feat(profiler/behave_shell): emit temporal.session_duration

Bucket ctx.duration_s against SESSION_DURATION_SHORT_MAX (60s) /
MEDIUM_MAX (600s) / LONG_MAX (3600s); else marathon. Direct
measurement, confidence 0.85. Skip emission only when no commands
and zero duration. New _features/temporal.py module opens Phase E.
This commit is contained in:
2026-05-04 00:10:57 -04:00
parent 46775fc0e5
commit 627fa59c15
4 changed files with 118 additions and 0 deletions

View File

@@ -24,6 +24,9 @@ from decnet.profiler.behave_shell._features.cognitive import (
inter_command_consistency,
inter_command_latency_class,
)
from decnet.profiler.behave_shell._features.temporal import (
session_duration,
)
from decnet.profiler.behave_shell._features.motor import (
command_chunking,
error_correction,
@@ -59,4 +62,5 @@ FEATURES: tuple[FeatureFn, ...] = (
error_resilience_retry_tactic,
error_resilience_frustration_typing,
error_resilience_fallback_to_man,
session_duration,
)

View File

@@ -0,0 +1,50 @@
"""``temporal.*`` feature functions — per-session subset.
Phase E ships the four ``temporal.*`` primitives that don't need
observation history. The other three (``session_timing``,
``persistence``, ``lifecycle_markers.idle_periodicity``) are Tier B
and computed by the attribution engine, not the extractor.
Step E.1: ``temporal.session_duration``.
"""
from __future__ import annotations
from typing import Iterator
from decnet_behave_core.spec.envelope import Observation
from decnet.profiler.behave_shell._ctx import SessionContext
from decnet.profiler.behave_shell._features._emit import make_observation
from decnet.profiler.behave_shell._thresholds import (
SESSION_DURATION_LONG_MAX,
SESSION_DURATION_MEDIUM_MAX,
SESSION_DURATION_SHORT_MAX,
)
def session_duration(ctx: SessionContext) -> Iterator[Observation]:
"""Emit ``temporal.session_duration`` ∈ {short, medium, long, marathon}.
Direct measurement off ``ctx.duration_s``. Skip emission only when
the session has neither commands nor any duration to speak of —
a one-event session with ``duration_s == 0`` and no commands has
nothing honest to bucket. Confidence is high — duration is a fact,
not an inference.
"""
if ctx.duration_s <= 0.0 and not ctx.commands:
return
d = ctx.duration_s
if d < SESSION_DURATION_SHORT_MAX:
value = "short"
elif d < SESSION_DURATION_MEDIUM_MAX:
value = "medium"
elif d < SESSION_DURATION_LONG_MAX:
value = "long"
else:
value = "marathon"
yield make_observation(
ctx,
primitive="temporal.session_duration",
value=value,
confidence=0.85,
)

View File

@@ -170,6 +170,20 @@ TOOL_VOCAB_BROAD_MIN: int = 10
FRUSTRATION_LOW_MAX: float = 0.10
FRUSTRATION_MODERATE_MAX: float = 0.30
# ── temporal.session_duration (Step E.1) ───────────────────────────────────
# Bucket edges (seconds) for ``ctx.duration_s``:
#
# duration_s < SESSION_DURATION_SHORT_MAX → short
# duration_s < SESSION_DURATION_MEDIUM_MAX → medium
# duration_s < SESSION_DURATION_LONG_MAX → long
# else → marathon
#
# 60s / 600s / 3600s are the BEHAVE-EXTRACTOR.md defaults; D.8-equivalent
# re-tune for E lands when calibration corpus is run.
SESSION_DURATION_SHORT_MAX: float = 60.0
SESSION_DURATION_MEDIUM_MAX: float = 600.0
SESSION_DURATION_LONG_MAX: float = 3600.0
# ── motor.keystroke_cadence (Step B.1) ──────────────────────────────────────
# Typing bursts split at gaps > IKI_THINK_MAX_S so think-pauses between
# commands don't inflate the within-burst CV. Mirrors the prototype's

View File

@@ -0,0 +1,50 @@
"""Step E.1: ``temporal.session_duration``."""
from __future__ import annotations
from decnet.profiler.behave_shell import extract_session
from decnet.profiler.behave_shell._parse import AsciinemaEvent
PRIMITIVE = "temporal.session_duration"
def _of(observations: list, primitive: str):
obs = [o for o in observations if o.primitive == primitive]
assert len(obs) == 1, f"expected exactly one {primitive}, got {len(obs)}"
return obs[0]
def test_empty_session_no_emission() -> None:
out = list(extract_session([], sid="dur-empty"))
assert [o for o in out if o.primitive == PRIMITIVE] == []
def test_under_60s_emits_short() -> None:
events: list[AsciinemaEvent] = [(0.0, "i", "a"), (30.0, "i", "b")]
obs = _of(list(extract_session(events, sid="dur-short")), PRIMITIVE)
assert obs.value == "short"
def test_under_600s_emits_medium() -> None:
events: list[AsciinemaEvent] = [(0.0, "i", "a"), (300.0, "i", "b")]
obs = _of(list(extract_session(events, sid="dur-med")), PRIMITIVE)
assert obs.value == "medium"
def test_under_3600s_emits_long() -> None:
events: list[AsciinemaEvent] = [(0.0, "i", "a"), (1800.0, "i", "b")]
obs = _of(list(extract_session(events, sid="dur-long")), PRIMITIVE)
assert obs.value == "long"
def test_over_3600s_emits_marathon() -> None:
events: list[AsciinemaEvent] = [(0.0, "i", "a"), (7200.0, "i", "b")]
obs = _of(list(extract_session(events, sid="dur-marathon")), PRIMITIVE)
assert obs.value == "marathon"
def test_high_confidence() -> None:
"""Duration is a fact, not an inference — confidence stays high."""
events: list[AsciinemaEvent] = [(0.0, "i", "a"), (30.0, "i", "b")]
obs = _of(list(extract_session(events, sid="dur-conf")), PRIMITIVE)
assert obs.confidence >= 0.80