diff --git a/docs/SUMMARY.md b/docs/SUMMARY.md
index 45e724747..f091cb82a 100644
--- a/docs/SUMMARY.md
+++ b/docs/SUMMARY.md
@@ -10,6 +10,7 @@
   - [Input Files](file_formats/input_files.md)
   - [Output Files](file_formats/output_files.md)
 - [Model Description](model/README.md)
+  - [Time](model/time.md)
   - [Units and Dimensions](model/units_and_dimensions.md)
   - [Dispatch Optimisation](model/dispatch_optimisation.md)
   - [Investment Appraisal](model/investment.md)
diff --git a/docs/model/README.md b/docs/model/README.md
index 01dcf8c9c..2a68b9f08 100644
--- a/docs/model/README.md
+++ b/docs/model/README.md
@@ -48,6 +48,8 @@ At a high level, the user defines:
    rules. Portions of demand for each commodity must be assigned to an agent, and the sum of these
    portions must be one.
 
+The temporal arrangements described above are explained in more detail in [Time](time.md).
+
 ## Framework Overview
 
 The model operates sequentially across a series of milestone years (MSYs). For the base year,
diff --git a/docs/model/time.md b/docs/model/time.md
new file mode 100644
index 000000000..91faa78ed
--- /dev/null
+++ b/docs/model/time.md
@@ -0,0 +1,90 @@
+# Time
+
+MUSE2 represents time in two related ways: over the long-term model horizon and within each
+milestone year.
+
+```text
+Time horizon
+└── Milestone years
+      └── Seasons
+          └── Time slices
+```
+
+Milestone years describe when the model makes long-term investment decisions and records results.
+Time slices describe how operation is represented within each of those years.
+
+## Time horizon and milestone years
+
+The time horizon is the overall period covered by a model run. It is represented by the ordered list
+of `milestone_years` in `model.toml`, for example:
+
+```toml
+milestone_years = [2020, 2030, 2040]
+```
+
+Milestone years must be positive, sorted and unique. MUSE2 evaluates the system at each milestone
+year. The first milestone year is the base year: existing assets and base-year demand describe the
+system being calibrated. Subsequent milestone years are investment and reporting points, at which
+assets, demand, dispatch and prices are evaluated.
+
+Milestone years need not cover every calendar year in the horizon. Use additional milestone years
+when technology, demand or policy changes need to be represented at a finer long-term resolution.
+
+## Seasons and time slices
+
+Each milestone year is divided into user-defined seasons and times of day. A *time slice* combines
+one season and one time of day, and is written as `season.time_of_day`, for example `winter.day` or
+`summer.peak`.
+
+The names are labels supplied by the modeller. MUSE2 does not attach a built-in meaning to names
+such as `winter`, `day` or `peak`.
+
+Time slices are defined in `time_slices.csv`, for example:
+
+```csv
+season,time_of_day,fraction
+winter,night,0.25
+winter,day,0.25
+summer,night,0.25
+summer,day,0.25
+```
+
+The `fraction` is the share of the year represented by a time slice. Every fraction must be
+positive, and the fractions for all slices must sum to one. If `time_slices.csv` is omitted, MUSE2
+uses one `all-year.all-day` slice covering the whole year.
+
+The input-file reference contains the complete [time-slice format](../file_formats/input_files.md#time-slices).
+
+## Time-slice selections
+
+A time-slice selection is the period over which MUSE2 applies a balance or limit. It can cover the
+whole year (`annual`), a season (`winter`), or one time slice (`winter.day`). The granularity of a
+selection comes from the thing it applies to: commodity balances use the commodity's
+`time_slice_level`, while explicitly provided constraints use the level of their `time_slice`.
+These two granularities are independent.
+
+- **SED balance:** Supply and demand for an SED commodity are balanced at the commodity's
+  `time_slice_level`. `annual` creates one balance for the whole year, `season` creates one balance
+  per season, and `daynight` creates one balance per time slice.
+- **SVD demand:** Annual service demand is distributed using `demand_slicing.csv`. Demand slicing
+  data should be provided at the granularity of the SVD commodity's `time_slice_level`.
+- **Commodity constraints:** A production or consumption limit uses the level of its own
+  `time_slice` entry, which may differ from the commodity's balance level. For example, an `annual`
+  commodity can have a production limit that applies only to `winter`.
+- **Availability constraints:** A process activity limit uses the level of its own `time_slice`
+  entry. It can apply to one time slice, a season, or the whole year, independently of the balance
+  level of the commodities produced or consumed by the process.
+
+## Choosing a resolution
+
+- **Define the time slices for the most detailed requirement:** The shared time-slice definition
+  must support the most detailed commodity balance or operational constraint in the model.
+- **Choose each commodity's balance level separately:** Set `time_slice_level` to `annual`, `season`,
+  or `daynight` according to the resolution needed for that commodity. Demand-slicing data should
+  use the same granularity.
+- **Keep explicit constraints in mind:** Availability limits and commodity constraints specify their
+  own time-slice selection, which can be more detailed than the balance level of the commodities
+  involved.
+
+More time slices provide more detail but increase the size of the optimisation problem, so use only
+the temporal resolution needed by the model.