is what transforms that visualization from illustration into evidence, something others can evaluate, reproduce, and build upon.

In forensic contexts, this distinction has legal weight: it's what separates an inadmissible graphic from defensible analysis. In academic contexts, it's what makes your work scholarly rather than speculative. In any investigative context, it's your audit trail.

Not every project needs to reach full evidentiary weight. Tier 0 visual insight is legitimate work, as long as it claims nothing it cannot support. The target tier you choose (below) sets how much rigor you owe.

This approach aligns with internationally recognized standards for digital investigation, particularly the Berkeley Protocol on Digital Open Source Investigations, which establishes that rigorous methodology, not just compelling visuals, is what makes digital evidence credible and potentially admissible in legal proceedings.

The Pipeline​

The module follows one ordered path from source to report:

Define → Collect → Preserve & Store → Verify → Analyze → Process → Validate & Report, with collaboration threaded through the whole.

Verify sits after the material is preserved and stored: the content is authenticated once the original is secured, before any processing touches it.

Evidence Types​

A reconstruction fuses several evidence types, and the audit trail records which claim comes from which one. The types this knowledge base works with, and what each one carries:

TypeWhat it carriesHandled in
Photographic imageryMeasurable geometry, EXIF data, chronolocation anchorsGeolocation
Video & footageMotion, time, perspective; frame extractionCamera Tracking
Drone captureNear-orthographic ground view, scaleCourse: site reconstruction
Satellite imageryWide context, change over timeCase studies
Maps & plansIntent, layout, dimensionsCase studies
3D scans & capturesPoint clouds, ready mesh assetsSourcing Models
TestimonyMemory, interpretation, lived experience. Carries meaning, not measurements. Record it, version it, and never let the model override it.Situated Testimony · Bilthoven Cold-Case

Provenance, Custody, and the Audit Trail​

The stages keep three records. They answer three different questions, and they can fail independently of each other.

Provenance is the origin or earliest known existence of the content: for online material, the earliest appearance online, or the original item before it was uploaded.[1] A photo has solid provenance when you can show who took it and when. For online content, record the first copy found online, not the first copy online: the original may already be removed, and closed channels — email, private messages — may have carried the item before it appeared publicly. Provenance is the one record you can never fully prove.

Chain of custody is the chronological documentation of the sequence of custodians of the evidence, and of its control, date and time, transfer, analysis, and disposition at each step.[1] For digital files, the hash carries the proof: compute it at acquisition, re-verify it at every handoff, and a match shows the copy is unaltered. The point is not ownership, it is integrity.

Provenance and chain of custody can fail independently. A photo can have solid provenance and a broken chain of custody — you can show who took it, but not that your copy is unaltered — or the reverse. This module keeps the two records in separate stages: provenance is documented in collect, custody in preserve-store.

Audit trail is the record of the investigation itself: the question, the sources, the searches, the decisions, and the dead ends, kept so another trained investigator can follow what you did and repeat it. The methodology section of your report is the audit trail made public.

The Berkeley Protocol: International Standards for Digital Investigation​

For those conducting investigations that may have legal, human rights, or accountability implications, it's valuable to understand how visual investigation methodology aligns with international standards.

The Berkeley Protocol on Digital Open Source Investigations[1] (published 2020 by the UN Office of the High Commissioner for Human Rights and UC Berkeley Human Rights Center) establishes minimum professional standards for using digital information in investigations of human rights violations and international law. While it focuses primarily on social media and online content, its core principles directly apply to 3D reconstruction work.

Key Principles​

The Protocol's five-phase framework (identification, collection, preservation, verification, and analysis) maps directly onto the stages of this Data section:

  • Identification = defining your research question and sources (this page)
  • Collection = documenting data sources with chain of custody (collect)
  • Preservation = working on copies, maintaining file integrity, version control (preserve-store)
  • Verification = authenticating the source material (verify) and testing model accuracy against independent sources (validate-report)
  • Analysis = using the model to answer your research question (analyze)

The Protocol emphasizes that methodology transforms digital content from illustration to evidence. It's not enough to create compelling visuals; you must document the reasoning, choices, and limitations that allow others to evaluate your work's credibility.

Why the Protocol Exists​

The Protocol was developed because digital information increasingly appears in courts, tribunals, and accountability proceedings[2]. Its standards help ensure that digital evidence is:

  • Admissible: Properly collected and documented
  • Reproducible: Others can understand and potentially verify your process
  • Credible: Transparent about limitations and uncertainties
  • Ethical: Respectful of privacy, safety, and appropriate use

A U.S. consensus standards body makes the same point from the practitioner side: “Digital and multimedia evidence, as well as the tools, techniques, and methodologies used in an examination, are subject to challenge in a court of law or other formal proceedings. Proper documentation is essential in providing individuals the ability to reproduce the forensic process and the results.”[10]

Even if your visual investigation work isn't destined for legal proceedings, following these principles strengthens your methodology. The Protocol's tool-agnostic approach (focusing on principles rather than specific software) reinforces that methodology is about reasoning, not just technical steps.

Methodology in the Field: What Real Reports Contain​

Direct lineage to the Berkeley Protocol is the common thread across all serious examples of investigation methodology. They differ in form — a court-facing manual, a technical methodology report, a published case-study walk-through, a peer-reviewed paper — but the methodology section contains the same five elements, regardless of document type:

  1. Research question + intended use of the findings, stated up front. Admissibility, publication, a committee of inquiry? The answer shapes everything that follows.
  2. Provenance and chain of custody, on equal footing with technique. Devices, accounts, tools, storage — documented with the same care as the analysis method.
  3. Reproducible method detail. Named tools with versions, parameters, reference points; measured versus estimated values. Written for an auditor who will try to repeat it.
  4. Explicit limitations + confidence language. "Appears to be," "suggests," quantified margins of error.
  5. Documented dead ends. Failed attempts kept on the record as evidence of rigor.

Neither the Protocol nor Bellingcat/GLAN covers 3D reconstruction specifically. For that: Villa et al. (2023)[9] is a peer-reviewed review of exactly this domain — and the only methodology source in this set that names Blender as a tool; Maksymowicz et al. (2014)[12] is a published case report of reconstruction from fragmentary evidence; the two Rangelov studies (2024, 2025) document the capture side for crime scenes; and the photogrammetry (2008) and 3DGS (2026) papers quantify accuracy. All appear below and are used in the framework that follows.

The Examples​

ExampleTypeWhat it shows
Bellingcat / GLAN — Justice & Accountability Unit Methodology (2022)[4]Court-facing OSINT methodology, self-declared Berkeley Protocol–compliantIntended use stated up front; first section documents the tools themselves; a full bias annex
Amnesty International — Forensic Methodology Report: How to catch NSO Group's Pegasus (2021)[5]Technical forensic methodology reportClaim under test in the first sentence; per-target traces; independent peer review by Citizen Lab
Bellingcat — How we Geolocated a Photo of a Russian Missile Programming Team (2022)[6]Published case-study walk-throughThe whole method as prose; dead ends kept on the record
Forensic Architecture / Forensis — Arena Bar methodology report (2021)[3]Commissioned investigation reportWorked example in Set Your Target Tier
Agosto et al. — Crime Scene Reconstruction Using a Fully Geomatic Approach (Sensors, 2008)[7]Peer-reviewed reconstruction paperGCPs vs. independent check points; redundancy for accuracy evaluation
Accuracy of 3D Gaussian Splatting for virtual crime scene reconstruction (Frontiers in Computer Science, 2026)[8]Peer-reviewed accuracy study of this course's own technologyQuantified millimeter-level error; failure modes named
Villa, Lynnerup & Jacobsen — A Virtual, 3D Multimodal Approach to Victim and Crime Scene Reconstruction (Diagnostics, 2023)[9]Open-access peer-reviewed review of 3D forensic reconstructionNames Blender as an animation tool; tested scenarios must depict several “realities”; bias and reproducibility expectations stated outright
SWGDE — Requirements for Report Writing in Digital and Multimedia Forensics v1.0 (2018)[10]U.S. consensus spec of the required report structureThe minimum elements: the request (details, purpose, scope); results for a layperson; the opinion and its basis; deviations from standard procedure disclosed
SWGDE — Best Practices for Digital Forensic Video Analysis v2.0 (2025)[11]Video-domain methodology workflow“Inconclusive” as a legitimate answer; margin of error tied to frame information and compression
Maksymowicz et al. — Crime event 3D reconstruction based on incomplete or fragmentary evidence material (Forensic Science International, 2014)[12]Published case report: reconstruction from fragmentary evidenceA long-modified scene rebuilt from testimony, a visit video, photos, Google Maps/Street View and construction archives — five candidate versions of the event, three ruled inconsistent (abstract; full text paywalled)
Rangelov et al. — Impact of Data Capture Methods on 3D Reconstruction with Gaussian Splatting (J. Imaging, 2025)[13]Open-access crime-scene capture study (NeRF + 3DGS)Camera path, speed and orientation as documented variables; a dedicated limitations section — the cleanest method → limitation structure in the modern stream

Also examined: Human Rights Watch's organization-level Our Research Methodology — purpose and scope, interview method, corroboration principles, and a “Specific Methodological Challenges” section published up front — as the model of an organization publishing its scoping and limits, not just its findings[15]; a companion study of camera settings (ISO, shutter, aperture → 3D quality) for the same crime-scene context[14]; and “Tracking People and Cars Using 3D Modeling and CCTV” (Forensic Science International) — on point for camera-matching and vehicle-speed work like the Hanau example, but paywalled[16].

The Core Question: What Are You Trying to Answer?​

Before opening Blender, write down your specific research question. Everything else follows from this.

Strong research questions are specific and answerable:

  • "What was the driver's line of sight at the moment of impact?"
  • "Could the witness positioned at the doorway have seen the events they described?"
  • "How did window placement affect natural lighting in the workshop space?"
  • "What was Vehicle A's approximate speed between these two points?"

Weak research questions are vague or unmeasurable:

  • "What happened during the incident?"
  • "What did the building look like?"
  • "How did people experience this space?"

These might be your ultimate goals, but they're too broad for methodology. Break them into specific, testable components.

This question is the exit criterion for Stage 0. You cannot pass the gate without a defined factual question — or an explicit Tier 0 declaration that the project makes no factual claim.

Set Your Target Tier​

Defensibility is not binary. Work climbs a ladder, and your project should know which rung it is aiming for:

TierNameWhat it demands
0Visual InsightA clear question and honest labeling. The work claims nothing it cannot support.
1Procedural SoundnessChain of custody and provenance, hashes, and a defined factual question the reconstruction answers.
2Methodological RigorA repeatable method, reported uncertainty, and independent verification of the key findings.
3Defensibility and Court-ReadinessA complete report with exhibits, explicit boundary awareness, and work that survives being challenged.

The Research Gate is the floor: passing it makes your work defensible — a sound method — regardless of your field. Your context adds requirements on top of that; jurisdiction-specific admissibility is not the method's target. Your tier determines how much documentation each stage owes you, and your final report should state which tier the project reached and what that tier does and does not claim.

Write it down now: your target tier (0–3) and why.

Worked Example: What a Tier 3 Define Looks Like​

The strongest public example of a top-tier define is the Hanau Arena Bar methodology report by Forensic Architecture and its Berlin partner Forensis[3] — a 21-page document presented to the Hanau Committee of Inquiry in the Parliament of Hessen (20 December 2021). Its central finding: if the bar's emergency exit had been open on the night of the attack, the five young men inside would very likely all have survived — directly rebutting the public prosecutor's decision to discontinue the emergency-exit investigation.

Its define part does four things your Stage 0 should mirror:

  1. The forum is declared up front. Cover: "Presented to the Hanau Committee of Inquiry in the Parliament of Hessen." Section 1: commissioned "to produce material… across legal, political, and cultural forums, as well as media, activism, and advocacy" — and the document labels itself preliminary throughout. The forum names the tier; the label states the confidence.

  2. Explicit negative scope. "FA's research does not examine whether the emergency exit was indeed locked on the night of the attack. Instead, our research examines the public prosecutor's claim, beginning from the assumption that the five young men… believed that they would find the emergency exit closed."

  3. One-sentence question, bounded and verbatim. "If any of those present in the Arena Bar at 10pm on 19 February 2020 had run toward the emergency exit, and it had been open, would they have had time to escape?" — with the materials declared ("a copy of the recordings made by the CCTV cameras") and the honesty marker: "preliminary research results, that may have to, or can be, updated as new facts become known."

  4. A conditional finding from page 1, held to the end. "If the emergency exit door… was open… and if the five young men… had attempted to escape through it… then they would very likely all have survived." The conclusion rebuts the prosecutor's specific claim, then points to the open question (was the door locked?) as the larger issue — boundary awareness in practice.

Note what the report does not contain: no decision log. Its audit trail is the named tools (Premiere Pro, Cinema4D), a footnote citation for every testimony, and a dedicated Caveats and considerations section that argues its speed estimates are conservative, citing running-performance literature. The decision log is this module's own mechanism; the principle is shared — every claim must be traceable to evidence.

Setting Up Your Methodology: A Practical Framework​

Methodology has four essential components. The complexity of each scales with your project; a forensic reconstruction requires more rigor than a historical exploration, but the basic structure remains the same. Each step closes with in the field — how the published investigations above actually did it.

The U.S. consensus spec for forensic report writing (SWGDE, 2018) requires the same four moves in its own structure: the request's details, purpose and scope; results described for a layperson, with any deviation from standard operating procedure disclosed; the opinion and the basis for it; and the disposition of the evidence[10].

1. Define Your Approach​

Ask yourself: Am I testing a specific claim (confirmatory) or exploring to understand spatial relationships (exploratory)?

Confirmatory investigations test hypotheses with measurable precision:

  • "The driver could not have seen the pedestrian from their position" (testable: yes or no)
  • "The reported speed is inconsistent with the observed distance" (testable: calculate and compare)
  • "Window X provided insufficient daylight for detailed work" (testable: lighting simulation)

Exploratory investigations examine open-ended spatial questions:

  • "How did the spatial layout facilitate different activities?"
  • "What sight lines were available from various positions?"
  • "How did lighting conditions change throughout the day?"

Most projects combine both: you might explore spatial relationships to generate hypotheses, then test specific claims confirmatorily.

Document:

  • Your research question (one clear sentence)
  • Your approach (confirmatory, exploratory, or mixed)
  • Why Blender specifically suits this question
  • What alternatives you considered and why you rejected them
  • Which techniques you will disclose in the report — including generative ones, if any. Decide now, not at report time

Example: "This investigation tests whether Witness B could have seen the traffic signal from their reported position (confirmatory). Blender was selected because it allows precise camera matching to the CCTV footage while simultaneously positioning a viewpoint at the witness location. Two-dimensional photo analysis cannot account for the three-dimensional relationship between elevated signal, roadside obstructions, and witness position. Specialized forensic software was considered but lacks the camera-matching flexibility needed for non-standard mounting angles."

In the field - the intended use stated up front (element 1). The Bellingcat/GLAN manual opens by saying exactly what the work is for: "Its aim is to ensure that any material discovered... is gathered in accordance with rules on admissibility of evidence so as to make it suitable for use in future legal proceedings and other accountability processes"[4]. The Amnesty Pegasus report puts the claim under test in its first sentence: "NSO Group claims that its Pegasus spyware is only used to 'investigate terrorism and crime' and 'leaves no traces whatsoever'. This Forensic Methodology Report shows that neither of these statements are true."[5] Both answers — for a court, for a publication — come before any method is described.

Villa et al. state the problem a define-stage must handle: “The tested scenarios should be able to depict several ‘realities’ and chain of events as it can be difficult to decide which ‘reality’ is the correct/most plausible one” — which is why the approach must name the alternative versions considered, before any method is described. The same authors add: “There is a potential for subconscious biases to come into play, which can potentially sway the verdict of a jury” — disclosure of stance and confidence (elements 4–5) is the counterweight[9].

2. Document Your Sources (Data Collection)​

The full collection methodology lives in Collect: the source registry, the Source Documentation Template, provenance, the authenticity gate, technical assessment, and metadata extraction.

Preservation principle: Work on copies of original evidence. Document any transformations or processing applied to source materials. This ensures that original evidence remains unchanged and that your process is transparent and reproducible.

In the field — provenance on equal footing with technique (element 2). The GLAN manual's first section is not about the case at all: it is about the tools. "Dedicated Device: You are expected to use a dedicated work device for your Justice and Accountability investigations… Ensure the hard drive of your laptop is encrypted and that you have set a very strong password" — plus a Device Log Sheet, a Mnemonic Preservation Sheet, and a browser plugin (Hunchly) that "tracks your online activities, preserving essential information about the webpages"[4]. The Amnesty report names the software used to extract evidence ("libimobiledevice and checkra1n were used extensively") and the data providers (Censys, RiskIQ passive DNS), and closes with per-target evidence: Appendix D, "Pegasus Forensic Traces per Target"[5]. The Bellingcat walk-through documents where the photo under investigation came from — a source who "contacted reporters via a burner email account" — and still requires the source's claim about the location to be independently verified[6].

3. Document Your Process (Method)​

Describe the specific techniques you used in Blender to transform sources into a spatial model.

Break this into:

a) Establishing spatial accuracy

  • How did you calibrate cameras? (fSpy, manual matching, reference points used)
  • How did you verify scale? (What known measurements did you use as references?)
  • How did you handle photogrammetry or point cloud data? (Software, settings, alignment method)

b) Reconstruction decisions

  • How did you model specific elements? (Which tools, which techniques)
  • When evidence was incomplete, how did you fill gaps? (Comparative examples, period-typical details, educated inference)
  • How did you prioritize conflicting sources?

c) Analysis techniques

  • How did you position viewpoints for sight line analysis?
  • What rendering settings simulated lighting conditions?
  • How did you measure distances, speeds, angles?
  • What physics simulations did you run?

The key principle: Someone with similar skills should be able to understand your process well enough to evaluate whether your approach was sound, even if they can't perfectly replicate every step. This aligns with the Berkeley Protocol's emphasis on reproducibility; others should be able to follow your reasoning and potentially verify your findings.

Document:

  • Blender version and any add-ons used
  • Specific techniques with tool names (not just "I tracked the camera" but "Motion tracking using Blender's built-in tracker, placing markers on static reference points X, Y, Z")
  • Where precision was possible versus where estimation was necessary
  • How you verified accuracy

Example: "Camera calibration used fSpy 1.0.3 with three vertical reference lines (building corners visible in frames 100-150) and two horizontal references (parking space edges). Resulting focal length: 8mm, sensor width: 13.2mm. Verification: rendered view overlaid with source footage showed alignment within 3 pixels for all reference points. Vehicle tracking: keyframes placed every 10 frames matching observed position using semi-transparent overlay. Position accuracy estimated ±20cm based on pixel resolution at vehicle distance."

In the field - reproducible method detail (element 3). Agosto et al. name every instrument by model: a total station (Leica TPS1200+), a time-of-flight laser scanner (Riegl LMS-Z420i), a triangulation scanner (Creaform Handyscan 3D). And they separate the points used to build the model from the points used to test it: reference points "can be used both as Ground Control Points (GCPs), required for the georeferencing procedures... and as independent Check Points (CPs), useful for the evaluation of accuracy," collected jointly "in order to have a redundant number of measurements... the availability of redundant data allows one to evaluate the accuracy of the estimated coordinates"[7]. The 2026 3DGS study describes its calibration the way an auditor can repeat it: a reference object "is selected and physically measured (Value A)" to rescale the whole scene, executed "within general-purpose 3D modeling programs that support standard units, such as 3ds Max," with virtual measurements compared against "actual objects using a tape measure"[8]. The Bellingcat walk-through works the same way in prose: cross-referenced "car-registration data and delivery orders," satellite imagery, geotagged photos - even the map's orientation is flagged ("note: map below has been reversed from north")[6]. The GLAN manual's one-line version of the same rule: “Make descriptive notes of your verification work performed on each item of content”[4].

Villa et al. state the peer expectation for a new method: “It is paramount to systematically test, validate and investigate the advantages and disadvantages before incorporating a new method as routine tools to ensure its implementation without compromising the judicial system and introducing bias.”[9]

4. Acknowledge Limitations​

Every approach has constraints. Stating them openly demonstrates critical thinking and actually strengthens credibility, a core principle of the Berkeley Protocol, which emphasizes that transparent acknowledgment of limitations enhances rather than undermines the value of evidence.

Common limitations:

Source quality: Low resolution footage, damaged documents, incomplete measurements, contradictory accounts

Technical constraints: Modeling simplifications, render approximations, occluded areas requiring inference

Scope boundaries: What you deliberately excluded (distant features, interior details, environmental factors)

Methodological limits: What your approach fundamentally cannot determine (what someone was actually looking at, cognitive factors, attention, decision-making)

Document:

  • Specific constraints and their implications
  • Margin of error where quantifiable (±X cm, ±Y km/h)
  • What remains uncertain or speculative
  • Why your approach is still valuable despite limitations

Example: "Source footage resolution (720x480) limits spatial precision to approximately ±15cm at relevant distances. The area behind the delivery truck is occluded in all footage; vehicle trajectory through this zone is interpolated based on entry/exit positions. This reconstruction establishes physical possibility of sight line, not proof of actual observation; driver attention and interior vehicle obstructions are unknown."

In the field - limitations as quantified statements (elements 4 and 5). The 3DGS study reports margins, not adjectives: "mean measurement errors ranging from 0.25 to 0.65 mm in the preliminary test and from 1.73 to 3.58 mm in the main test," names the failure modes ("artifacts caused by reflective surfaces (glass, glossy materials) and reduced visibility of minute bloodstains"), and draws the exact line between what holds and what doesn't: thin objects show higher relative errors, but their "absolute physical precision remained intact"[8]. The Hanau report holds its honesty marker from page 1 to the conclusion: "preliminary research results, that may have to, or can be, updated as new facts become known," and keeps its finding conditional[3]. Dead ends stay on the record: “With the Frunze Naberezhnaya facility ruled a less-likely candidate, despite the red herring of the white walls…” — keeping a ruled-out path in the text shows the audit trail of your judgment[6].

SWGDE's video best practices turn not-overclaiming into a norm: “In instances where the opinion cannot definitively answer the question being asked, an answer of ‘inconclusive’ may be the only appropriate response. In instances where a quantitative response is required, a margin of error may be expected, based on the frame information and compression of the images.”[11]

Documentation Practices: What to Keep Track Of​

Methodology isn't just written at the end; it's built throughout your process. Set up documentation habits from day one.

Decision Log​

Keep a running document (simple text file, or spreadsheet) of every significant modeling choice. One entry per decision, written when you make it:

DATE:
ELEMENT: [What you're modeling/deciding]
DECISION: [What you chose to do]
EVIDENCE: [What sources informed this]
ALTERNATIVES: [What else you considered]
RATIONALE: [Why you chose this option]
CONFIDENCE: [High/Medium/Low]
NOTES: [Uncertainties, assumptions]

Example entries:

2024-05-15: Window height set at 2.1m based on measurement in Figure 3 of survey report. Alternative: 1.9m suggested in written description, but figure is more recent and detailed. Confidence: High.

2024-05-18: Door placement inferred from typical period practice; no direct evidence for this wall. Positioned based on circulation logic and comparison with similar buildings documented in Jones (1995). Confidence: Low (speculative).

2024-05-20: Traffic signal height: 5.8m verified by on-site measurement with laser distance meter. Matches specification documents for this model. Confidence: Very high.

This running log creates the transparent audit trail that professional investigation requires.

Organize Your Blender File​

Use clear naming conventions:

  • Collections named by evidence type: Verified_Elements, Inferred_From_Photos, Speculative_Context
  • Objects named descriptively: Vehicle_A_Honda_Civic, Signal_Post_Measured, Building_North_Simplified

Use custom properties: Right-click object → Custom Properties → Add

  • Source: "Site measurement 2024-05-15"
  • Confidence: "High" / "Medium" / "Low"
  • Evidence: "Survey document p.12"

Include documentation in the file:

  • Add a Text object in a dedicated collection with project notes
  • Use empty objects with names like NOTE_Camera_Calibration_Details to mark important reference points

Save development versions: Project_v01_Initial_Blocking.blend Project_v02_Camera_Matched.blend Project_v03_Analysis_Complete.blend

This versioning creates a record of how the model evolved and allows you to revisit earlier states if needed.

Visual Documentation​

Create comparison images:

  • Source photo alongside your rendered view
  • Overlay of model on source footage (at 50% opacity)
  • Before/after showing modeling stages
  • Annotated diagrams showing measurement points

Document sight lines:

  • Render from analyzed viewpoints
  • Add translucent planes or colored lines showing sight angles
  • Label key elements in the image

Show your verification:

  • Screenshot showing how you aligned reference points
  • Image with measurements labeled
  • Comparison of multiple interpretations if uncertainty exists

Write As You Go​

Don't wait until the end to document methodology. Keep a working document where you:

  • Drop in source citations as you use them
  • Paste screenshots of technical steps
  • Write quick notes about decisions
  • Record problems encountered and how you solved them

When it's time to write the formal methodology, you'll have all the raw material ready.

Practical Examples: Scaling the Framework​

The same four-component structure works at different scales. Here's how methodology looks for different project types:

Example: Incident Reconstruction from Video​

Research Question: "What was the approximate speed of Vehicle A between intersection entry and collision point?"

Approach: Confirmatory, calculating measurable speed to test claims.

Sources:

  • CCTV footage: 1920x1080, 25fps, specific timestamp range
  • File hash for integrity: a3f5d8c9e42b1a7f6d8c3e9b2a1f4d7c (SHA-256)
  • Obtained from: City Traffic Management, April 2, 2024
  • Chain of custody: Original SD card → department server → forensic copy → project import
  • On-site measurements: intersection width (12.4m), camera height (8.2m)
  • Vehicle specifications: 2019 Honda Civic (4.63m length)
  • Weather data: clear conditions verified

Method: "Camera matched using fSpy with verified reference points (building edges, parking meter at known position). Vehicle tracked frame-by-frame (keyframes every 5 frames = 0.2 second intervals). Position matched using overlay method. Distance measured in Blender: 28.4m traveled in 6.0 seconds = 4.73 m/s (17.0 km/h). Margin of error: ±0.3 m/s based on positioning uncertainty of ±20cm."

Limitations: "Footage resolution creates ±20cm positioning uncertainty, yielding ±1.1 km/h speed uncertainty. Between-frame positions interpolated (25fps = 0.04 second intervals). Verification: calculated speed consistent with observed motion; no physics violations detected."

The Stage 0 Gate​

Pass this list before you start modeling. It is the exit criterion for this stage:

  • Target tier (0–3) chosen and written down, with reason
  • Research question written as one defined factual sentence — or an explicit Tier 0 (no factual claim)
  • Approach defined (confirmatory/exploratory/mixed)
  • Disclosure decision made: which techniques will be named in the report, including generative ones if any
  • Source list started
  • Decision log file created (format in Decision Log)
  • Blender file organized with named collections
  • Version control plan (file naming scheme)
  • Documentation folder structure created

Summary​

A methodologically sound investigation with acknowledged limitations beats a polished visualization with undocumented choices. Methodology — not rendering skill — is what turns illustration into defensible evidence: define the question before opening Blender, work the four components (approach, sources, process, limitations) at the scale your tier demands, and document as you go. The failures that sink otherwise sound work repeat in every project: claiming certainty you don't have, writing methodology after the fact, mistaking a beautiful render for verification, and ignoring what didn't work. When in doubt, document more rather than less.

Resources​

  • Research Gate — the checklist of stages this module implements
  • Collect — next stage: acquiring and documenting your sources

Sources​

  1. Berkeley Protocol on Digital Open Source Investigations (OHCHR)
  2. The Future of Digital Evidence Authentication at the International Criminal Court (JPIA – Princeton)
  3. Forensic Architecture / Forensis — Methodology Report: Preliminary Investigation, Potential Escape from the Arena Bar, Hanau (2021)
  4. Bellingcat & GLAN — Justice and Accountability Unit: Methodology for Online Open Source Investigations into Ukraine (2022) (PDF: JA Manual for Publication)
  5. Amnesty International — Forensic Methodology Report: How to catch NSO Group's Pegasus (2021)
  6. Bellingcat — How we Geolocated a Photo of a Russian Missile Programming Team (2022)
  7. Agosto E, Ajmar A, Boccardo P, Lingua A — Crime Scene Reconstruction Using a Fully Geomatic Approach (Sensors, 2008)
  8. Accuracy of three-dimensional Gaussian Splatting for virtual crime scene reconstruction (Frontiers in Computer Science, 2026)
  9. Villa C, Lynnerup N, Jacobsen C — A Virtual, 3D Multimodal Approach to Victim and Crime Scene Reconstruction (Diagnostics, 2023, 13(17), 2764)
  10. SWGDE — Requirements for Report Writing in Digital and Multimedia Forensics v1.0 (November 20, 2018)
  11. SWGDE — Best Practices for Digital Forensic Video Analysis v2.0 (December 10, 2025)
  12. Maksymowicz K et al. — Crime event 3D reconstruction based on incomplete or fragmentary evidence material, case report (Forensic Science International, 2014, 242, e6–e11)
  13. Rangelov et al. — Impact of Data Capture Methods on 3D Reconstruction with Gaussian Splatting (J. Imaging, 2025, 11(2), 65)
  14. Rangelov et al. — Impact of Camera Settings on 3D Reconstruction Quality: Insights from NeRF and Gaussian Splatting (Sensors, 2024, 24(23), 7594)
  15. Human Rights Watch — Our Research Methodology
  16. Tracking people and cars using 3D modeling and CCTV (Forensic Science International, 2010)