Explore before-and-after comparisons

The Lab’s comparison view answers a specific question:

Among these included observations, which before/after combinations are not outperformed on every chosen objective?

This is a method for inspecting tradeoffs. It is not a detector of surface change, a registration service, or a universal ranking of image quality.

Define the population deliberately

Selecting a row inspects it. An inclusion checkbox changes whether it contributes to exploratory pair construction. Use visible rows for analysis deliberately promotes the current view into an included population; simply filtering the table does not do so.

The Lab partitions included acquisitions against the reference interval. It constructs the Cartesian product of strictly before and strictly after observations. Acquisitions with missing or invalid time, overlap with the reference interval, or a start outside the query window are reported separately.

The population line distinguishes:

included PRE × included POST → temporal pairs → evaluable pairs → nondominated pairs

Each transition has a different meaning. Fewer evaluable pairs means required metadata was missing; fewer frontier pairs means another eligible pair dominated them.

Choose method objectives

The current exploratory engine supports five objectives, all minimized:

Objective	Pair value	Unit in the Lab
Δ incidence	Absolute difference between the two incidence values	Degrees
Δ emission	Absolute difference between the two emission values	Degrees
Δ phase	Absolute difference between the two phase values	Degrees
Before-event distance	Nominal reference time minus the before acquisition start	Days
After-event distance	After acquisition start minus the nominal reference time	Days

Resolution is retained as raw source metadata but is not offered as a Lab objective until its units are established by the adapter. The Python protocol supports worst resolution when valid metre-per-pixel values exist. The browser’s available dimensions therefore must not be described as a full implementation of all historical V1–V4 variants.

An unknown active value excludes the pair from screening. Deselecting that objective can legitimately make the pair evaluable; no value is imputed.

Read nondominance correctly

A pair is nondominated when no other included, evaluable pair is no worse on every active objective and strictly better on at least one. Exact ties survive. The frontier can contain many members because different comparisons trade temporal proximity against acquisition conditions.

For a dominated pair, the inspector shows an actual witness and the objective values supporting that relation. It does not invent a composite quality score.

Illustrative method example—not lunar data: vectors \((2, 5)\) and \((3, 7)\) under two minimization objectives show strict domination of the second by the first. Vectors \((2, 8)\) and \((3, 7)\) instead express a tradeoff. Equal vectors do not strictly dominate one another.

Separate method from view

The objective checkboxes define the screening calculation. The horizontal and vertical selectors define only the plot projection.

A point may look superior in two plotted dimensions while being inferior in another active dimension. Always inspect the full objective vector before interpreting membership. The exact pair table remains available; rows missing a selected plot-axis value can be omitted from the plot without being scientifically excluded for that reason.

Exactness and resource limits

The interactive implementation allows at most 4,000 temporal pairs and 10,000,000 dominance comparisons. These are execution guards, not sampling rules. Exceeding a guard returns a named limitation rather than a partial frontier presented as complete.

To continue, explicitly include fewer observations, narrow the query, or use the Python workflow. Do not describe a frontier over an included or partially retrieved population as archive-global.

What the analysis does not establish

Nondominance does not establish image registration, matching ground coverage, photometric comparability, a detected change, causal attribution, or scientific admission. Browse inspection may help a researcher decide what to investigate next, but it does not supply missing processing or evidence.

The browser result is labeled exploratory and uses Lab-local pair identifiers. Canonical Python pair identities include scientific context and are not replaced by those display identifiers. See Event-conditioned comparison methodology and From exploration to reproducible computation.