IAI CFVE common mistakes
IAI CFVE · Understand the traps and how to avoid them.
Treating the video file as the evidence instead of the original storage device, so the candidate copies a file off a DVR or phone and starts analysis from that copy without documenting the acquisition of the source device.
Candidates come from an IT or editing background where copying a file is routine, and the working copy looks identical to the original, so the chain-of-custody gap only surfaces when the report is challenged.
What to do instead
For every case, state the source device, its make and model, the acquisition method (native export, write-blocked imaging, or logical extraction), the hash values of both the acquired container and any derived working copies, and who handled the device at each step. Practise writing this paragraph until it is automatic before any frame is examined.
Reporting frame-accurate conclusions from a player timeline or a timecode overlay without verifying the file's actual frame rate and frame indexing, then quoting a timestamp that does not map to the frame examined.
Media players display timecode that can be rounded, offset by a start-time field, or based on a nominal rather than real frame rate, and candidates trust the on-screen number because it is convenient.
What to do instead
Determine the real frame rate and total frame count from the container and stream metadata, reconcile any start offset, and index frames yourself. Quote conclusions by frame number and by the corresponding absolute time, and state the frame rate and time base used. Where variable frame rate is present, say so and explain how timestamps were derived.
Applying enhancement, sharpening, or super-resolution filters and then presenting the processed image as if it were the original frame, without keeping or disclosing the unprocessed source.
Enhancement makes plates and faces look clearer, and candidates want to give the investigator a usable image, so the processed version becomes the exhibit by default.
What to do instead
Always retain and present the unaltered frame alongside any processed derivative, document every filter and parameter applied, and state explicitly that the enhanced image is an interpretation, not new evidence. Never draw an identification conclusion from an enhanced image alone.
Estimating a subject's height, speed, or distance from a single frame or from the video alone, without camera calibration, without accounting for perspective, and without stating assumptions.
These measurements look like simple arithmetic, and candidates answer the question asked rather than the question the method can support, so error bounds and calibration are omitted.
What to do instead
Use photogrammetric calibration with known reference points in the scene, report a measurement plus an uncertainty range, and list every assumption (lens, sensor, mounting height, ground plane). If calibration references are absent, state that a reliable measurement cannot be made rather than giving a point estimate.
Describing a codec or container as lossless or as preserving original quality, or claiming a transcode is bit-exact, without checking the actual codec, bit rate, and generation loss.
Candidates memorise that some formats are 'high quality' and apply the label generically, missing that re-encoding to H.264 at a lower bit rate destroys detail relevant to identification.
What to do instead
Identify the codec, profile, bit rate, resolution, and chroma subsampling from the stream metadata before commenting. State plainly that any re-encode is generational loss, that export settings must be documented, and that analysis should be performed on the least-processed copy available.
Failing to recognise and document the difference between the recorder's clock time and real-world time, then building a timeline of events on uncorrected DVR timestamps.
DVR clocks drift, are set to the wrong time zone, or are never synchronised, but the overlay looks authoritative and candidates accept it as ground truth.
What to do instead
Compare recorder time against an independent reference (a known event, a synchronised phone, a network time source, or the file's creation metadata), calculate the offset, and present both raw and corrected times with the correction method stated. Note daylight-saving and time-zone handling explicitly.
Rendering an opinion on whether a face or vehicle in a low-quality image matches a known person or plate, rather than limiting the conclusion to what the imagery can support.
Investigators ask directly, and candidates feel pressure to be helpful, so they drift from describing features into asserting identity without a validated comparison method.
What to do instead
Keep the conclusion at the level the image supports: describe observable features, state the limitations (resolution, angle, compression, occlusion), and use a graded conclusion scale such as 'cannot be determined', 'similar', or 'consistent with', reserving identification for comparisons that meet a documented standard.
Skipping validation of the tools and scripts used, and not verifying that a frame extraction, conversion, or hash utility produces reproducible results across runs.
Open-source and command-line tools are trusted because they are widely used, and candidates do not test them against known inputs or record versions and parameters.
What to do instead
Record exact tool names, versions, and command lines; test against a known reference file; and confirm that repeated runs produce identical output hashes. Where a tool's behaviour is uncertain, describe the uncertainty in the report rather than presenting output as definitive.
Writing a report that mixes observation, interpretation, and conclusion in one narrative, so a reader cannot tell which statements are directly observable in the imagery.
Candidates write reports as storytelling, and the analytical steps feel like clutter, so the reasoning trail is lost and the opinion appears unsupported.
What to do instead
Separate the report into acquisition, examination, analysis, and opinion sections, keep observation statements factual and frame-referenced, and tie each opinion to the specific frames and method that support it. Include limitations and a clear statement of what was not examined.
Assuming a video export from a proprietary player or a cloud service is complete, when it may omit frames, drop audio, or re-encode silently.
The export plays correctly in a player, so it appears faithful, and candidates never compare the exported frame count and duration against the source recording.
What to do instead
Compare exported duration, frame count, and resolution against the source device's own record, request native or proprietary format exports where possible, and document any discrepancy. Treat player exports as derived evidence and state their limitations in the report.