updated the docs

Author: stefdoerr

doc/source/explanation/adaptive-sampling.md

@@ -44,6 +44,8 @@ Simulations are named with the pattern `e4s3_e2s1p0f45`. Parsed:
 - `e4s3` - generated in **epoch 4**, the **3rd simulation** of that batch.
 - `e2s1p0f45` - the starting conformation came from **epoch 2**, **simulation 1**, **piece 0**, **frame 45**.
 
+The `e<n>s<m>` prefix (epoch / spawn) is the only part that is always present; everything after the `_` records provenance and varies in shape. Epoch-1 sims descend directly from a generator rather than from another sim, so they take the form `e1s3_<generator-name>` (epoch 1, spawn 3, descended from the named generator) instead of the `e<n>s<m>p<piece>f<frame>` parent reference used from epoch 2 onward.
+
 Some MD engines split long simulations into pieces; the piece index is usually 0 and can be ignored.
 
 ## Simulation length

doc/source/how-to/adaptive-inspect-run.md

@@ -21,7 +21,7 @@ for s in sims[:5]:
           s.molfile)     # topology file(s) for this sim
 ```
 
-Each {py:class}`~htmd.simlist.Sim` records the path to its `input/` directory; the directory **name** is the parent-tracking key. For an epoch-1 sim it looks like `e1s3_<generator-name>` (epoch 1, spawn 3, descended from the named generator); for later epochs it looks like `e3s7_e1s2p0f120` (epoch 3 spawn 7, parent = the sim named `e1s2`, piece `p0`, frame `f120`). The `e<n>s<m>` prefix is the only part that's stable; everything after `_` encodes provenance and varies in shape.
+Each {py:class}`~htmd.simlist.Sim` records the path to its `input/` directory; the directory **name** is the parent-tracking key (e.g. `e3s7_e1s2p0f120`). The leading `e<n>s<m>` is the epoch / spawn, and everything after the `_` records which parent frame the sim spawned from. See {doc}`Adaptive sampling explanation <../explanation/adaptive-sampling>` ("Naming scheme") for the full breakdown of the pattern.
 
 ## Common variations
 

doc/source/how-to/msm-bootstrap-errors.md

@@ -33,7 +33,7 @@ print("std  top-3 timescales (ns):", slowest.std(axis=0))
 
 {py:meth}`~htmd.metricdata.MetricData.bootstrap(ratio)` returns a new `MetricData` containing a random `ratio` fraction of the trajectories (no replacement by default). The kept count is `int(floor(numtraj * ratio))` - on small N this rounds down (e.g. 7 trajs × 0.8 = 5 kept, not 5.6), so bootstrapping ratios are coarser than they look. Each bootstrap is independent - re-cluster + re-fit from scratch.
 
-`model.msm.timescales()` (deeptime API) returns the implied timescales in **the same units as `model.lag`** - and `markovModel(..., units="ns")` stores the lag in frames after the unit conversion, so the timescales come back in frames. Multiply by `model.data.fstep` to convert to whatever time unit you set `fstep` to (ns by convention).
+`model.msm.timescales()` returns timescales in frames, so each row above is multiplied by `model.data.fstep` to get ns - see {doc}`How to read off and interpret an ITS plot <msm-interpret-its-plot>` ("Compare lag times by direct call") for the full explanation.
 
 ## Parameters that matter