Before Calling It Extraordinary: The Measurement Problem Behind a UAP Video

A small object crosses a screen. Someone estimates its speed, another person proposes an ordinary explanation, and a third argues that the explanation cannot be right because the object appears to turn too quickly. Before any of those claims can be evaluated, there is a less glamorous question: what, exactly, did the recording measure?

A video can show a changing image position without providing the object’s distance. It can preserve a sequence of pictures without preserving reliable capture timing. It can be authentic as a file while being misinterpreted as a measurement. These distinctions are not arguments that every unidentified anomalous phenomenon has already been explained. They are the conditions needed to know which claims an observation can support.

The NASA independent study team’s September 2023 report identified missing sensor metadata, weak calibration, inadequate baseline data and a shortage of multiple measurements as obstacles to studying UAP. It proposed a better data-collection approach rather than a universal explanation for individual sightings. That diagnosis is a useful starting point for a public reader: ask what information would turn an intriguing recording into a testable physical claim. NASA UAP Independent Study Team Report.

This article works through that question with explicit calculations and fictional examples. None of the example cameras, trajectories, files or observations represents a real sighting. The purpose is to show how conclusions change when distance, timing, geometry or processing assumptions change. It is not a claim to have identified a particular object, and it does not estimate the probability of extraterrestrial life.

1. An unidentified observation is not a measured performance envelope

The first task is to separate three statements that often travel together: something was observed; the observer could not identify it; and the object performed a particular maneuver at a particular speed. The first can be supported by a record of an observation. The second describes an identification result. The third requires measurements and a model connecting those measurements to motion.

For a fictional clip, write the claims in separate rows. “A bright region shifts rightward across the image” is a description of pixels. “An object moved east” adds a coordinate frame. “The object traveled at three hundred meters per second” adds a distance scale and elapsed time. “It accelerated beyond a known aircraft’s capability” adds a trajectory estimate, uncertainty and a comparison standard. Each row needs more than the row before it.

This separation protects witnesses as well as readers. A person can report an unfamiliar appearance accurately without having enough information to estimate range. A pilot’s account may be important to an investigation without turning every visual impression into a calibrated measurement. Treating uncertainty as dishonesty is as unwarranted as treating confidence as a substitute for missing data.

A useful observation record preserves the witness’s description in its own terms and keeps later analysis separate. If an analyst converts “it seemed very fast” into a numerical estimate, the record should show how that conversion was made. If no defensible conversion is available, the qualitative description should remain qualitative.

The same discipline applies to proposed explanations. Calling a shape a balloon, aircraft or optical artifact is an identification claim, not a measurement. It should be supported by the relevant evidence rather than offered as a reflexive answer. The aim is to make competing explanations testable, not to choose which side receives the benefit of missing information.

2. Calibration means a documented relationship, not an impressive camera

NIST describes metrological traceability as a property of a measurement result connected to a reference through a documented calibration chain, with uncertainty accounted for along that chain. It also emphasizes that traceability alone does not guarantee fitness for a particular purpose. A result can be traceable yet insufficiently precise for the question being asked. NIST: metrological traceability.

Consider a fictional camera called C1. Its brochure advertises high resolution, and a technician has checked that it records recognizable images. Neither fact, by itself, supplies the relationship between a pixel coordinate and a viewing direction. For that, the analyst needs the relevant camera geometry and the state in which the camera was used.

A filled measurement record might look like this:

The important feature is not the camera’s invented specification. It is the difference between a known value, an unknown value and a value whose meaning has not been validated. A blank field should not inherit a plausible default merely because a calculation requires an input.

An investigation can still proceed with partial information. It might estimate image-plane motion, bound a range of possible geometries or identify a question that another sensor could answer. What it should not do is conceal the missing inputs inside a polished animation. A reconstruction is only as informative as the evidence and assumptions used to construct it.

3. Work the angular-speed problem before making a speed claim

Suppose, in an original fictional example, a calibrated observation establishes an angular rate of one degree per second. Assume the observer is stationary, the motion is locally transverse to the line of sight, and the object’s range is approximately constant over the short interval being considered. Under those assumptions, transverse speed is approximately range multiplied by angular rate expressed in radians per second.

One degree is approximately 0.017453 radians. The same measured angular rate therefore produces three very different conditional speeds:

These are calculations, not observations. The table deliberately changes only the assumed range to expose how strongly the result depends on it. Without an independent range estimate, the angular observation does not select one row as the correct physical speed.

There is a common circular argument to avoid. An analyst may assume the object is large, infer that it must be distant, and then use that distance to infer a remarkable speed. The speed can then be presented as evidence that the object is an extraordinary vehicle. But the extraordinary interpretation helped determine the distance in the first place. The calculation has not independently established what the argument claims.

A transparent reconstruction would instead display the assumed range and show how the answer changes across plausible values. If a separate measurement constrains range, include that measurement and its uncertainty. If the constraint comes from an identification assumption, label it as such. This allows readers to distinguish a genuine measurement from a conditional scenario.

The formula also has boundaries. It is a local relation under the stated geometry, not a complete three-dimensional trajectory solution. Motion toward or away from the observer, observer movement, changing camera orientation and significant range changes require additional treatment. The worked example is useful because its assumptions are visible, not because it can be pasted onto every video.

4. Apparent size has the same missing-distance problem

A second fictional observation measures an angular width of one tenth of a degree. In a simple geometry with an object facing the observer, its physical width is twice the range multiplied by the tangent of half the angular width. At five hundred meters, that works out to approximately 0.87 meters. At five thousand meters, it becomes approximately 8.73 meters.

The tenfold change is not a camera malfunction. It is the ordinary consequence of looking at different physical scales through the same angular opening. A small nearby object and a larger distant object can occupy the same image width. A familiar-looking silhouette may suggest a hypothesis, but it does not remove the geometric ambiguity without additional evidence.

Now add an image-processing complication. Suppose the measured bright patch spans four pixels, but the optical response spreads light from a smaller source across those pixels. Measuring the patch’s width is not necessarily measuring a solid object’s edge. A claim about physical dimensions needs a reason to interpret the visible boundary as the boundary of the object rather than the boundary of the recorded response.

USGS guidance on camera calibration describes the importance of focal length, principal-point location and lens-distortion parameters in relating images to geometry. The guidance concerns imagery calibration, not UAP identification, but the underlying point is directly relevant: the mapping from a recorded image to a physical ray is something to characterize, not assume. USGS: guidelines for calibration of uncrewed-aircraft imagery.

For a public explanation, show the original pixel footprint alongside any enlargement. State whether the analysis measures a resolved outline, a light distribution or a manually selected region. Those are different quantities. A caption that simply says “object width” can hide the difference and give a more definite impression than the measurement supports.

A useful result might be an angular extent with a stated method, while physical dimensions remain unresolved. That is still information. It is better than selecting a convenient distance and presenting the resulting number without its dependence on that choice.

5. Two cameras help only when the geometry and timing cooperate

NASA’s explanation of stellar parallax illustrates a basic geometric principle: an apparent positional shift observed from a known baseline can be used to estimate distance. Applying related geometry to a moving object near Earth requires the relevant viewing positions and directions to be known at the relevant times. The astronomical example supplies the principle, not a ready-made solution for an arbitrary pair of videos. NASA: stellar parallax.

Imagine two fictional stationary cameras separated by one hundred meters. Put the object directly in front of the midpoint between them, with all three locations in a flat horizontal plane. If the object is one thousand meters from the baseline’s midpoint, the two viewing directions converge by approximately 5.725 degrees. At ten thousand meters, the convergence is only about 0.573 degrees.

For this deliberately symmetric geometry, the perpendicular distance equals half the baseline divided by the tangent of half the convergence angle. That relationship allows a second calculation. If the measured convergence is around six tenths of a degree, small angle changes produce substantial distance changes:

This table is a sensitivity illustration, not a confidence interval. It shows why a second camera is not automatically decisive. The precision of the angle measurements matters, especially when the viewing rays are nearly parallel. The geometry of the baseline matters too; the symmetric formula cannot be used unchanged when the object is elsewhere.

The two images must also depict the same object at the intended time. If one camera’s selected frame is delayed, the viewing rays may point toward different positions along a moving object’s path. Their apparent intersection can then be misleading. A pair of videos is an opportunity for triangulation, not proof that triangulation has been successfully performed.

A strong report would publish the camera positions, coordinate convention, frame-selection method, timing alignment and uncertainty treatment needed to reproduce its calculation. Where some information cannot be released, the public conclusion should acknowledge which parts outsiders cannot independently check. An attractive three-dimensional animation does not replace those inputs.

6. Observer motion can turn a quiet object into a dramatic screen crossing

AARO’s 2024 information paper on forced perspective and parallax states that these effects can explain some reports of apparently large objects or high speeds, while explicitly declining to present a single explanation for every UAP case. That is a bounded methodological claim, not a verdict on every unfamiliar recording. AARO: forced perspective and parallax.

To see the issue without relying on a disputed clip, imagine a fictional observer moving sideways past two stationary markers. One marker is nearby and one is far away. The observer’s view of the nearby marker changes more rapidly. If a camera keeps the nearby marker centered, the distant background appears to sweep across the frame. Nothing in the setup requires the nearby marker to move.

Now imagine receiving only the centered video, without the observer’s path or the camera’s pointing history. A viewer might describe the central feature as racing over the background. That description concerns relative image motion. It cannot, by itself, determine how much came from the observer, the camera or the feature.

The appropriate analysis uses a coordinate frame and asks which motion is being estimated. Image coordinates describe where a feature falls on the sensor. Camera-relative coordinates describe its direction relative to the instrument. World coordinates require information about the instrument’s position and orientation. Moving between these frames is a calculation with inputs, not a change of vocabulary.

In a fictional reconstruction, first hold the object stationary and move the observer according to the known path. Then compare the predicted image motion with the recorded motion. A good match would show that observer motion can account for that feature of the clip under the model. It would not, by itself, identify the object. A poor match would challenge that particular reconstruction, not prove every extraordinary alternative.

This distinction prevents two opposite mistakes: interpreting all apparent motion as object motion, and treating the mere existence of parallax as sufficient to explain a case. The effect must be calculated for the actual geometry when the necessary information is available.

7. Frame count is not a substitute for elapsed time

The Scientific Working Group on Digital Evidence’s frame-timing guidance distinguishes overall frame rate from the interval between specific frames. It warns that proprietary formats can be misread by playback software and places limits on timing analysis of transcoded or screen-captured material. Those cautions matter whenever a speed calculation uses a video’s apparent playback rate. SWGDE: frame-timing analysis.

Here is a fictional calculation with a known displacement of twelve meters. An analyst assumes that twelve frame intervals represent four tenths of a second and reports thirty meters per second. Later, a reliable timing record shows that the selected endpoints are eight tenths of a second apart. The same displacement then gives fifteen meters per second.

The example does not assert that every video has incorrect timing. It demonstrates that distance and time deserve separate verification. A correct distance measurement divided by an incorrect interval is still an incorrect speed estimate. The number of displayed frames alone cannot resolve a disagreement about how the file was recorded, processed or presented.

Keep three clocks distinct. Capture time concerns when the sensor acquired information. File timestamps or timing tables describe how the recording represents time. Playback time concerns what a viewer’s software displays. Those clocks may correspond closely in a well-characterized system, but the correspondence is an empirical question when material has passed through several export and editing steps.

For the fictional case, a defensible worksheet would identify the two endpoint frames, preserve their original timing information and state which interval was used. If only a screen recording survives, the analyst should describe that limitation rather than silently substitute the screen recorder’s settings for the source camera’s capture behavior.

A timeline also needs a clear origin. An elapsed time of three seconds from the start of a file is not a universal time that can automatically align with another camera. Synchronizing separate records requires evidence about their clocks or a shared event that can be matched with an appropriate uncertainty. A display overlay is a clue to investigate, not an automatic guarantee.

A speed spike can also be an identity switch

Timing is not the only hidden assumption in a track. The analysis must also establish that the positions being compared belong to the same feature. Consider a separate, entirely synthetic example with two points in a one-dimensional image. At the first frame, point A is at pixel ten and point B is at pixel eighty. At the next frame, A is at twelve and B is at seventy-eight. At the third, A is at fourteen and B is at seventy-six.

Both points move two pixels per interval in opposite directions. Suppose the selected track follows A from ten to twelve, then accidentally switches to B at seventy-six. The final step appears to cover sixty-four pixels. If the known synthetic frame interval is one thirtieth of a second, the correct track for A implies sixty pixels per second, while the switched step implies nineteen hundred and twenty pixels per second. Neither number is a physical speed in meters per second because no distance scale has been supplied.

The apparent jump is a property of the association error, not of either point’s generated motion. This example does not establish that a real UAP video contains such an error. It shows why a reported acceleration should be accompanied by evidence about how the same target was followed across frames, especially where features overlap, fade or become difficult to distinguish.

A review can expose the issue by displaying the selected position on each original frame and marking intervals where identity is uncertain. Where two associations remain plausible, carry both possibilities into the analysis rather than choosing the one that creates the smoothest or most dramatic path. If manual intervention changes a track, preserve that intervention in the processing record.

The useful distinction is between a noisy position estimate and uncertainty about which object was measured. Small coordinate error bars around the wrong feature do not solve the second problem. An analysis that reports precise positions while hiding ambiguous associations can look mathematically careful and still answer the wrong question. The synthetic test therefore adds a specific check to the workflow: confirm identity continuity before interpreting a discontinuity as physical motion.

8. Put uncertainty into the worked result, not just the disclaimer

NIST’s guidance on propagation of uncertainty treats a measured output as a function of input quantities and their uncertainties, including relevant corrections and correlations. It is not enough to attach a generic “may be inaccurate” sentence after a numerical result. The important question is how uncertain inputs affect the quantity being claimed. NIST Technical Note 1297, Appendix A.

Consider another original hypothetical: a displacement estimate of 120 meters with a standard uncertainty of four meters, divided by a time interval of two seconds with a standard uncertainty of one tenth of a second. The nominal speed is sixty meters per second. If the two input uncertainties are independent and a first-order approximation is suitable, the relative standard uncertainties combine in quadrature.

The distance contribution is four divided by 120, or about 3.33 percent. The timing contribution is 0.1 divided by two, or five percent. Squaring those fractions, adding them and taking the square root gives about 6.01 percent. Multiplying by the nominal speed gives a standard uncertainty of about 3.61 meters per second.

That calculation does not automatically produce a ninety-five-percent interval. It also does not cover an unmodeled error such as using the wrong object range or matching the wrong frames. The stated uncertainty is conditional on the measurement model and its assumptions. A narrow-looking numerical band can be misleading if a large source of uncertainty was omitted entirely.

The example suggests a practical priority. Improving the distance estimate while ignoring the larger timing contribution may produce less benefit than expected. A sensitivity calculation can show which input deserves additional measurement. That turns uncertainty analysis into a planning tool rather than a ceremonial warning.

For a real claim of exceptional acceleration, uncertainty must also be carried through the changes in velocity over time. Differencing estimates can magnify the influence of noisy measurements. A graph with sharp peaks is not self-authenticating evidence of sharp physical maneuvers. The analyst needs to show the measurement process, the smoothing or fitting choices and the effect of reasonable alternatives before the peaks are interpreted as performance.

9. An infrared image is not a direct picture of surface temperature

NIST’s thermal-imaging discussion in a fire-research report explains that radiation reaching a sensor can include emitted, reflected and transmitted contributions, and that material properties and wavelength matter. This is a general measurement principle illustrated in a different application; it does not identify the cause of any UAP image. NISTIR 7094, infrared-camera discussion.

A manufacturer’s current thermal-camera guidance similarly notes that measurement accuracy depends on factors including distance, ambient conditions and emissivity. Those factors cannot be assumed known for an unidentified object merely because the displayed image uses a temperature-like color scale. FLIR: using a thermal camera.

Imagine a fictional exported infrared clip that shows a dark patch against a lighter background. Before saying the patch is “cold,” the analyst should establish what the display mapping means. Does darker indicate lower recorded radiance in that rendering? Was the mapping fixed or adjusted? Does the file preserve quantitative sensor values, or only a presentation image? Without those answers, a description of relative displayed brightness is safer than an absolute temperature claim.

A second issue is how much of the sensor’s view the target fills. If a recorded region combines signal from the target and its background, interpreting it as the target’s isolated surface condition requires additional assumptions. A tiny patch is not automatically a resolved thermographic measurement of a solid object.

This is why visible and infrared images can complement one another without serving as interchangeable photographs. They record different aspects of a scene through different instruments and processing chains. Agreement about the presence of a feature can be valuable, while disagreement in appearance may require an explanation of the measurement systems rather than a claim that the object changed shape.

A careful public caption could say that a feature has a specified contrast in a specified infrared rendering, with absolute temperature not determined from the available file. That is more informative than calling it either “ice cold” or “white hot” without a defensible conversion. The point is not to forbid interpretation; it is to require the conversion from display appearance to physical quantity to be shown.

10. Enlargement can clarify pixels without discovering new edges

SWGDE’s revised guidance on resizing imagery describes how interpolation changes the displayed representation and can affect apparent edges. It distinguishes making an image easier to inspect from introducing detail absent from the original recording. Its discussion also cautions about the reproducibility of novel machine-learning interpolation methods. SWGDE: fundamentals of resizing imagery.

In a fictional example, the target occupies a four-by-four-pixel patch. Enlarging it tenfold in each dimension creates a forty-by-forty display patch. The display now contains sixteen hundred positions, but the original patch supplied only sixteen pixel values. The expansion has changed presentation, not the number of independent observations originally captured.

Different interpolation methods can make the same patch look blockier, smoother or more sharply outlined. A viewer may find one result more recognizable. Recognition is not a measurement of newly recovered physical structure. If a small feature appears only after a particular enhancement, the analysis should show the original and identify the operation responsible for the new appearance.

A useful comparison panel would contain the native-scale region, a simple enlargement and any processed version, all labeled with their transformations. The scale should be stated rather than allowed to change with the viewer’s screen. If brightness or contrast has been adjusted, disclose the settings and preserve the unadjusted image in the evidence package.

The same principle applies to an animation reconstructed from estimated positions. It may help explain a hypothesis, but its smooth trajectory and three-dimensional object model are authored presentation choices unless independently measured. A rendering should not be introduced as if it were another camera view.

For a reader, the key question is whether the conclusion survives reasonable processing choices. If an asserted edge or turn vanishes when a different defensible method is used, that instability belongs in the result. A picture should not become more certain merely because the processing made it more persuasive.

11. Authenticate the file without overclaiming the event

SWGDE’s video-authentication guidance addresses whether a file’s provenance, content and structure align with what is claimed about it. It distinguishes examining a digital recording from establishing that everything depicted or asserted about its context is true. File authenticity and event interpretation are related but different questions. SWGDE: digital-video authentication.

The group’s updated forensic-video-analysis guidance recommends working copies, integrity checks and reproducible documentation of processing. It also calls for communicating limitations when material is not supplied in its native format. Those practices help preserve what an analyst actually received and did; they are not a certificate that the analyst’s physical explanation is correct. SWGDE: digital forensic video analysis.

Suppose a fictional investigator receives a clip by email. Computing a hash establishes an identifier for those received bytes. Recomputing the same hash later can help show that the working file has not changed. It does not establish who originally recorded the scene, whether the file had been edited before receipt, or whether the accompanying caption identifies the correct date and place.

The acquisition record should therefore distinguish the received file, the stated source and any independently verified context. If an earlier version becomes available, compare it rather than silently replacing the first copy. Differences may reflect an ordinary export, a crop, added labels or something more significant. The analysis should identify the difference before assigning a motive to it.

This produces a useful two-column conclusion: what has been established about the file, and what has been established about the depicted event. The first column might be strong while the second remains incomplete. That is not a contradiction. It is a map of the work still needed.

12. A complete fictional worksheet: the camera was turning too

Bring the pieces together in a deliberately simplified fictional case. The observer remains stationary. The target stays near the center of the camera’s view, and the analyst works only with a short interval in a horizontal plane. A separate, validated range measurement places the target at approximately one thousand meters for that interval. These assumptions are part of the invented exercise, not facts about any public UAP clip.

The camera is rotating to the right. Its orientation log and image measurements provide the following values, all expressed in a compatible angular coordinate system:

The rightmost column adds the camera orientation and the target’s camera-relative angle in this simplified planar setup. The target’s world bearing changes by one degree, not four tenths of a degree. At the assumed range, the local transverse-speed estimate is about 17.45 meters per second. Using image motion alone would give about 6.98 meters per second.

The mistake in this example would underestimate speed. Measurement errors do not always make a sighting look more dramatic; they can move a result in either direction. That is one reason to analyze the geometry rather than adopt “ordinary explanation” as a predetermined numerical answer.

Now remove one input at a time. Without the orientation log, the world-bearing change is unresolved. Without the range measurement, the physical-speed estimate is conditional. Without compatible timestamps, the two angular readings may not describe the same interval. The worksheet makes each dependency visible instead of giving a single answer that survives even after its supporting inputs disappear.

A real analysis would need to extend this simple model to the actual geometry, including three-dimensional orientation, observer motion and changes in range where relevant. It would also evaluate uncertainties and correlations. The exercise is intentionally small enough that a reader can check the arithmetic; it is not a universal reconstruction algorithm.

The publication-ready conclusion within the fictional case would be limited: under the specified geometry, synchronized readings and range assumption, the calculated local transverse speed is approximately 17.45 meters per second. It would not be appropriate to identify an aircraft type, infer an origin or announce an extraordinary capability from that calculation alone. Those are additional questions requiring additional evidence.

13. Test the processing chain with inputs whose answers are known

A useful way to evaluate an analysis pipeline is to give it controlled inputs before relying on it for an unfamiliar observation. The following is an original proposed test exercise. It does not claim that any agency or laboratory has run these tests or achieved these results.

Construct a synthetic sequence with a marker moving exactly two image pixels per frame at known, regular time intervals. This is a software-generated test pattern, not a fabricated sighting to be presented as real. Run the intended extraction and tracking steps on that sequence. The pipeline should recover the known image displacement within its stated tolerance. If it does not, the discrepancy needs to be understood before physical interpretations are attached to an actual clip.

Next, add a known camera pan to the synthetic scene while retaining the same world trajectory. A pipeline intended to reconstruct world motion should distinguish the two contributions when supplied with the correct orientation data. Repeat the test with orientation data deliberately omitted and verify that the output becomes appropriately limited rather than silently assuming a stationary camera.

Then vary the timing. Create a sequence with a known longer interval between two frames. The analysis should not interpret the larger image displacement over that longer interval as an unexplained burst of speed merely because it assumes equal spacing. A good test report records what the software did, not just whether the final graph looked smooth.

Finally, change presentation without changing the underlying trajectory. Make a separately labeled resized or recompressed test copy and compare the recovered measurements. This can reveal which outputs are sensitive to a particular transformation. It does not establish that the same transformation occurred in a real recording; it establishes a limitation or behavior of the analytical method.

A compact proposed acceptance table would look like this:

Tolerances should be chosen for the intended measurement task, not adjusted after the test to guarantee a pass. A pipeline can pass these elementary checks and still fail on a more difficult real scene. Passing is evidence about the tested behaviors, not certification of every possible output. Failing, however, provides a concrete reason to investigate the method before treating its most dramatic result as a discovery.

14. Ask whether a second sensor adds an independent constraint

Two instruments are often more useful than one, but the important word is not “two.” It is the relationship between their measurements. A second view can constrain geometry. A separate timing reference can constrain synchronization. A range measurement can resolve an ambiguity that an image alone leaves open. Two displays derived from the same underlying stream may add little independent information.

In an original fictional review, label the dependencies explicitly. Camera C1 supplies image coordinates. Its orientation log supplies pointing direction. Range instrument R1 supplies a distance estimate. A second video, C2, is a screen capture of C1’s monitor. C2 may help document what an operator saw, but it is not a second viewing location for triangulation. Calling this a “two-camera confirmation” would conceal the dependency.

Now replace C2 with an independently positioned camera whose orientation and timing are characterized. That changes the evidence because it introduces a different line of sight. But the analyst must still determine whether both cameras recorded the same target, whether their timestamps align and whether their error sources are independent enough for the intended calculation.

A shared clock is another example of a useful feature that can also create a shared vulnerability. If both systems inherit the same incorrect time offset, their mutual agreement does not prove agreement with an external event time. If both use the same flawed calibration assumption, repeating the analysis twice does not remove that flaw. Independence is a property to examine, not a label conferred by having two screens.

The practical question is what ambiguity the additional record resolves. If the answer is “none; it is another rendering of the same information,” the record may still be useful for presentation or provenance, but it should not be counted as a new physical measurement. If it constrains distance, direction or timing, explain that contribution precisely.

15. The strongest conclusion may be a well-defined limit

An investigation does not fail merely because an object remains unidentified. It fails when it claims a measurement that its inputs cannot support or treats an unresolved identity as proof of a preferred explanation. The worked examples here show several different outcomes: a conditional speed table, a geometry-sensitive distance, a corrected timing calculation and a motion estimate that changes when camera rotation is included.

Those are substantive results. They tell a reader which conclusions are stable and which depend on missing information. They also identify where additional evidence would be valuable: an original file, a validated interval, an orientation record, a separate range measurement or a genuinely independent view.

The public should not have to choose between ridicule and credulity to ask about UAP. A measurement-centered approach takes observations seriously enough to examine them and conclusions seriously enough to demand support. The interesting question is not how confidently someone can describe a moving dot. It is whether the distance, time, geometry and processing behind that description can survive a check.