feat: implement Pillars 2-4 and fix a production false-positive in Pillar 1

Pillar 1 v1.1 (shipped earlier) flagged benign browsing bursts as
beaconing: on short captures, grid periods longer than the observed
span never complete a cycle, so significance saturates to 1 and the
score sticks at exactly 0.5. v1.2 caps the grid at a third of the
observed span, requires a minimum number of completed cycles, and
excludes protocols that are periodic by nature (NTP, DNS, mDNS,
ICMP). Validated against 300 seeds per scenario: false positives on
bursty traffic drop from the 290-300/300 that v1.1 produced to 0/300,
with every beacon-detection scenario scoring unchanged.

Add Pillar 2 (persistent/long-lived sessions, with Zeek's fragmented
TCP sessions rejoined after a 5-minute gap), Pillar 3 (DNS tunneling
via Shannon-entropy information volume per subdomain per hour; the
Jensen-Shannon divergence was measured and does not separate tunnels
from benign CDN subdomains, so it stays in the evidence at zero
weight instead of in the score), and Pillar 4 (deterministic
noisy-OR fusion across the three pillars).

The fusion formula was validated against 8 labeled real-traffic
vectors (IoT-23, CTU-Normal) with tolerance <= 1e-4. A known
limitation is kept, not hidden: a benign Philips HUE bulb's strict
periodic polling and a 21.9h session score it above three real C2
hosts. No weight or threshold was tuned to mask this; it is
documented in the architecture docs instead.

Also fixes real bugs found along the way: ingest_dir's partition
count summed file names instead of distinct dates, DuckDB errors
were counted as bad rows, /api/alerts always returned an empty list
regardless of the last analysis, and the dashboard still claimed
every pillar was unimplemented.

Restores four design corrections that were missing from
docs/architecture.md (KL to Jensen-Shannon, the persistent-session
detector, the JA4 licensing boundary, and deterministic reduction
order), and catalogs the public datasets used for validation
(IoT-23, CTU-Normal, Mendeley DTQBC) with their licenses and SHA-256
checksums under research/validation/.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VAyr8d7HuAdUGB6aA5xmDD

research/validation/p2real.py

@@ -0,0 +1,52 @@
+# Pillar 2 candidate metrics on real Zeek conn.log: chained session duration per tuple, persistence score,
+# per-host robust z of ln(duration), PCR, failed attempts.
+import sys, math, collections, statistics
+def read(path):
+    fields = None
+    for line in open(path, errors='replace'):
+        line = line.rstrip('\n')
+        if line.startswith('#fields'): fields = line.split('\t')[1:]; continue
+        if line.startswith('#') or not line: continue
+        yield dict(zip(fields, line.split('\t')))
+num = lambda x: float(x) if x not in ('-', '', '(empty)') else 0.0
+GAP = 3600.0
+def sessions(rows):
+    by4 = collections.defaultdict(list)
+    for r in rows:
+        if r['proto'] != 'tcp': continue
+        by4[(r['id.orig_h'], r['id.orig_p'], r['id.resp_h'], r['id.resp_p'])].append(r)
+    out = []
+    for k, rs in by4.items():
+        rs.sort(key=lambda r: float(r['ts']))
+        cur = None
+        for r in rs:
+            s = float(r['ts']); e = s + num(r['duration'])
+            cont = cur is not None and 'S' not in r['history'] and s - cur['end'] <= GAP
+            if cont:
+                cur['end'] = max(cur['end'], e); cur['ob'] += num(r['orig_bytes']); cur['rb'] += num(r['resp_bytes']); cur['parts'] += 1
+            else:
+                if cur: out.append(cur)
+                cur = dict(key=k, start=s, end=e, ob=num(r['orig_bytes']), rb=num(r['resp_bytes']), parts=1)
+        out.append(cur)
+    return out
+for path in sys.argv[1:]:
+    rows = list(read(path))
+    ss = sessions(rows)
+    fails = collections.Counter((r['id.orig_h'], r['id.resp_h'], r['id.resp_p']) for r in rows if r['proto'] == 'tcp' and r['conn_state'] in ('S0', 'REJ', 'RSTOS0', 'SH'))
+    host_ld = collections.defaultdict(list)
+    for s in ss:
+        d = s['end'] - s['start']
+        if d > 0: host_ld[s['key'][0]].append(math.log(d))
+    best = {}
+    for s in ss:
+        t = (s['key'][0], s['key'][2], s['key'][3]); d = s['end'] - s['start']
+        if t not in best or d > best[t]['end'] - best[t]['start']: best[t] = s
+    ranked = sorted(best.items(), key=lambda kv: -(kv[1]['end'] - kv[1]['start']))
+    print(f"== {path}: tcp sessions={len(ss)} tuples={len(best)}")
+    for t, s in ranked[:5]:
+        d = s['end'] - s['start']; xs = host_ld[t[0]]
+        xs = xs or [0.0]; med = statistics.median(xs); mad = statistics.median(abs(x - med) for x in xs) or 1e-9
+        z = (math.log(d) - med) / (1.4826 * mad) if d > 0 else 0
+        pers = min(1, max(0, math.log(d / 3600) / math.log(24))) if d > 0 else 0
+        tot = s['ob'] + s['rb']; pcr = (s['ob'] - s['rb']) / tot if tot else 0
+        print(f"  {t} L={d:.0f}s ({d/3600:.2f}h) parts={s['parts']} persistence={pers:.3f} host_robust_z={z:.1f} ob={s['ob']:.0f} rb={s['rb']:.0f} PCR={pcr:+.2f} B/h={(tot/(d/3600)) if d>0 else 0:.0f} failed_attempts={fails[t]}")