Publish AriaSOC synthetic laboratory cases

README.md

@@ -0,0 +1,56 @@
+---
+license: cc-by-4.0
+pretty_name: AriaSOC laboratory cases
+task_categories:
+  - text-classification
+  - other
+tags:
+  - cybersecurity
+  - soc
+  - alert-triage
+  - siem
+  - synthetic
+language:
+  - en
+  - fa
+size_categories:
+  - n<1K
+---
+
+# AriaSOC laboratory cases
+
+Synthetic SIEM alerts and case labels used by [AriaSOC](https://huggingface.co/spaces/alirezaaminzadeh/aria-soc).
+
+**This is generated data** (seed `24`). It is for reproducible demos and tests, not a customer extract. Addresses are documentation-range or RFC1918 laboratory values. Tokens and mailboxes are fake and redacted in the demo.
+
+## Files
+
+| File | Grain | Notes |
+|---|---|---|
+| `data/alerts.csv` | one alert | Required SOC fields plus source, title, optional ATT&CK |
+| `data/cases.csv` | 20 generator scenarios | Incident id, importance, gold label |
+| `data/splits.json` | time cut | Train = earlier 70% of timestamps |
+| `data/evaluation.json` | one run | Recorded baseline vs correlation metrics |
+| `data/sample_preview.json` | UI snapshot | Precomputed KPI HTML; labeled as a sample |
+
+## Schema
+
+`alerts.csv`: `event_id, timestamp, asset_id, rule_id, severity, user_id, source, title, message, src_ip, dest_host, process, technique, case_id, incident_id, importance, gold_label`
+
+Required for the demo parser: `event_id, timestamp, asset_id, rule_id, severity, user_id`. Empty `asset_id` or `user_id` is valid and means **unknown**, not a pass.
+
+Sources: `identity`, `edr`, `network`.
+
+Importance: `important`, `benign`, `noise`, `insufficient`.
+
+## Split
+
+Do **not** shuffle rows of the same incident into train and test. Use `timestamp <= train_until` from `splits.json` as history. The optional LightGBM ranker is fit only on that earlier window.
+
+## Generator
+
+`projects/aria-soc/src/generator.py` in the Space repo, seed `24`. Twenty scenarios include a finance compromise chain, look-alike benign activity, missing-schema events, and a two-day-old recon burst that sits outside the default correlation window.
+
+## License
+
+CC-BY-4.0. Keep the synthetic-data label if you reuse the tables.

data/cases.csv

@@ -0,0 +1,21 @@
+case_id,incident_id,importance,gold_label,title,sources
+C01-credential-spray,I-FIN-001,important,true_positive,Credential spray then success,identity
+C02-ransom-like,I-FIN-001,important,true_positive,Mass rename and backup deletion,edr
+C03-lateral-smb,I-FIN-001,important,true_positive,SMB lateral from finance host,"network,edr"
+C04-impossible-travel,I-HR-004,important,true_positive,Two countries in 40 minutes,identity
+C05-beacon,I-DEV-005,important,true_positive,Periodic outbound beacon,network
+C06-encoded-ps,I-JUMP-006,important,true_positive,Encoded PowerShell on jump host,edr
+C07-phish-oauth,I-HR-007,important,true_positive,Mail click then OAuth grant,"identity,edr"
+C08-dns-nx,I-DEV-005,important,true_positive,High NXDOMAIN rate,network
+C09-mfa-disable,I-ADM-009,important,true_positive,Privileged MFA disabled,identity
+C10-svc-interactive,I-BAK-010,important,true_positive,Backup service account interactive logon,identity
+C11-vuln-scan,I-BEN-011,benign,benign,Approved vulnerability scan,network
+C12-av-update-storm,I-NOI-012,noise,noise,Antivirus definition storm,edr
+C13-travel-vpn,I-BEN-013,benign,benign,Travel plus corporate VPN,identity
+C14-helpdesk-rdp,I-BEN-014,benign,benign,Scheduled helpdesk RDP,network
+C15-software-deploy,I-BEN-015,benign,benign,Change-window process storm,edr
+C16-broken-schema,I-UNK-016,insufficient,insufficient,Events missing required context,"identity,edr"
+C17-orphan-low,I-UNK-017,insufficient,insufficient,Single low-severity orphan,network
+C18-shared-ip-unrelated,I-MIX-018,benign,benign,"Shared NAT IP, unrelated users",network
+C19-dup-c02,I-FIN-001,important,true_positive,Near-duplicate ransom alerts that must survive,edr
+C20-early-recon,I-FIN-001,important,true_positive,Recon two days earlier; outside default window,network

data/evaluation.json

@@ -0,0 +1,13 @@
+{
+  "has_ground_truth": true,
+  "input_events": 643,
+  "cluster_count": 92,
+  "baseline_queue_size": 48,
+  "noise_reduction": 0.8569,
+  "unwitnessed_claim_rate": 0.0,
+  "precision_at_k": 0.875,
+  "pairwise_precision": 0.8578,
+  "important_pairwise_recall": 0.8034,
+  "important_incident_recall": 1.0,
+  "top_k": 8
+}

data/sample_preview.json

@@ -0,0 +1,26 @@
+{
+  "run_id": "aria-soc-9d0709194d",
+  "status": "ok",
+  "status_label": "Live analysis finished",
+  "metrics": {
+    "has_ground_truth": true,
+    "input_events": 643,
+    "cluster_count": 92,
+    "baseline_queue_size": 48,
+    "noise_reduction": 0.8569,
+    "unwitnessed_claim_rate": 0.0,
+    "precision_at_k": 0.875,
+    "pairwise_precision": 0.8578,
+    "important_pairwise_recall": 0.8034,
+    "important_incident_recall": 1.0,
+    "top_k": 8
+  },
+  "baseline_metrics": {
+    "queue_size": 48,
+    "noise_reduction": 0.9253,
+    "precision_at_k": 1.0,
+    "important_incident_recall": 0.8571
+  },
+  "kpi_html": "\n<div class=\"kpi-grid\">\n  <article class=\"kpi-card\"><p class=\"kpi-label\">Needs review</p><p class=\"kpi-value\">32</p><p class=\"kpi-sub\">Cases in the analyst queue</p></article>\n  <article class=\"kpi-card\"><p class=\"kpi-label\">Events kept</p><p class=\"kpi-value\">643</p><p class=\"kpi-sub\">Dropped 0 unusable rows</p></article>\n  <article class=\"kpi-card\"><p class=\"kpi-label\">Incomplete</p><p class=\"kpi-value\">53</p><p class=\"kpi-sub\">Missing user, asset, or second source</p></article>\n  <article class=\"kpi-card\"><p class=\"kpi-label\">Queue compression</p><p class=\"kpi-value\">86%</p><p class=\"kpi-sub\">Always read with important-case recall</p></article>\n  <article class=\"kpi-card\"><p class=\"kpi-label\">Runtime</p><p class=\"kpi-value\">7.2s</p><p class=\"kpi-sub\">Warm CPU target < 30s</p></article>\n</div>\n",
+  "precomputed": true
+}

data/splits.json

@@ -0,0 +1,5 @@
+{
+  "seed": 24,
+  "train_until": "2026-09-18T08:23:00",
+  "note": "Time split by alert timestamp. Do not shuffle rows of the same incident."
+}