# `cube_trend()`
**Category:** Time and climatology
**Status:** experimental

## Purpose
Compute a descriptive linear temporal slope independently for every longitude,
latitude, depth, and variable cell of a historical ocean cube.

## Real signature
```r
cube_trend(
  x,
  method = "linear",
  period = NULL,
  time_unit = "year",
  min_n = 3L,
  diagnostics = FALSE
)
```

## Scientific contract
The predictor is actual elapsed time, not observation position. Date uses exact
civil days and POSIXct uses exact UTC instants, including subdaily and
fractional-second spacing. Output units may be per year (exactly 365.2425 days),
day, hour, or second. Every finite stored observation has equal weight;
irregular sampling is accepted and recorded, and explicit preaggregation is
required when equal-period weighting is desired.

Only `method = "linear"` is available. `min_n` is a per-cell finite count.
Optional diagnostics contain aligned finite counts, cell time spans, fitted
values at the analysis anchor, R2, and residual SD. No Sen slope,
Mann--Kendall test, p-value, confidence interval, change metric, breakpoint, or
regime detection is calculated.

## Input and output
Raw, temporally aggregated, difference-anomaly, z-anomaly, and `signal_noise()`
cubes retain historical time and are valid inputs. Recurrent
`cube_climatology()` pseudo-time and identifiable descendants are rejected.
The output is an in-memory `ocean_cube` with time collapsed to the
representational midpoint of the selected stored timestamp range. Other axes,
variable order, and source-unit meaning are preserved.

## Backend support
Memory: **TRUE**. NetCDF: **TRUE**. Lazy values are reduced through bounded
spatial, depth, variable, and time blocks in one scientific pass. The full lazy
source is never materialised at once; the singleton-time output is materialised
in memory.