wulf-pulse/lib/services/campaign-grouping-service.ts
lorentz 5f308f836f fix(phishing-recipient-seubert): scope campaign grouping Tier 3 to company, not reporting contact
Tier 3 is the only tier automatic (webhook-triggered) grouping ever
reaches, since grouping runs before message parsing. It was scoped to
reports.requester_contact_id, so the same campaign reported by
different employees at the same company never consolidated into one
campaign — each report's evidence/blast-radius view silently
under-reported the campaign's true recipients.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-22 12:34:33 -04:00

493 lines
22 KiB
TypeScript

/**
* Campaign Grouping Service
*
* Shared grouping core called by all 3 trigger sites: the webhook path
* (`webhook-service.ts`'s `triggerPhishingDetection()`), the cron sweep
* (`phishing-sweep-service.ts`), and the on-demand
* `POST /api/phishing/tickets/{id}/analyze` route. One deterministic,
* testable function computes a tiered campaign key (CAMP-01) and
* finds-or-creates a `campaigns` row (CAMP-02) — no duplicated matching
* logic between callers, mirroring `phishing-detector.ts`'s shared-core
* architecture.
*
* D-07 limitation (load-bearing, stated explicitly): `groupReportIntoCampaign`
* always runs BEFORE `parseAndStoreMessage` on the automatic webhook path
* (see `webhook-service.ts`'s `triggerPhishingDetection()` — grouping happens
* first, parsing happens afterward inside `runGatedPhishingStages()`). That
* means at grouping time the CURRENT report never has its own `messages`/
* `indicators` row yet, so Tier 1 (Message-ID) and Tier 2 (attachment-hash/
* URL-domain) — both of which require the report's OWN signal to search
* for candidates — can never match on the automatic path's one-and-only
* grouping call (subsequent webhook events short-circuit via
* `skipIfAlreadyGrouped`). Automatic grouping therefore always resolves via
* Tier 3 (normalized subject + company + 24h window).
*
* Bug fix (debug session phishing-recipient-seubert): Tier 3 previously
* scoped its match to `reports.requester_contact_id` — i.e. it only ever
* merged reports filed by the SAME reporting employee. Since Tier 3 is the
* only tier automatic grouping can ever reach (see D-07 above), that meant
* the same phishing campaign sent to and reported by MULTIPLE different
* employees at the same company could never be consolidated into one
* campaign — each recipient's report silently became its own single-report
* campaign, so any single ticket's evidence/blast-radius view under-reported
* the campaign's true recipient list. Tier 3 now scopes to company + subject
* only (no contact/requester restriction), matching its `sender_subject_
* client` name's original intent of grouping the same external campaign
* across a company, independent of who reported it. (`sender` isn't
* literally available yet at this point — see D-07 — so "client" scoping is
* company-wide, deliberately wider than a single reporter.)
*/
import type { PoolClient } from 'pg';
import { postgresClient } from './postgres-client';
// =============================================================================
// Pure logic — tier-key helpers (mirrors matchesPhishingPatterns / computePhishingContentHash)
// =============================================================================
/**
* D-03: strip leading Re:/Fwd:/Fw: (repeated, case-insensitive), lowercase, trim.
*/
export function normalizeSubject(subject: string | null): string {
let s = (subject ?? '').trim();
const prefixRe = /^(re|fwd|fw):\s*/i;
while (prefixRe.test(s)) {
s = s.replace(prefixRe, '').trim();
}
return s.toLowerCase();
}
/**
* Pitfall 4: indicators.value for indicator_type='url' is a bare URL string,
* not a domain — extract at read time, guarded (malformed/relative URLs are
* possible in real-world phishing emails).
*/
export function extractUrlDomain(url: string): string | null {
try {
return new URL(url).hostname || null;
} catch {
return null;
}
}
// =============================================================================
// Orchestration — groupReportIntoCampaign (mirrors detectPhishingTicket)
// =============================================================================
export interface GroupReportResult {
campaignId: string;
groupMethod: 'message_id' | 'attachment_or_url' | 'sender_subject_client';
created: boolean;
}
interface OwnReportRow {
title: string | null;
company_id: number | null;
created_at: string;
campaign_id: string | null;
}
interface OwnMessageRow {
id: string;
message_id: string | null;
}
interface IndicatorRow {
indicator_type: string;
value: string;
}
interface CandidateIndicatorRow extends IndicatorRow {
message_id: string;
}
interface Tier2CandidateRow {
report_id: string;
campaign_id: string;
title: string | null;
message_id: string;
}
/** Builds a `message_id:...` key — the strongest available signal (Tier 1). */
function computeTier1Key(messageId: string | null): string | null {
return messageId ? `message_id:${messageId}` : null;
}
/** Builds an `attachment_or_url:...` key from own indicators + subject + sender (Tier 2). */
function computeTier2Key(
attachmentHashes: string[],
urlDomains: string[],
normalizedSubject: string,
senderValue: string | null
): string | null {
if (!senderValue || !normalizedSubject) return null;
if (attachmentHashes.length === 0 && urlDomains.length === 0) return null;
const keyParts = [...attachmentHashes, ...urlDomains].sort();
return `attachment_or_url:${keyParts.join(',')}:${normalizedSubject}:${senderValue}`;
}
/**
* Builds a `sender_subject_client:...` key from normalized subject + company
* (Tier 3) — deliberately company-wide, NOT scoped to a single reporting
* contact, so the same campaign reported by different employees at the same
* company still consolidates into one campaign (see file-level bug-fix note).
*/
function computeTier3Key(normalizedSubject: string, companyId: number | null): string | null {
if (!normalizedSubject || !companyId) return null;
return `sender_subject_client:${normalizedSubject}:${companyId}`;
}
/**
* CR-02: decrements a now-abandoned origin campaign's report_count. Shared
* by both places a report can be abandoned from its origin campaign — the
* sibling-migration branch (moving to a DIFFERENT existing campaign) and the
* signal-diverged create-new fall-through (moving to a brand-new campaign) —
* so neither is left with a stale, inflated count. Decrement only; never
* deletes the origin row even if its count reaches 0 (out of scope — see
* CR-02 scope note at the call sites).
*/
async function decrementOriginCampaign(client: PoolClient, originCampaignId: string): Promise<void> {
await client.query(
`UPDATE campaigns
SET report_count = report_count - 1, updated_at = NOW()
WHERE id = $1`,
[originCampaignId]
);
}
/**
* Shared entry point called by the webhook path, the cron sweep, and the
* on-demand `/analyze` route. Looks up the report's own signal (message,
* indicators, requester/company), tries Tier 1 -> Tier 2 -> Tier 3 matching
* queries in order inside a single transaction (Pitfall 2 — `campaigns.
* campaign_key` has no UNIQUE constraint), and either links the report to an
* existing campaign (bumping report_count/last_seen_at, CAMP-02) or inserts a
* new campaigns row.
*
* D-08: `opts.skipIfAlreadyGrouped` short-circuits (returns null) when the
* report is already linked — used by the automatic webhook/cron path to
* avoid redundant work on every re-fire/sweep pass. The `/analyze` route
* omits this option so it always re-runs full tiered matching (it may
* upgrade a Tier-3-only report to Tier 1 after `parseAndStoreMessage` has
* just populated `messages`/`indicators` for the first time).
*
* D-04: never merges campaigns — attaches to the first/best tiered match
* found, or creates a new campaign. Multi-campaign-merge logic is explicitly
* out of scope this phase.
*/
export async function groupReportIntoCampaign(
reportId: string,
opts?: { skipIfAlreadyGrouped?: boolean }
): Promise<GroupReportResult | null> {
try {
if (opts?.skipIfAlreadyGrouped) {
const existing = await postgresClient.query<{ campaign_id: string | null }>(
`SELECT campaign_id::text AS campaign_id FROM reports WHERE id = $1`,
[reportId]
);
if (existing.rows[0]?.campaign_id) {
return null; // already grouped, short-circuit (D-08)
}
}
return await postgresClient.transaction(async (client) => {
const ownReportRes = await client.query<OwnReportRow>(
`SELECT title, company_id, created_at, campaign_id::text AS campaign_id
FROM reports
WHERE id = $1`,
[reportId]
);
const ownReport = ownReportRes.rows[0];
if (!ownReport) {
throw new Error(`groupReportIntoCampaign: report ${reportId} not found`);
}
const normalizedSubject = normalizeSubject(ownReport.title);
// Own messages row, if `parseAndStoreMessage` has already run for this
// report (D-07 — not the case for the automatic webhook/cron path
// until an explicit /analyze call has happened at least once).
const ownMessageRes = await client.query<OwnMessageRow>(
`SELECT id::text AS id, message_id
FROM messages
WHERE report_id = $1
LIMIT 1`,
[reportId]
);
const ownMessage = ownMessageRes.rows[0] ?? null;
let matchCampaignId: string | null = null;
let matchGroupMethod: GroupReportResult['groupMethod'] | null = null;
// ---------------------------------------------------------------------
// Tier 1: Message-ID. Self-exclusion (`r.id != $2`) is REQUIRED — without
// it this query would trivially match the report's own messages row
// against itself, double-incrementing its own already-linked campaign
// on every re-run (plan-checker finding).
// ---------------------------------------------------------------------
if (ownMessage?.message_id) {
const tier1 = await client.query<{ campaign_id: string }>(
`SELECT r.campaign_id::text AS campaign_id
FROM messages m
JOIN reports r ON r.id = m.report_id
WHERE m.message_id = $1
AND r.campaign_id IS NOT NULL
AND r.id != $2
ORDER BY r.created_at ASC
LIMIT 1`,
[ownMessage.message_id, reportId]
);
if (tier1.rows[0]?.campaign_id) {
matchCampaignId = tier1.rows[0].campaign_id;
matchGroupMethod = 'message_id';
}
}
// ---------------------------------------------------------------------
// Tier 2: attachment-hash / URL-domain + normalized subject + sender,
// within a 24h window (D-02). Only reachable if the own report has its
// own indicators (i.e. parseAndStoreMessage already ran for it).
// Self-exclusion (`r.id != $1`) required for the same reason as Tier 1.
// ---------------------------------------------------------------------
let ownAttachmentHashes: string[] = [];
let ownUrlDomains: string[] = [];
let ownSenderValue: string | null = null;
if (!matchCampaignId && ownMessage) {
const ownIndicatorsRes = await client.query<IndicatorRow>(
`SELECT indicator_type, value
FROM indicators
WHERE message_id = $1`,
[ownMessage.id]
);
ownAttachmentHashes = ownIndicatorsRes.rows
.filter((i) => i.indicator_type === 'attachment_hash')
.map((i) => i.value);
ownUrlDomains = ownIndicatorsRes.rows
.filter((i) => i.indicator_type === 'url')
.map((i) => extractUrlDomain(i.value))
.filter((d): d is string => d !== null);
ownSenderValue =
ownIndicatorsRes.rows.find((i) => i.indicator_type === 'sender')?.value ?? null;
if (
(ownAttachmentHashes.length > 0 || ownUrlDomains.length > 0) &&
ownSenderValue &&
normalizedSubject
) {
const tier2Candidates = await client.query<Tier2CandidateRow>(
`SELECT r.id::text AS report_id, r.campaign_id::text AS campaign_id, r.title,
m.id::text AS message_id
FROM messages m
JOIN reports r ON r.id = m.report_id
WHERE r.campaign_id IS NOT NULL
AND r.id != $1
AND r.created_at BETWEEN $2::timestamptz - INTERVAL '24 hours'
AND $2::timestamptz + INTERVAL '24 hours'`,
[reportId, ownReport.created_at]
);
if (tier2Candidates.rows.length > 0) {
const candidateMessageIds = tier2Candidates.rows.map((c) => c.message_id);
const candidateIndicatorsRes = await client.query<CandidateIndicatorRow>(
`SELECT message_id::text AS message_id, indicator_type, value
FROM indicators
WHERE message_id = ANY($1::uuid[])`,
[candidateMessageIds]
);
const indicatorsByMessage = new Map<string, CandidateIndicatorRow[]>();
for (const ind of candidateIndicatorsRes.rows) {
const arr = indicatorsByMessage.get(ind.message_id) ?? [];
arr.push(ind);
indicatorsByMessage.set(ind.message_id, arr);
}
for (const candidate of tier2Candidates.rows) {
if (normalizeSubject(candidate.title) !== normalizedSubject) continue;
const candidateIndicators = indicatorsByMessage.get(candidate.message_id) ?? [];
const candidateSender = candidateIndicators.find(
(i) => i.indicator_type === 'sender'
)?.value;
if (candidateSender !== ownSenderValue) continue;
const candidateHashes = candidateIndicators
.filter((i) => i.indicator_type === 'attachment_hash')
.map((i) => i.value);
const candidateDomains = candidateIndicators
.filter((i) => i.indicator_type === 'url')
.map((i) => extractUrlDomain(i.value))
.filter((d): d is string => d !== null);
const hashMatch = ownAttachmentHashes.some((h) => candidateHashes.includes(h));
const domainMatch = ownUrlDomains.some((d) => candidateDomains.includes(d));
if (hashMatch || domainMatch) {
matchCampaignId = candidate.campaign_id;
matchGroupMethod = 'attachment_or_url';
break;
}
}
}
}
}
// ---------------------------------------------------------------------
// Tier 3: normalizeSubject(title) + company + 24h window (D-02, fixed
// per file-level bug-fix note). Company-wide — deliberately NOT scoped
// to `reports.requester_contact_id` — so the same campaign reported by
// different employees at the same company still consolidates into one
// campaign. Self-exclusion (`r.id != $2`) required for the same reason
// as Tiers 1-2.
// ---------------------------------------------------------------------
if (!matchCampaignId && ownReport.company_id && normalizedSubject) {
const tier3 = await client.query<{ campaign_id: string; title: string | null }>(
`SELECT r.campaign_id::text AS campaign_id, r.title
FROM reports r
WHERE r.company_id = $1
AND r.campaign_id IS NOT NULL
AND r.id != $2
AND r.created_at BETWEEN $3::timestamptz - INTERVAL '24 hours'
AND $3::timestamptz + INTERVAL '24 hours'
ORDER BY r.created_at ASC`,
[ownReport.company_id, reportId, ownReport.created_at]
);
const match = tier3.rows.find((r) => normalizeSubject(r.title) === normalizedSubject);
if (match) {
matchCampaignId = match.campaign_id;
matchGroupMethod = 'sender_subject_client';
}
}
// ---------------------------------------------------------------------
// Own-signal tier keys, computed once here and reused by both the
// own-campaign revalidation guard (CR-03, below) and the create-new
// fall-through further down — avoids computing them twice.
// ---------------------------------------------------------------------
const tier1Key = computeTier1Key(ownMessage?.message_id ?? null);
const tier2Key = computeTier2Key(
ownAttachmentHashes,
ownUrlDomains,
normalizedSubject,
ownSenderValue
);
const tier3Key = computeTier3Key(normalizedSubject, ownReport.company_id);
const currentKeys = [tier1Key, tier2Key, tier3Key, `report:${reportId}`].filter(
(k): k is string => k !== null
);
// ---------------------------------------------------------------------
// Find-or-create against `campaigns` (CAMP-02).
// ---------------------------------------------------------------------
if (matchCampaignId && matchGroupMethod) {
// CR-01/CAMP-02 no-op guard: tier queries only exclude the report's
// OWN row (`r.id != $x`), not its siblings already in the same
// campaign. Since `/analyze` re-runs grouping unconditionally
// (D-08), a report already linked to campaign X can re-match X via
// a sibling row on every repeat call, inflating `report_count`
// without bound. If the match resolves to the report's own current
// campaign, it's already correctly linked — skip both UPDATEs.
if (matchCampaignId === ownReport.campaign_id) {
return { campaignId: matchCampaignId, groupMethod: matchGroupMethod, created: false };
}
await client.query(
`UPDATE campaigns
SET report_count = report_count + 1, last_seen_at = NOW(), updated_at = NOW()
WHERE id = $1`,
[matchCampaignId]
);
await client.query(
`UPDATE reports SET campaign_id = $1, updated_at = NOW() WHERE id = $2`,
[matchCampaignId, reportId]
);
// CR-02: if the report had a DIFFERENT prior campaign (the
// same-campaign case returned above; a never-grouped report has no
// origin — D-08), decrement it so it doesn't keep a permanently
// stale report_count for a report it no longer holds. Scope:
// decrement only, never delete the origin row even if its count
// reaches 0 (deleting introduces FK-cascade/detail-route risk out of
// scope for this gap fix; an empty campaign is a strictly lesser
// problem than a stale count).
if (ownReport.campaign_id) {
await decrementOriginCampaign(client, ownReport.campaign_id);
}
return { campaignId: matchCampaignId, groupMethod: matchGroupMethod, created: false };
}
// ---------------------------------------------------------------------
// CR-03: no sibling matched on any tier (matchCampaignId is null).
// Every tier query above self-excludes the report's own row
// (`r.id != $x`), so a single-report campaign (no sibling report
// exists yet — the common case for any brand-new phishing lure before
// a second person reports it) always reaches here with matchCampaignId
// still null. Pre-fix, control fell straight into "create new
// campaign" below, abandoning the report's still-valid campaign and
// producing a SECOND campaigns row with the identical campaign_key on
// every repeat /analyze call. This guard sits OUTSIDE the sibling-
// match branch above precisely because it must only run when NO
// sibling matched — it never suppresses a legitimate D-08 sibling
// upgrade, which returns from the branch above instead.
// ---------------------------------------------------------------------
if (!matchCampaignId && ownReport.campaign_id) {
const ownCampaignRes = await client.query<{ campaign_key: string; group_method: string }>(
`SELECT campaign_key, group_method
FROM campaigns
WHERE id = $1`,
[ownReport.campaign_id]
);
const ownCampaign = ownCampaignRes.rows[0];
if (ownCampaign && currentKeys.includes(ownCampaign.campaign_key)) {
// The report's own campaign still satisfies its freshly-recomputed
// current tier key — it's still validly linked, reuse it as-is.
return {
campaignId: ownReport.campaign_id,
groupMethod: ownCampaign.group_method as GroupReportResult['groupMethod'],
created: false,
};
}
// CR-02: own campaign row is gone, or the report's signal has
// genuinely diverged from its stored key — this report is about to
// be abandoned from its origin campaign in favor of a brand-new one
// below (the same class of abandonment as the sibling-migration
// branch above, just landing on a new row instead of an existing
// sibling's). Decrement now so the origin doesn't show a stale
// count. Falls through to create-new unchanged.
await decrementOriginCampaign(client, ownReport.campaign_id);
}
// No match on any tier — create a new campaign. Prefer the strongest
// available key (Tier 1 > Tier 2 > Tier 3) so future duplicates of THIS
// report can match it; do NOT set/transition campaigns.status (leave
// migration default 'open' — Phase 19/20's concern).
let newCampaignKey: string;
let newGroupMethod: GroupReportResult['groupMethod'];
if (tier1Key) {
newCampaignKey = tier1Key;
newGroupMethod = 'message_id';
} else if (tier2Key) {
newCampaignKey = tier2Key;
newGroupMethod = 'attachment_or_url';
} else if (tier3Key) {
newCampaignKey = tier3Key;
newGroupMethod = 'sender_subject_client';
} else {
newCampaignKey = `report:${reportId}`;
newGroupMethod = 'sender_subject_client';
}
const newCampaign = await client.query<{ id: string }>(
`INSERT INTO campaigns (campaign_key, group_method, first_seen_at, last_seen_at, report_count)
VALUES ($1, $2, NOW(), NOW(), 1)
RETURNING id::text AS id`,
[newCampaignKey, newGroupMethod]
);
const newCampaignId = newCampaign.rows[0].id;
await client.query(
`UPDATE reports SET campaign_id = $1, updated_at = NOW() WHERE id = $2`,
[newCampaignId, reportId]
);
return { campaignId: newCampaignId, groupMethod: newGroupMethod, created: true };
});
} catch (error) {
console.error('[CAMPAIGN-GROUPING] Failed to group report into campaign', reportId, error);
throw error;
}
}