Endogenous catalogue

Endogenous catalogue

amorem exposes 74 numeric endogenous statistics — 66 with a compiled C++ fast path, 8 in pure R — plus the sender_receivers_set list-column, organised into six families × seven variant axes. The same history can be summarised many ways; the variant axis decides which way. The simulator and the post-hoc endogenous_features() engine cover the catalogue together, and every shared statistic is parity-tested on common event logs.

---

The six variant axes, side by side

Eight events between two actors X and Y (blue ticks: Y → X, red ticks: X → Y):

Event stream

For the focal dyad X → Y, every reciprocity variant computes a different summary of this exact history:

Variant traces

Variant	Reads as	What changes at each tick
count	cumulative count of past Y → X events	+1 at each blue tick
binary	“any past Y → X?”	jumps 0 → 1 at the first blue tick, then stays
exp_decay	weighted count with half-life h	adds 1 at each blue tick, decays in between
time_recent	t − t_last (Y → X)	resets to 0 at each blue tick, climbs linearly between
time_first	t − t_first (Y → X)	monotone linear ramp once the first blue tick fires
count_interrupted	count since the most recent X → Y event	resets to 0 at each red tick
time_recent_interrupted	time_recent reset at each red tick

The *_interrupted family is the central methodological distinction of Juozaitienė & Wit (2024): a closure event “resets” the relevant counter, so the statistic measures only the most recent unclosed cycle. On Manufacturing it carries information distinct from *_time_recent (Real-data analysis).

Driver script: paper/wiki/experiments/variant_traces.R.

---

The six families

Family	Captures	Closure pattern
Per-actor	sender outdegree, receiver indegree, last-event recency	(no closure — per-actor totals)
Reciprocity	has the receiver previously fired toward the sender?	$s \to r$ cycles $r \to s$
Transitivity	closure of two-paths	$s \to k$, $k \to r$ ⇒ $s \to r$
Cyclic	closure of three-cycles	$r \to k$, $k \to s$ ⇒ $s \to r$
Sending balance	joint sending behaviour vs alters	$s \to k$, $r \to k$ co-occur
Receiving balance	joint receiving behaviour vs alters	$k \to s$, $k \to r$ co-occur

For every closure family, all six variant axes are available; some families additionally expose an ordered axis (constrained direction of two-path traversal). The full set is:

Family	Statistic names
Per-actor	sender_outdegree, receiver_indegree, recency
Reciprocity	reciprocity_binary, reciprocity_count, reciprocity_exp_decay, reciprocity_time_recent, reciprocity_time_first, reciprocity_binary_interrupted, reciprocity_count_interrupted, reciprocity_exp_decay_interrupted, reciprocity_time_recent_interrupted, reciprocity_time_first_interrupted
Transitivity	transitivity_binary, transitivity_count, transitivity_binary_ordered, transitivity_count_ordered, transitivity_exp_decay, transitivity_exp_decay_ordered, transitivity_time_recent, transitivity_time_first, transitivity_time_recent_ordered, transitivity_time_first_ordered, transitivity_time_recent_interrupted, transitivity_time_first_interrupted, transitivity_count_interrupted†, transitivity_binary_interrupted†, transitivity_exp_decay_interrupted†
Cyclic	cyclic_binary, cyclic_count, cyclic_binary_ordered, cyclic_count_ordered, cyclic_exp_decay, cyclic_exp_decay_ordered, cyclic_time_recent, cyclic_time_first, cyclic_time_recent_ordered, cyclic_time_first_ordered, cyclic_time_recent_interrupted, cyclic_time_first_interrupted, cyclic_count_interrupted†, cyclic_binary_interrupted†, cyclic_exp_decay_interrupted†
Sending balance	sending_balance_binary, sending_balance_count, sending_balance_binary_ordered, sending_balance_count_ordered, sending_balance_exp_decay, sending_balance_exp_decay_ordered, sending_balance_time_recent, sending_balance_time_first, sending_balance_time_recent_ordered, sending_balance_time_first_ordered, sending_balance_time_recent_interrupted, sending_balance_time_first_interrupted, sending_balance_count_interrupted†, sending_balance_binary_interrupted†, sending_balance_exp_decay_interrupted†
Receiving balance	receiving_balance_binary, receiving_balance_count, receiving_balance_binary_ordered, receiving_balance_count_ordered, receiving_balance_exp_decay, receiving_balance_exp_decay_ordered, receiving_balance_time_recent, receiving_balance_time_first, receiving_balance_time_recent_ordered, receiving_balance_time_first_ordered, receiving_balance_time_recent_interrupted, receiving_balance_time_first_interrupted, receiving_balance_count_interrupted†, receiving_balance_binary_interrupted†, receiving_balance_exp_decay_interrupted†

† Post-hoc only at the moment; the simulator’s inner loop does not yet generate from these variants.

---

Bipartite support

The simulator’s closure-family state matrices are sized |U| × |U| over the unified actor universe; endogenous_features() is universe-agnostic via its string-keyed history. Both paths handle bipartite and arbitrarily overlapping sender / receiver sets for every closure-family statistic, and are cross-validated on bipartite seeds by the parity test suite.

---

Parity status

The same closure-family statistics produced by the simulator “on the fly” and re-computed post-hoc agree row-for-row on timing variants (*_time_recent, *_time_first, *_time_*_interrupted), but the count and exp_decay families currently show an O(events)-scale gap that needs investigation — see Validation experiments / E4. Until that gap closes, treat the post-hoc engine as authoritative for count statistics in downstream model fitting.