UC feedback: add MITRE TA-codes, disposition, attack reports, per-incident GraphML

README.md

@@ -42,7 +42,7 @@ configs:
 
 ## Overview
 
-A large-scale, labeled cybersecurity dataset derived from production Security Operations Center (SOC) data processed by [WitFoo Precinct](https://www.witfoo.com/) version 6.x. The dataset contains **2.07 million sanitized security events** (signal logs) and **provenance graphs** (10,442 incident graphs with 30,092 nodes and 595,618 edges) from real enterprise network monitoring across multiple organizations.
+A large-scale, labeled cybersecurity dataset derived from production Security Operations Center (SOC) data processed by [WitFoo Precinct](https://www.witfoo.com/) version 6.x. The dataset contains **2.1 million sanitized security events** (signal logs) and **provenance graphs** (13,119 incident graphs with 35,133 nodes and 634,190 edges) from real enterprise network monitoring across multiple organizations. Each incident has a natural-language threat-hunting report describing the attack chain, and per-incident GraphML files for direct loading in graph analysis tools.
 
 **Available in two sizes:**
 - [`witfoo/precinct6-cybersecurity`](https://huggingface.co/datasets/witfoo/precinct6-cybersecurity) — 2M signals (this dataset)
@@ -124,12 +124,18 @@ Tabular format ideal for ML classification, anomaly detection, and feature engin
 | `vendor_code` | string | Vendor-specific event code (e.g., `ASA-4-106023` for Cisco) |
 | `message_sanitized` | string | Full sanitized raw log message (syslog, XML, JSON, CSV depending on source) |
 | `label_binary` | string | `malicious`, `suspicious`, or `benign` (see [Labeling](#labeling-methodology)) |
-| `label_confidence` | float | Confidence score for the label (0.0–1.0) |
-| `attack_techniques` | string | JSON array of MITRE ATT&CK technique IDs |
-| `attack_tactics` | string | JSON array of MITRE ATT&CK tactic names |
+| `label_confidence` | float | Confidence score for the label (0.0–1.0). See [Scoring](#scoring). |
+| `attack_techniques` | string | JSON array of MITRE ATT&CK technique IDs (e.g., `["T1041", "T1567"]`) |
+| `attack_tactics` | string | JSON array of MITRE ATT&CK tactic IDs in `TA0001`-style (e.g., `["TA0009", "TA0010"]`) |
+| `defense_techniques` | string | JSON array of MITRE D3FEND defense technique IDs |
 | `mo_name` | string | Modus operandi / attack campaign type (e.g., `Data Theft`, `Phishing`) |
-| `suspicion_score` | float | WitFoo-computed suspicion score (0.0–1.0) |
+| `suspicion_score` | float | WitFoo-computed suspicion score (0.0–1.0). See [Scoring](#scoring). |
 | `lifecycle_stage` | string | Kill chain stage (e.g., `initial-compromise`, `complete-mission`) |
+| `disposition` | string | Raw Precinct status (`Disrupted`, `Investigating`, `Resolved`, `Dismissed`, `False Positive`, `Unprocessed`). See [Ground Truth](#ground-truth-and-disposition). |
+| `disposition_category` | string | Bucketed disposition (`confirmed-malicious`, `false-positive`, `dismissed`, `automated`) |
+| `is_false_positive` | bool | True if SOC analyst marked this incident as a false positive |
+| `status_name` | string | Same as `disposition` (raw Precinct status); kept for clarity |
+| `incident_ids` | string | JSON array of incident UUIDs referencing this artifact (empty for benign/suspicious) |
 | `matched_rules` | string | JSON array of WitFoo lead rule descriptions that matched this event |
 | `set_roles` | string | JSON array of WitFoo classification set roles (e.g., `Exploiting Host`, `C2 Server`) |
 | `product_name` | string | Security product that generated this event (e.g., `ASA Firewall`, `Falcon`) |
@@ -201,8 +207,10 @@ Complete incident records as produced by WitFoo Precinct's threat detection engi
 
 ### Additional Files
 
-- **`graph/graph.graphml`** — Full provenance graph in GraphML format for NetworkX, Gephi, or graph databases
+- **`graph/graph.graphml`** — Full provenance graph in GraphML format (streaming write; ~477 MB)
 - **`graph/graph.json`** — NetworkX node-link JSON format
+- **`graph/incidents_graphml/{incident_id}.graphml`** — 13,119 per-incident GraphML files (~1 MB total). Each is small and loadable in Gephi, NetworkX, igraph, or DGL. Useful for graph-based research where loading the entire dataset is impractical.
+- **`graph/attack_reports.jsonl`** — Natural-language threat-hunting reports, one per incident. See [Attack Reports](#attack-reports).
 - **`reference/lead_rules_catalog.json`** — Complete catalog of 261 WitFoo lead detection rules, 158 security products, 106 classification sets, and 216 stream-to-product mappings
 
 ## Data Provenance
@@ -323,9 +331,17 @@ Labels are derived from two sources: WitFoo Precinct's incident analysis and lea
 
 | Label | Count | Percentage |
 |-------|-------|------------|
-| `benign` | 1,899,723 | 91.7% |
-| `malicious` | 125,780 | 6.1% |
-| `suspicious` | 45,420 | 2.2% |
+| `benign` | 1,899,587 | 90.4% |
+| `malicious` | 155,520 | 7.4% |
+| `suspicious` | 45,256 | 2.2% |
+
+Disposition breakdown for malicious records (raw `status_name`):
+
+| Disposition | Count | Meaning |
+|-------------|-------|---------|
+| `Disrupted` | 86,760 | Analyst confirmed and intervened |
+| `Unprocessed` | 68,740 | Automated detection, not yet reviewed |
+| `Dismissed` | 20 | Analyst dismissed (low confidence) |
 
 **114M Dataset (`witfoo/precinct6-cybersecurity-100m`):**
 
@@ -354,9 +370,77 @@ The `matched_rules` column contains JSON arrays of rule descriptions matched for
 | User Account Created | Windows Event ID 4720 | Exploiting Target | Exploiting Host |
 | Special privileges assigned | Windows Event ID 4672 | Exploiting Target | Exploiting Host |
 
+### Ground Truth and Disposition
+
+**All labels in this dataset are derived from WitFoo Precinct's automated incident correlation engine — there is no independent, analyst-verified ground truth.** Researchers should treat Precinct's analysis as a strong but imperfect oracle. The `disposition` column lets you assess label quality on a per-record basis:
+
+| `disposition` | Meaning | Confidence in label |
+|---------------|---------|---------------------|
+| `Disrupted` | SOC analyst confirmed the incident and intervened | High — human-confirmed malicious |
+| `Investigating` | SOC analyst is actively investigating | Medium — analyst engaged |
+| `Resolved` | SOC analyst confirmed and resolved | High — human-confirmed malicious |
+| `Dismissed` | SOC analyst dismissed the incident | Negative — analyst rejected |
+| `False Positive` | SOC analyst confirmed false positive | Negative — analyst rejected |
+| `Unprocessed` | Automated detection, no human review | Lower — Precinct-confidence only |
+
+The `disposition_category` column buckets these into four values for easier filtering: `confirmed-malicious`, `false-positive`, `dismissed`, `automated`. For experiments where ground-truth quality matters, restrict to records where `disposition` ∈ {`Disrupted`, `Resolved`} to compare against analyst-confirmed labels.
+
+For benign and suspicious records, `disposition` is `Unprocessed` (no incident association). For malicious records, `disposition` reflects the parent incident's status at extraction time.
+
+### Scoring
+
+The dataset exposes two related score fields:
+
+- **`suspicion_score`** (float, 0.0–1.0) — Precinct's proprietary suspicion score. Populated for malicious records from the parent incident; zero for benign and suspicious records. Range observed in this dataset: 0.25 to 0.98, mean 0.55.
+
+- **`label_confidence`** (float, 0.0–1.0) — Confidence in the assigned `label_binary` tier. Computed deterministically from corroborating signal:
+
+  | Label | Formula |
+  |-------|---------|
+  | `malicious` | `max(0.6, suspicion_score)` clamped to 0.95; lowered to 0.3 if `is_false_positive` |
+  | `suspicious` | `0.4 + 0.1 × n_matched_rules + 0.05 × n_set_roles`, clamped to [0.5, 0.85] |
+  | `benign` | `0.5` (no positive evidence either way) |
+
+  Note: `label_confidence` is **not** the probability the activity is malicious — it indicates how much corroborating evidence supports the assigned tier. See `compute_label_confidence` in [`src/precinct6_dataset/label.py`](https://github.com/witfoo/dataset-from-precinct6/blob/main/src/precinct6_dataset/label.py).
+
+### Attack Reports
+
+The `graph/attack_reports.jsonl` file contains one natural-language threat-hunting report per incident (13,119 reports). Each report is deterministically composed from structured incident metadata (modus operandi, set roles, lead descriptions, MITRE mappings, timestamps) and explicitly documents that it **reflects Precinct's automated correlation engine output, not an independent threat-hunting investigation**.
+
+Each record contains:
+
+| Field | Description |
+|-------|-------------|
+| `incident_id` | Unique incident identifier (matches `incidents.jsonl`) |
+| `report_text` | Multi-sentence narrative paragraph |
+| `report_source` | Always `"template"` — sentences produced from a deterministic template |
+| `mo_name` | Modus operandi |
+| `suspicion_score`, `disposition`, `disposition_category` | Same as signal columns |
+| `attack_techniques`, `attack_tactics` | MITRE ATT&CK mappings |
+| `lead_count`, `node_count`, `edge_count` | Graph structure |
+| `first_observed_at`, `last_observed_at` | Attack chain time bounds |
+| `set_role_names` | WitFoo classification roles assigned |
+| `matched_rules` | Detection rule descriptions that triggered |
+| `products_observed` | Security products that detected activity |
+| `lifecycle_stage` | Kill-chain stage |
+
+Researchers can audit exactly how each sentence is derived by reading [`src/precinct6_dataset/attack_reports.py`](https://github.com/witfoo/dataset-from-precinct6/blob/main/src/precinct6_dataset/attack_reports.py).
+
 ### MITRE ATT&CK Mappings
 
-Attack technique and tactic labels are derived from WitFoo's framework mapping of incident patterns. The `lifecycle_stage` field maps events to the APT kill chain:
+Attack technique and tactic labels are derived from three sources, with deduplication:
+
+1. **WitFoo set role names** attached to the incident (e.g., `C2 Server` → `TA0011` Command and Control, `T1071` Application Layer Protocol)
+2. **Modus operandi** name on the incident (e.g., `Ransomware` → `TA0001`, `TA0002`, `TA0040`; `T1486` Data Encrypted for Impact)
+3. **Per-product framework data** embedded in `incident.nodes.products.frameworks` (when present)
+
+Tactic IDs use the standard MITRE ATT&CK Enterprise format (`TA0001` through `TA0043`). Technique IDs are top-level techniques (no sub-techniques) representing the most likely category for a given role. Researchers wanting precise per-event technique attribution should treat these as priors.
+
+**Per-edge/per-node attribution in graph output:** `attack_tactics`, `attack_techniques`, `set_roles`, `lifecycle_stage`, `label_binary`, `label_confidence`, `suspicion_score`, `disposition` are attached at the **edge** level in both NDJSON and GraphML output. Nodes in `incidents.jsonl` carry their own `sets` and `products` dicts with per-entity information.
+
+The full mapping tables are in [`src/precinct6_dataset/mitre_mapping.py`](https://github.com/witfoo/dataset-from-precinct6/blob/main/src/precinct6_dataset/mitre_mapping.py).
+
+The `lifecycle_stage` field maps events to WitFoo's internal kill-chain model:
 
 1. `initial-compromise` — Initial access to the network
 2. `establish-foothold` — Execution and establishing persistence