Results Index
Alongside measure-specific text outputs, PILOTS writes a machine-readable
results.json index that captures:
input fingerprints (config/input/topology)
model scale and mapping spec hash
group and topo_group audits (including dynamic size stats)
topology sections loaded
measure metadata and profiling
This is intended to make experiments reproducible and auditable.
Schema version
The schema is versioned; see schema_version in the file.
Dynamic selections
When a selection depends on a dynamic group/topo_group, PILOTS records
size_min/size_max/size_mean for the selection spec.
Audits and quality control (M0)
PILOTS always writes a small set of run-level audits that make it easy to sanity-check a run and keep results reproducible.
These audits are produced even if you only compute a single measure.
Run-level frame stats
run.frame_stats is computed over frames that were actually processed by the
Runner (i.e. after resume_from, it refers to the resumed segment).
Fields:
frames: number of processed frames.timestep_first/timestep_last: first/last processed timestep.timestep_delta_mode: the most common timestep stride between adjacent processed frames (often the dump interval).irregular_stride_count: how many adjacent timestep deltas differ from the mode. This can indicate non-uniform output sampling or gaps, but it is not a guarantee of missing data.natoms_min/natoms_max: a quick check for atom loss/growth.box_changes: how many processed frames have a box different from the first processed frame (common for NPT).
Selections and topology audits
groups.atom_groupsandgroups.topo_groupsrecord each group expression and, for dynamic groups,size_min/size_max/size_mean.groups.combined_selectionsrecords audit stats for combined selections (AtomGroup + TopoGroup + boolean combine).run.topologyincludesloaded_sectionsand thebond_graphdefinition used to build adjacency.
Mapping audit
When mapping is enabled, mapping.spec_hash_fnv1a64 is recorded so that a
checkpoint/resume cannot silently change bead identity.