Investigative Context
Introduction
A reconstruction is only as strong as its documentation. This page is about turning a 3D model into something that can be scrutinized and defended: being honest about scale limits, accounting for distortion, quantifying uncertainty, and recording the methodology so another investigator could reproduce it. This is what separates a reconstruction that holds up from one that collapses under questioning.
Scale Limits
No reconstruction is infinitely precise. The model inherits the limits of the references used to build it. Two limits decide how accurate your scale can be:
- The scale anchor. Every dimension traces back to the scale pillar from Module 02. A reference that measures true to ±5 cm puts ±5 cm into every dimension derived from it.
- Imagery resolution. One satellite pixel covers roughly 30 to 50 cm of ground. You cannot place an edge finer than the pixel that shows it.
Report the limit with the number. "Dimensions accurate to within ±5 cm, given the door-frame reference" tells a reviewer what to expect. "Exactly 24.0 m wide" claims a precision your references do not provide.
Distortion
Distortion moves the match while the model still looks right. Three sources matter in this pipeline:
- Lens distortion in the source photo. Real lenses bend straight lines. Vanishing-point matching reads that curve as scene geometry and bends the camera to fit it.
- Satellite projection error. Satellite tiles arrive re-projected onto a map grid, so a straight ground edge can arrive slightly curved.
- The pinhole assumption. Camera matching assumes one center of projection and no lens offset. Real lenses approximate that best at the center of the frame.
Watch for one symptom: the match fits the center of the frame and drifts at the corners. When corner alignment fails while the center holds, check distortion before you rebuild the model.
Presenting Uncertainty
A reconstruction that states its limits can be checked. Make every number checkable:
- Report ranges. Give key dimensions as intervals: "23.8 to 24.2 m" carries its own accuracy.
- Label every number. Mark each value as measured (taken from a reference), derived (computed from measurements), or assumed (chosen without evidence).
- Disclose your cleanup. Vegetation removal and simplified clutter change what the model shows. Say what you removed.
This connects to Foundation's Validate & Report. In validation, you test your stated limits against the evidence again.
Documentation Requirements
Keep a record with every run. This is the minimum:
- Capture conditions: date, time, platform, weather, and what the imagery shows.
- Source provenance: where each image comes from, its resolution, and its identifier.
- Software trail: programs and versions, camera-solve parameters, and the scale reference with its source level.
- Decisions: what you cleaned up, what you assumed, and what you left out.
- Date performed: when you made the reconstruction, so reviewers can re-check it against the sources available at the time.
The Berkeley Protocol on Digital Open-Source Investigations sets the standard for this record. Use it whenever others will examine your work.
Knowledge Check
Test what you just read. A few questions, immediate feedback on every answer, and a different set each attempt.
Knowledge check
Check what you just read. Every answer gets feedback.
Summary
Every number in your reconstruction has a limit, a source, and a label. Scale ends at your weakest reference. Distortion bends the fit at the frame edges. Cleanup changes what the model shows. Report the ranges, label each value as measured, derived, or assumed, and keep the run record. A reviewer can then re-check each claim against that record.
Resources
- Validate & Report — the validation workflow
- Measurements, Scale & Dimension — scale hierarchy and limits
- Scale Estimation — the scale pillar in full
Sources
- Berkeley Protocol on Digital Open-Source Investigations — international standards for digital open-source investigations