test(clustering): fixture 4 paused_campaign + active_days/time_window

Adds the actor.active_days primitive to the campaign factory so a
DSL actor can be bound to specific day indexes. Falls back to the
non-paused day pool when absent (existing fixtures unchanged).
Intersects with pause_windows so the campaign-wide silence still
wins if both are set.

Adds time_window_clusterer reference to fixture_harness — union-find
over attackers, edge if their session time-ranges are within
gap_days of each other. Deliberately-bad reference for fixture 4:
multi-day silent stretches fragment a single campaign because the
clusterer has no signal that bridges the gap.

Fixture 4 (paused_campaign): one campaign modeled as two DSL actors
representing the operator's two operational windows (active days
1-2 and 6-7), separated by a silent stretch (days 3-5). Both share
JA3 + HASSH + payload + C2 callback; only their active_days differ.

Five tests: corpus shape (rows in their windows, shared signals),
pipeline pass via fingerprint_clusterer at level=campaign,
adversarial fragmentation via time_window_clusterer (1-day union
threshold cannot bridge the 4-day silence → completeness collapses),
huge-gap sanity (gap_days=10 unions both halves), silent-stretch
invariant (no session leaks into the configured pause window).

Identity-level scoring is fixture 2's job; this fixture is
campaign-level only — modeling caveat documented in the YAML.
This commit is contained in:
2026-04-26 07:39:46 -04:00
parent 0def6f7e37
commit 304592abfe
5 changed files with 334 additions and 11 deletions

View File

@@ -0,0 +1,24 @@
# Bounds for fixture 4 (paused_campaign).
#
# Ground truth at campaign-level: 1 campaign of 2 observation rows
# (one per DSL actor — modeling the operator's two operational
# windows). A correct algorithm scores 1.0 on every metric.
#
# Completeness is the load-bearing metric: a clusterer that lets a
# multi-day silent period split the campaign tanks completeness
# (the one true class is split across two predicted clusters,
# matching the gap). The adversarial time_window_clusterer
# demonstrates this and the bound below rejects it.
#
# This fixture is CAMPAIGN-LEVEL ONLY (see the fixture YAML for
# why). No identity-level scoring.
#
# Bounds are loose at v1; tighten as the algorithm matures.
adjusted_rand_index:
min: 0.85
homogeneity:
min: 0.90
completeness:
min: 0.80
singleton_recall:
min: 0.95

View File

@@ -0,0 +1,85 @@
# Fixture 4 (paused_campaign) — see development/CAMPAIGN_CLUSTERING.md §2.
#
# One campaign that operates in two sprints with a multi-day silence
# between them:
#
# active days 1-2 (0-indexed [0, 1]) — Delivery, Exploitation
# silent days 3-5 (0-indexed [2, 3, 4]) — pause window
# active days 6-7 (0-indexed [5, 6]) — Discovery, Lateral Movement,
# Exfiltration
#
# Modeled as TWO DSL actors representing the same operator's two
# operational windows. Both share JA3, HASSH, payload, and C2
# callback — the stable signals a fingerprint-driven clusterer
# resolves on. Their ``active_days`` differ so each operator-half
# emits sessions in disjoint time ranges, which is what makes the
# adversarial time-window clusterer fragment the campaign.
#
# Two-actor modeling caveat: the factory mints a separate
# ``truth_identity_id`` per DSL actor by design (see IDENTITY_
# RESOLUTION.md — identities are recovered from signals, not
# declared in the DSL). This is a CAMPAIGN-LEVEL fixture only;
# identity-level scoring is fixture 2's job. The bound floors below
# apply at level=campaign.
#
# Pass condition: a fingerprint-driven clusterer must fold both
# operational windows into one cluster (shared JA3 + HASSH +
# payload). A clusterer that lets a multi-day quiet period split
# the campaign fails the completeness floor.
#
# Adversarial condition: ``time_window_clusterer`` (union sessions
# within ≤1 day of each other) is unable to bridge the 4-day silent
# stretch and splits the campaign into "before pause" and "after
# pause" clusters. Completeness collapses; the bound floor rejects
# this clusterer.
campaign:
id: paused-campaign-001
duration_days: 7
pause_windows:
- [2, 4] # campaign-wide silence days 3-5 (0-indexed)
actors:
- id: ops-sprint-1
asn: 64520
ip_pool: sticky
ja3: "771,4865-4866-4867-49195-49199-49196-49200,0-23-65281-10-11-35-16-5-13-18-51-45-43-27,29-23-24,0"
hassh: "paused-op-dddddddd-dddddddd-dddddddd"
hours_active_utc: [9, 10, 11, 12, 13, 14, 15, 16]
jitter_seconds: 60
active_days: [0, 1]
- id: ops-sprint-2
asn: 64520 # same ASN — operator stays on same egress
ip_pool: sticky
ja3: "771,4865-4866-4867-49195-49199-49196-49200,0-23-65281-10-11-35-16-5-13-18-51-45-43-27,29-23-24,0"
hassh: "paused-op-dddddddd-dddddddd-dddddddd"
hours_active_utc: [9, 10, 11, 12, 13, 14, 15, 16]
jitter_seconds: 60
active_days: [5, 6]
phases:
- name: delivery
actor: ops-sprint-1
target_selector: { service: ssh, count: 2 }
dwell_seconds: 1
- name: exploitation
actor: ops-sprint-1
tool_signature:
payload_hash: "paused-op-stage1-payload"
c2_callback: "c2.paused-op.example"
target_selector: { service: ssh, count: 2 }
dwell_seconds: 5
- name: discovery
actor: ops-sprint-2
target_selector: { service: ssh, count: 2 }
dwell_seconds: 5
- name: lateral_movement
actor: ops-sprint-2
tool_signature:
payload_hash: "paused-op-stage1-payload"
c2_callback: "c2.paused-op.example"
target_selector: { service: ssh, count: 2 }
dwell_seconds: 5
- name: exfiltration
actor: ops-sprint-2
tool_signature:
c2_callback: "c2.paused-op.example"
target_selector: { service: ssh, count: 2 }
dwell_seconds: 5