103 mailboxes, 2.66 M messages, 1.43 M threads — full org coverage. support@ 112K, accounts@ 77K, hr@ 72K, collabs@ 61K, cs@ 49K — the functional aliases are the densest archives.

1. Coverage is the entire org — 103 mailboxes, 2.66 M messages, 1.43 M threads. Every individual contributor mailbox plus every functional alias captured at the same moment. The biggest individual mailbox runs 238K messages; the second runs 236K. There's no missing perspective.
2. The functional aliases are exactly as dense as expected. support@contigo.com 112K messages, accounts@contigo.com 77K, hr@contigo.com 72K, collabs@contigo.com 61K, cs@contigo.com 49K, careers@contigo.com 47K, admin@contigo.com 24K, facility@contigo.com 1.4K. These shared inboxes are the departmental archive — far denser than per-individual mailboxes.
3. The same person identities show up across Gmail, Drive, Calendar, and Zoho. Pivot from 'who answered this brand' to 'what files they shared with that brand last week' to 'what calls they have scheduled' — the whole footprint of an individual is reconstructable across tools.

76.8% of threads are single-message; only 0.3% (~4,375 threads) are 31+ messages — but that small slice is where the substantive customer / negotiation arcs live.

1. 1.43 M threads with their length distribution exposed directly. 76.8% are single-message threads (notifications, announcements, automated mail). 15.6% are 2-3 messages (quick exchanges). 6.0% are 4-10 messages. 1.3% are 11-30 messages. 0.3% — about 4,375 threads — are 31+ messages: active negotiations, customer-incident war-rooms, and brand-relationship arcs. The deep-tail threads are where the substantive business actually lives.
2. Each thread already carries the org's own labels. What people called 'Sales' or 'Brand-XYZ' or 'Recruiting' or 'Customer-Support' is right there on the thread. You inherit a categorization that took years of mailbox curation.
3. Threads are the right grain for sales analysis — one negotiation, one thread. Aggregating at thread level instead of message level matches the unit of buyer-decision-making. Median thread length per industry, gaps over 30 days that signal dormant relationships — all visible.

2.67 M messages spanning six years — exact brand-correspondence volumes per customer, 50+ named campaign hashtags, peak email day 2025-04-17 (4,148 emails).

1. Reconstruct any major customer relationship in pure email — exact correspondence volumes per brand. Top counts (sender or recipient): Brand-08 parent brand-77.com 21K, Vendor-03 20K, Brand-15 18K, Brand-23 16K, Brand-09 12K, Brand-10 9.5K, Brand-04 9.5K, Brand-03 8K, Brand-07 8K, Brand-26 3.9K, Brand-35 3.2K, Brand-25 3K, Brand-38 2.7K, Brand-41 2.5K, Brand-21 1.8K. Every major customer's negotiation transcript sits in this corpus.
2. Subject-line hashtags name 50+ specific campaigns by volume. #Brand-95-vedic (1,008), #Brand-91 (786), #Brand-92-women (660), #Brand-93-amigo-campaign (491 — Brand-93), #Brand-94 (381), #Brand-23-v (311), #Brand-23-weekend (294), #Brand-08-festival (294), #Brand-09-campaign (277), #Brand-04 (273), #Brand-01-creators (250), #Brand-08-birthday (238), #Brand-10-5G-campaign, #Brand-08-launch, #Brand-14-MegaJune25. Each hashtag opens the door to that campaign's full email arc — brief, proposal, execution.
3. The peak email days line up with the major events captured elsewhere in the corpus. All-time peak: 2025-04-17 (4,148 emails). The Meta-enforcement day 2025-01-22 saw 2,903 emails — well above the daily baseline — matching the Confluence response-page drafting that day. Heaviest email stretches are Jul-Oct 2023 and Apr-Jul 2025 — the same windows as the largest engineering stress events. 36.6% of all messages are RE: replies, 25.1% carry attachments.

2,156 label entries tagging 7.4 M messages. ~251 of them are individual-employee-name folders — sales managers organize their inboxes by the people reporting to them.

1. 2,156 label entries in total, tagging 7.4 M messages collectively. 1,574 are uppercase system labels (INBOX, SENT, IMPORTANT, STARRED, DRAFT, TRASH, SPAM). The rest are user-defined topic / brand / workflow labels (Pending, Completed, Followup, Platform-01, BD- Internship, Travel Desk, Tax invoice, etc.) plus a meaningful tail of person-name labels.
2. About 251 labels look like individual-employee-name folders. A strong cultural signal that managers organize their inboxes by the person who reports to them — rather than by topic. This 'mail-from-this-person' folder pattern is common in Indian sales orgs and shows up clearly in the data.
3. System labels combine with custom labels to encode urgency. A thread tagged 'Sales' AND 'IMPORTANT' AND 'STARRED' carries a much stronger priority signal than label inference would suggest. The user's own urgency model is already in the data.

104 send-as identities + 103 vacation responders captured — every individual has tuned their inbox identity at least once.

1. The settings table captures 104 send-as identities, 103 vacation responders, and 4 auto-forwarding rules across the org — roughly one configured setting per mailbox. Every individual has tuned their inbox identity at least once.
2. Send-as identities surface multi-persona reps. Sales reps frequently send 'as' a brand-team alias; without seeing this, the same physical message looks like it came from two different identities. The send-as table is the alias-to-person mapping the rest of the email data joins through.
3. Forwarding rules and vacation responders are tenure breadcrumbs. Long-term vacation responders, mailbox-forward-to-manager rules, out-of-office templates surface departures, leaves, and continuity patterns that would otherwise need HR data.

The contracts, proposals, and decks that flow through email — invisible to anyone with only a Drive view.

1. Pitch decks, contracts, and invoices live in attachments — not Drive. Many B2B handoffs (proposal PDFs, PO scans, signed agreements) are emailed and never landed in Drive. This is the only place to see them.
2. The mime mix per mailbox is a workflow fingerprint. Sales and legal mailboxes are PDF and Word heavy (proposals, NDAs); finance and accounts run on .xlsx (invoices, ageing reports); marketing mailboxes carry images (brand assets, screenshots).
3. Attachment timing inside a thread suggests deal stage. A thread with one PDF mid-conversation is usually 'proposal sent'; a later second PDF (the signed counterpart) usually means 'closed'. You can sense deal-stage transitions without reading any body.

Two named shared drives — Operations (sales-ops) and Gen AI App Builder Client Libraries (engineering) — the split mirrors the company's two worlds.

1. Two top-level shared drives capture two distinct operating modes. Operations is the sales-ops + customer-delivery hub — brand contracts, post-campaign rosters, creative assets. Gen AI App Builder Client Libraries is engineering's shared workspace. The two-drive split is the first piece of architecture you see in the data.
2. Personal drives layer behind the shared drives. Personal drives hold work-in-progress and pre-shared drafts; shared drives hold final approved artifacts. You can tell where a document is in its lifecycle from where it lives.
3. About 87% of files are owned by @contigo.com users; the rest belong to outside partners. That outside slice is a measurable footprint of vendor staff embedded inside the workflow — the kind of operating-model signal you can't read off any org chart.

Image dominates by count (44.1%), video dominates by size (77.6% / ~2.2 TB) — creator-economy raw material plus a thicket of campaign briefs, contracts, and post-campaign reports.

1. The corpus is image-heavy by count, video-heavy by size — both are real signals. Image files are 44.1% of the count but only 10.6% of the size. Video is 6.1% of the count but 77.6% of the total size (~2.2 TB of mp4/mov/mkv). Then PDFs (16.1% of count), Sheets (13.3%), Docs (5.1%). The corpus is fundamentally a creator-content delivery archive (videos are the deliverable a brand pays for) plus a thicket of campaign images, brand briefs, contracts, and post-campaign reports.
2. About 87% of files are owned by @contigo.com users; the rest are external partners with elevated ownership. Top external owner-domains: vendor-05.com (1.3%, ~7K files — the office partner with embedded ops), vendor-10.com (0.5%), vendor-11.com (0.3%), and notably university-01.in + university-03.edu (universities — likely intern programs). The owner-share itself is a measurable footprint of vendor-embedded operating model.
3. Top folder names give you the campaign-asset map. Brand-67 | Business Party - shots (6,912 files), Brand-17-event-stories (5,856), Creator insights Ext-Brand-26 (1,851 — the Brand-26-specific folder). Plus monthly buckets (10). OCT 2025, 13). JAN 2026) showing how delivered campaigns are organized chronologically.

The directory tree alone is a customer atlas — /Sales/Clients/<brand>/... lets you walk the corpus customer by customer.

1. The directory tree is a customer atlas. Folders organized by /Sales/Clients/<brand>/Active Campaigns/... and /Operations/... — every brand client has its own folder lineage. You see the customer footprint without ever touching CRM.
2. Folder-modification timestamps are an account-health signal. Recently-touched brand folders flag active accounts; long-stale folders flag dormant ones — quietly visible, no CRM query required.
3. Folder permission patterns reveal which folders are 'hot'. Folders shared to many external users are typically the customer-delivery pickup points; folders kept private hold contracts and financials. The sharing breadth per folder is the sensitivity tell.

Top external grantees with exact share — Brand-25, Vendor-03, Vendor-10, Vendor-11, Brand-26, Brand-03, Brand-23. Universities (University-02, PES, University-01) also show up — intern programs measurable in the data.

1. The external sharing graph is led by personal gmail.com addresses (15.3% of all permissions) — likely creator personal emails. Then named brands and partners: vendor-05.com 22.8K (0.7%), brand-25.com 19.5K (0.6%), vendor-03.com 10.3K (0.3%), vendor-10.com 7.1K, vendor-11.com 5.6K, brand-26.com 5.0K, brand-03.com 3.3K, vendor-07.com 2.2K, vendor-06.com 1.7K, brand-23.com 1.3K, investor-02.com + investor-01.com 1.3K each (investors).
2. Universities show up in the sharing graph — the intern-program footprint. university-01.in (~1K perms), university-02.ac.in (798), university-03.edu (448) all have hundreds of permission grants — sustained collaboration with engineering colleges, not one-off campus visits.
3. Permission roles partition the sharing intent. Writer 2.03M, Owner 627K, Reader 393K, Commenter 91K. The writer-heavy distribution reflects collaborative editing — creator content is co-owned, not just published. Plus 290K anyone-type grants (public-link sharing) and 50K domain-level grants (whole-org access for partners).

Every save, every editor, every minute — the canonical productivity timeline for sales-ops.

1. The most granular productivity timeline in the whole corpus. Every save on every file is stamped with the user who made it. Per-person editing cadence, per-document edit density, and per-team productive-hour heatmaps fall out naturally.
2. Living documents and sealed documents become obvious. A file with hundreds of revisions across many editors is an SOP that's actively maintained; a file with one revision and one author is a contract or invoice, sealed at creation. You can tell which is which without opening anything.
3. Time-of-day editing patterns reveal organizational tempo. Sales-ops revisions cluster 9am–6pm IST; engineering revisions stretch into the evenings. The heatmaps per department validate office-vs-remote-vs-hybrid working patterns without an HR survey.

Comment threads where authors explicitly write 'approved', 'budget cleared', 'go ahead' — the smallest unit of business decision in the corpus.

1. The smallest grain of business decisions. People write 'approved', 'budget cleared', 'go ahead', 'not yet', 'see legal' in comments — these are unambiguous decision records, often missing from the doc body itself.
2. Resolved-vs-unresolved comments tell you which documents are still in flight. A doc with many open comments is mid-review; a doc with all comments resolved is finalized. Per-folder open-comment density is a real-time view of work in motion.
3. Per-author comment patterns reveal the informal review hierarchy. Senior reviewers comment broadly across many folders; junior contributors comment within their own scope. The comment graph reveals de-facto seniority better than job titles.

Storage usage per user separates heavy-document creators from light consumers — a visible role indicator.

1. Storage usage per user is a visible role indicator. Sales-ops and creative-ops users typically run heavy (video deliverables); engineers run light (text-heavy work); executives are highly variable. The usage histogram exposes which roles handle which content types.
2. Trash storage separates hoarders from clean operators. Heavy trash users churn through drafts; low-trash users are deliberate creators — a small, clean style indicator.
3. The connected user list is the spine of the Workspace identity graph. Each user shows up in Drive, Gmail, Calendar, and Zoho with the same identity — so an analysis of one person's footprint can span every tool they touched.

At least a dozen brand-named teams — Brand-48, Brand-49, Pod-NN, Brand-04 execution, Brand-08 Install (×3), Brand-45/Brand-46/Brand-47, Brand-42, Brand-43, Brand-44 — reveal a deeply embedded brand-pod operating model.

1. At least a dozen brand-named teams reveal a dedicated-team-per-brand operating model. Brand-48, Brand-49, Pod-NN, Brand-04 execution, Brand-08 Install, Brand-08 sub-team-NN, Brand-08 Internal, Brand-45/Brand-46/Brand-47, Brand-42 <> Contigo, Brand-43-X-Contigo, Brand-44, Pod-NN — each a dedicated pod for one customer engagement. The brand-pod model isn't a one-off; it's how the company operates.
2. Contigo_Ops_Celeb is the revenue-celebration channel — itself a cultural artifact. Posts in this team typically reference closed-won deals, named brand wins, and rep-level achievements. The celebration messages amount to an org-self-reported revenue-event log, often well ahead of the public-facing reporting.
3. The visibility split (Private 35 / Public 15) tells you what's gated. Brand-specific work, leadership channels, and live-issue rooms are private. Cross-functional channels like Data Team, Contigo, Leads, Dev_Alerts, Prod_Alerts are public. The customer-confidentiality posture is visible in the channel-type distribution alone.

Go live Campaign is the launch-coordination war room; brand-named channels often appear weeks before the first Zoho deal closes.

1. Go live Campaign is the launch-coordination war room. Every product launch — the Brand-09 Brands App go-live on June 11 2025, the Brand-26 app kickoff in late June 2025, the Brands App user-management rebuild in May 2025 — is coordinated in this channel. The message archive is a working operations playbook for B2B SaaS launches.
2. Brand-named channels often appear weeks before the first Zoho deal closes. When a private channel exists for a brand, that brand has cross-functional commitment from the company — and the channel typically pre-dates the deal by a couple of weeks. The channel inventory becomes an early-warning signal for incoming pipeline.
3. Public vs. private channel split shows what's open and what's gated. Public channels carry cross-functional discussion; private channels gate brand-specific or sensitive workstreams. Customer-confidentiality posture is visible in the channel-type distribution per team.

Six years of internal nervous system — including the Brand-09 launch warroom and customer-delight escalations, day by day.

1. The internal nervous system of the company for over six years. The chatter concentrates in Customer Delight, Go Live Campaign, HR, and the per-brand channels — the running operational discussion that surrounds every customer engagement, every launch, every escalation.
2. You can reconstruct the Brand-09 launch day by day. The June 11 2025 Brand-09 go-live shows up in Go live Campaign as a sequence of coordinated messages across product, engineering, and sales. The post-launch defect spike — a wave of P1 bugs over the following two weeks — appears as escalation messages with explicit references to the Jira tickets. A complete real-world incident-response transcript.
3. Reply trees and message types reveal decision dynamics. Threaded conversations are intact, and the type field separates regular messages from system events (member added, file shared, call started). You can see which decisions got debated in long reply trees vs. which announcements got no replies at all.

The back-channel where exec decisions and cross-team bridges happen — usually invisible to outside analysts.

1. The back-channel where exec decisions actually happen. 1:1 and group chats sit outside the formal team-channel structure. Cross-functional bridges, ad-hoc decisions, and exec sync that bypasses the documented record happen here — usually invisible to outside analysts.
2. Group-chat topics name purpose explicitly. Many group chats are titled after a project or initiative — 'Brand-09 onboarding', 'Brand-26 scoping' — a parallel taxonomy of in-flight work that complements the formal channel structure.
3. Meeting chats preserve conversation tied to specific calls. Every Teams meeting that produced chat output (Q&A during the call, links shared mid-meeting) leaves a record that's tied back to the calendar event — meeting context recoverable months later.

Membership patterns that reveal informal coordination networks the team-channel structure misses.

1. The informal collaboration network is visible. Who chats with whom outside formal team boundaries — aggregate the pairings and you have a cross-functional network that's far more honest than the org chart.
2. Owners vs. members distinguish coordination leads from participants. Group-chat owners are de-facto coordination leads, even when they hold no formal title — that becomes measurable here.
3. Membership-start timestamps capture late arrivals. A person added halfway through a chat's life is a late-arriving stakeholder; useful for tracing when decisions changed hands.

Who pulls who into which decisions — observable as an explicit attention graph.

1. The @-mention graph is the explicit attention layer of the org. Whoever gets pulled into conversations far more than peers is a hub — typically a sales lead, a customer-delight lead, or a senior engineer. A clean measure of de-facto leadership independent of title.
2. Team-mentions and user-mentions tell you culture. @TeamName mentions are announcements; @PersonName mentions are direct asks. The ratio per team is a culture indicator — broadcast vs. individual accountability.
3. Mention bursts coincide with customer events. Spikes of mentions in Go live Campaign map to launch days; spikes in Customer Delight map to brand escalations. Plotted against the customer-event timeline, mention frequency is a self-validating dataset.

Emoji reactions — the lowest-friction approval signal — reveal which announcements got endorsed vs. merely seen.

1. The lowest-friction approval signal in any chat tool. People react to messages they don't reply to. A senior-leader announcement that gets many 'celebrate' reactions but zero replies is a strong endorsement signal — invisible without reaction data.
2. Reaction-type breakdown reveals cultural temperament. Mostly 'like' / 'thumbs-up' = formal acknowledgement culture. Heavy 'heart' / 'celebrate' / 'fire' = high-engagement culture. Rare 'sad' / 'angry' / 'confused' reactions flag posts that hit a nerve.
3. Per-user reaction patterns are a personal-engagement fingerprint. Some people react to every announcement; others only on milestones. Reaction frequency by role distinguishes engaged team members from disengaged ones.

The handoff layer — files and links shared inline in chat that bypass Drive and Jira entirely.

1. The cross-tool handoff layer. When someone shares a Figma link in Go live Campaign, a Jira ticket in Customer Delight, or a Drive deck in HR — this is where it's captured. The 'I'll just share this in the chat' moments that actually move work along.
2. Content type tells you which teams operate by-link vs. by-file. Engineering-leaning teams share Jira and Bitbucket links inline; brand teams share videos and PDFs. The mix per team is a tooling fingerprint.
3. Per-author attachment frequency separates information sharers from consumers. A high attachment-share rate is typical of leads and coordinators; a low share rate is typical of individual contributors. A role indicator that survives even when titles are missing.

Teams-native meetings with subject lines that name the customer or initiative.

1. Online-meeting share is a remote-work signal you can compute exactly. Teams-native meetings are flagged; everything else is third-party or in-person.
2. Subject lines name customers and initiatives. 'Brand-10 QBR', 'Brand-26 sync', 'Brand-09 launch readiness' — when a brand name shows up as a meeting subject, that brand has scheduled stakeholder commitment. Meeting cadence per customer is plain to read.
3. Per-organizer meeting load is a workload signal. Some users organize many meetings per week (coordinators, sales leads); others organize none (individual contributors). Sustained high meeting-organizing rate is a measurable attention-fragmentation signal.

The most profile-rich employee directory in the corpus — job title, department, office, city, country populated.

1. The most profile-rich employee directory in the corpus. Job title, department, office location, city, country — the kind of HR-grade detail that's usually missing from collaboration-tool exports. CEO, AVP-Finance, Sales Lead, Customer Delight Lead are all visible.
2. Department coverage spans the full functional org. Engineering, Product, Customer Delight, HR, Data Intelligence, Sales, Finance, Security — each represented with member counts that approximate true team size. An org-design dataset without ever doing a survey.
3. The active-vs-deprovisioned flag separates current employees from departed ones. Current-employee analyses, historical-team analyses, and tenure analyses all become possible from one user list.

Customer-named projects (Brand-83-key for Brand-83), security incidents (CSI), and a revenue-target initiative literally named M8C for Mission 8 cr.

1. Customer-named projects expose dedicated tracks. Brand-83-key is the Brand-83 (Brand-83) engagement; M8C is Mission 8 cr — a revenue-target initiative literally codified into the work-tracker. Project keys are an org's most honest statement of strategic priority.
2. Service-desk projects (SUP, CSI) are the customer-facing surfaces. Tickets in Support are external requests; tickets in Security Incidents are real-world security events. The internal vs. external work split is right there in the project type.
3. Project-level metadata reveals process maturity. Some projects use software style with sprints; others are product_discovery (lightweight); service-desks have queue-based workflows. You can see which parts of the org practice mature agile vs. lightweight workflow.

Sept 2023 was the biggest defect spike — 165 P1 / Critical bugs in one 2-week window, more than any later release including Brand-09. Brand-tagged epics anchor customer work; Brands App is the most-epic'd surface.

1. The biggest defect spike in the dataset was September 2023. 165 P1 / Critical bugs were filed in 2023-09-W2 alone, followed by 105 in 2023-10-W1 and 75 in 2023-09-W1. The Aug-Sep-Oct 2023 stretch is the most stress-tested engineering period the data captures — even bigger than the June 2025 Brand-09 launch wave.
2. The Brands App is by far the biggest internal product investment. Epic bracket-tags: [Brands App] (6 epics), [Backend] (2), [Creators App] (2), [Profile] (2), [Onboarding Module] (2), plus singletons including [Brand-26] (the Brand-26 customer epic ET-4440), CLCM, OCB, Payment Module, Social Cashback. The brand-side product gets multiple tracked epics; the Brand-26 customer build is a single epic.
3. The project mix shows what the org tracks. Issue distribution: ET (Engineering) 44.0%, CAAA (Contigo Application archive) 24.3%, KA (Contigo app) 12.6%, K2AA (2022 archive) 6.2%, DIT (Data Intelligence) 5.3%, then PK / RCT (Brand-104) / CZAA / PROD / PT / M8C (Mission 8 cr revenue initiative) / PEDE / Brand-83-key (Brand-83 engagement). Three explicit archives suggest disciplined project-history retention.

The substantive design discussion behind every issue — where decisions actually got debated, not just announced.

1. The substantive design discussion that no other artifact preserves. Commit messages are terse; PR titles are coarse; the actual debate — 'should we cache this at the edge?', 'who owns the auth migration?' — happens in Jira comments. The densest source of engineering decision rationale in the corpus.
2. Comment density flags the issues worth analyzing. High-comment issues concentrate in customer-onboarding epics (Brand-09, Brand-26) and post-launch defect spikes. The high-comment subset captures most of the substantive engineering discussion — a built-in signal-to-noise filter.
3. Per-author commenting frequency identifies the de-facto technical reviewers. Engineers commenting heavily across many issues are senior reviewers; sparse commenters are deep-focus individual contributors. Reveals technical leadership independent of formal org structure.

The complete record of what happened on every issue and when.

1. The complete record of what happened on every issue. Every status transition, assignee change, priority bump, sprint move — timestamped to the actor. The full history of any ticket is recoverable.
2. Sprint velocity reads off the data directly. Throughput per sprint, per team, per quarter — the engineering-management metrics every CTO wants — are immediate. Per-team variation reveals which squads ship vs. which struggle.
3. Re-assignment churn is a managerial-quality signal. Issues with many assignee changes typically signal scoping problems or capacity issues. The churn distribution per epic is a leading indicator of project trouble — visible months before the symptoms show up in product.

Self-reported engineering effort attached to issues — actual time spent per customer-tagged epic.

1. Self-reported engineering effort, attached to the customer it served. Worklogs aren't just timestamps — they're the only way to assign actual cost to customer-tagged work. The Brand-26 epic and the Brands App user-management rebuild each have their own worklog totals; engineering cost per customer is directly visible.
2. Worklog comments often capture work substance better than commit messages. Engineers tend to write what they did before logging time. When you want to know what actually got built, the worklog text is the more reliable source.
3. Per-person hours-logged distribution flags both over- and under-trackers. Engineers logging consistently high weekly hours are at capacity; consistently low loggers are likely under-tracking. A useful data-quality signal when reasoning about productivity.

Sprint goals as free text double as a customer-narrative timeline.

1. Sprint goals are a customer-narrative timeline. Goal text often references the customer or initiative driving the sprint — 'Brand-09 onboarding stabilization', 'Brands App user-mgmt rebuild' — read chronologically, the sprint goals tell a self-narrated story of engineering priorities.
2. Active vs. closed sprint distribution measures process discipline. Sprints lasting well past their stated end date are red flags; the duration-overrun rate per team is a real agile-process-health metric.
3. The gap between stated sprint goal and what actually closed is measurable. When the goal mentioned customer X and the closed issues are mostly tech-debt cleanup, you have a plan-vs-execution gap — repeatable, observable across teams.

The blocking-and-relating graph behind every release wave — including the multi-repo Brand-09 launch.

1. The critical path of any release reads directly. blocks, is blocked by, relates to, duplicates — compose them and the multi-repo Brand-09 launch's blocking dependencies become a real graph. Every release wave's risk is visible.
2. Epic-to-sub-task hierarchies are walkable. ET-4440 (Brand-26 app) links out to a tree of dependent stories and bugs; the full implementation plan is recoverable as a connected sub-graph.
3. Direction is preserved. The fact that A blocks B (not B blocks A) is in the data — critical-path inference doesn't have to guess at directionality.

Per-issue screenshots, logs, and design refs — most of the substance lives here, not in body text.

1. The attachment is usually where the bug actually lives. Screenshots for bug reports, log files for incidents, design exports for stories, spreadsheets for data tasks — most issue context is here, not in the body text.
2. Bugs almost always have a screenshot; story tickets often have none. The attachment ratio per issue type is a quick read on bug-report discipline.
3. Per-author upload patterns separate roles. Designers upload Figma exports; QA engineers upload screen-recordings; PMs upload sheets. Useful when the display name doesn't make role obvious.

Bot-to-human ratio reveals a tool-integrated engineering org.

1. The bot-to-human ratio reveals tool-integration maturity. App accounts (Bitbucket-to-Jira automation, Teams-to-Jira webhooks, custom workflow apps) make up a substantial share — a measurable proxy for how integrated the engineering toolchain is.
2. Service-desk customer accounts are separate from engineering staff. Support and CSI projects spin up external customer accounts; engineering projects use only internal staff. The two are clearly distinguishable.
3. Active vs. former staff are both preserved. Former employees still appear in old issues but can be filtered out for current-org analyses. Both subsets are useful, depending on the question.

Org-specific fields reveal what dimensions Contigo's project-management discipline actually tracks beyond standard Jira.

1. Custom fields reveal what the org actually values tracking. Story Points, Epic Link, Customer-tag, Severity, Sprint Cohort — what's defined here tells you which dimensions matter to the project-management discipline at Contigo.
2. Story Points + Sprint linking + Customer-tagging together signal a mature agile org. Absence would point to a less-evolved Jira deployment. The custom-field set is itself a maturity score.
3. The structure preserves history through reorganizations. Fields can be added or deprecated over the years without breaking historical lookups — old issues retain whatever metadata existed at the time they were created.

A first-class Security Incidents space (CSI) at the same level as Product and Engineering — incident response is structurally institutionalized here.

1. A first-class Security Incidents space (CSI), at the same level as Product and Engineering. Not a buried page; a top-level space. Its very existence tells you the org operationalized incident response. Mirrored by a Jira project of the same name — incident response is institutionalized in two tools.
2. Sometimes the most sensitive content lives in personal spaces. The January 2025 Facebook-policy response pages were drafted in a personal space, not a global one. Where confidential drafting actually happens is itself a useful audit-finding.
3. Global spaces map cleanly to functional teams. Product, Engineering, Product Ops, Data Intelligence, Mobile App Revamp, Support, Security Incidents — the space taxonomy is a self-described functional org chart, validated by what shows up in MS Teams.

High-signal documentation — operational SOPs, design docs, and a real January 2025 platform-policy crisis response captured as it happened.

1. The most-iterated page in the corpus is User Stories | Brand Interface — 203 versions over a year. The runner-up is User Stories | Creators Interface at 154 versions. Then User Story - CLCM (52), Vanity URL Enhancement (50), User Stories - Profile Module (32), Fuzzy Search (31), Payment System - Influencer (26). The version counts per page tell you which product-spec docs the team actually treats as living.
2. The compliance-response pages are a real crisis-management artifact. Contigo Response to violations 3.b.i-iii and two related compliance pages were drafted in response to a major platform's enforcement event. Crisis comms preserved as it happened.
3. Pages are deliberate, not a meeting-notes dump. They concentrate in operational SOPs (sales playbooks, onboarding runbooks, payment ops, brand-call procedures), product/engineering specs, and security-incident records (CSI). Compared to the Jira ticket volume, Confluence is signal-dense — what's there was meant to be there.

Comment timeline doubles as a decision timeline — bursts of comments coincide with customer launches and crisis events.

1. The comment timeline doubles as an event timeline. Bursts of comments around a single date often coincide with external events — a customer launch, a crisis response, an executive-driven pivot. The comments are a self-validating record of org attention.
2. Inline vs. footer comments distinguish granular review from high-level reaction. Inline comments target a specific phrase; footer comments are general acknowledgement. Both styles are preserved.
3. Comment density per page flags the pages worth reading. A heavily-commented page is one where multiple stakeholders aligned (or fought); a 0-comment page is a one-off.

Inline checkbox tasks — lightweight workflow that bypasses Jira entirely.

1. Inline checkbox tasks are workflow that bypasses Jira entirely. Action items 'too small to ticket' but 'too important to forget' get embedded directly in pages — workflow data hiding inside docs, usually invisible to project tooling.
2. Per-assignee outstanding-task count is a real workload signal. An engineer with many open Confluence tasks plus open Jira issues has a different total load than someone with just Jira. Including doc-tasks gives a more honest workload picture.
3. Doc-task completion ratio per page is a doc-health metric. Pages with high task-completion are mature; pages with stale open tasks are abandoned.

Where doc bodies are sparse, attachments carry the actual content — embedded specs, exported decks, vendor PDFs.

1. Where the page body is sparse, the attachment carries the actual content. Specs reference attached wireframes; vendor briefs reference attached PDFs; training docs reference attached decks. Without the attachment, those pages look empty.
2. Attachment mime mix per page is a page-type classifier. PNG-heavy = wireframes and specs; PDF-heavy = vendor briefs or compliance docs; .pptx-heavy = exec decks.
3. Pages with no attachments are usually body-only; pages with many attachments are typically design or training material. A quick 'is this a real artifact?' check.

Engineers and PMs over-represented as authors; sales reads but rarely writes Confluence.

1. Confluence accounts skew toward writers, not readers. Engineers and PMs are over-represented; sales rarely authors but consumes heavily. The author distribution tells you which functions value documentation as work.
2. The split between human accounts and app integrations. App accounts handle automated content sync (Jira-to-Confluence links); humans author and comment. Useful when separating 'what was actually written by people' from 'what was synced by a bot'.
3. Public name and display name are both preserved. External-facing display name and internal directory name are kept separately — useful when reconciling identities across tools.

The most honest signal of current page ownership — who actually opted in to be notified when this changes.

1. Watching is the most honest signal of current page ownership. Authorship reflects who created a page; watching reflects who currently cares. The watch graph is a more accurate 'who owns this' than the author field.
2. Watch counts per page are a quick importance read. Pages with many watchers are de-facto org canon; pages with no watchers are stale or abandoned.
3. Per-user watch-set reveals knowledge interests. A person watching many product pages but no sales pages is clearly a product specialist — useful when role classification is fuzzy.

Per-space access rules naming which workstreams (CSI, customer-specific spaces) are deliberately gated.

1. The full governance audit trail across spaces. Who can view, edit, or admin each space — at the user, group, and anonymous-access levels. Any compliance review starts here.
2. The presence of restrictions is itself the security signal. CSI (Security Incidents) and customer-specific spaces have restricted access. Whether they're gated — not the values of the gates — is what tells you the org takes confidentiality seriously.
3. Group-based grants vs. individual grants. Group-based access indicates role-based-access maturity; pages of individual grants indicate ad-hoc access growth — a maturity signal.

Folder titles named after initiatives or customers — the per-launch organizational unit, independent of any one page's lifecycle.

1. Folder titles are organizational atomic units. Folders named after initiatives or customers are persistent organizing structures, independent of any one page's lifecycle. Useful for 'which initiatives have any documentation at all?'-type questions.
2. Created and updated timestamps give a project pulse. Recently-created folders mark new initiatives; long-stale folders mark abandoned ones. A second project-activity signal that complements Jira sprints.
3. Folder hierarchy nests pages. Parent-child folder relationships preserve the org's preferred mental model for organizing content, even when individual pages move around.

Visual brainstorming canvases — pre-document ideation rarely captured anywhere else.

1. Visual brainstorming canvases — pre-document ideation rarely captured anywhere else. Most companies' brainstorming happens on Miro or paper that doesn't persist. These whiteboards capture the moment when ideas became formal enough to document but not yet text.
2. Title, space, and created-at metadata only — not the canvas geometry. Adequate to track 'how often do teams brainstorm formally?' without rendering anything visual.
3. Whiteboard usage concentrates in product and engineering spaces. A signal of which teams ideate visually vs. textually — useful when planning what kind of design tool to invest in next.

Plugin-set metadata that exposes the doc-management toolchain the team actually relies on.

1. Plugin-set typed metadata exposes the doc-management toolchain. Compliance / scaffold / template add-ons leave distinctive property keys; the property-key inventory tells you which tools the team relies on.
2. Pages can be filtered by structured criteria — 'all SOPs reviewed in the last 90 days' — without scanning page bodies. Properties are the structured-data layer beneath the unstructured pages.
3. Property values are typed and stable. Robust input for downstream automation that depends on doc-state being machine-readable (e.g., 'this page was approved by legal').

Page-level overrides flag the sensitive-content hotspots within otherwise-open spaces.

1. Page-level overrides flag the sensitive-content hotspots. Within otherwise-open spaces, individual pages with read or update restrictions reveal where legal, financial, or HR-sensitive content lives.
2. Restricted-page density per space yields a sensitive-content map. You don't have to read pages to know where the sensitivity concentrates.
3. User-level vs. group-level restrictions both captured. Distinguishes 'restricted to a person' (often legacy or ad-hoc) from 'restricted to a group' (governed access).

Plugin-defined non-page artifacts — the structured product-management surface beyond free-form docs.

1. Plugin-defined non-page artifacts surface advanced workflows. Decisions, requirements, glossary entries — when these custom content types exist, the team uses Confluence as a structured product-management surface, not just free-form docs.
2. Their presence indicates tooling-investment maturity. Custom decision or requirement content types imply investment in formal product-management Confluence apps.
3. Same body-storage pattern as pages. Analytical pipelines built for pages work for custom content without modification.

22 naming clusters reveal the full product surface — Brand-104- (7 repos), Brand-26- (5), DIY- (4), data- (4), ai-* (4), plus customer-named tracks. Python dominates 72%.

1. There are 22 repository naming clusters — and each tells you what got engineering investment. The biggest customer-named clusters: Brand-104- (7 repos — a premium creator program), Brand-26- (5 repos — the Brand-26 customer build). Product-line clusters: DIY- (4), data- (4), ai- (4), youtube- (3), social- (3), flink- (3), campaign- (3), Contigo- (3), ContigoGoogleHackathon-* (3 hackathon teams). Walk the cluster list and you see the whole product surface laid out.
2. The languages tell you what kind of engineering organization built this. Python dominates at 72.4% (data services, ML, scrape pipelines). TypeScript 9.2% (web frontends). Dart 6.6% (mobile apps — Brands App + Verve Studio creator app). JavaScript 6.6%, Java 3.9%. The platform is overwhelmingly a Python-data company with a Dart mobile surface, not a web-first SaaS.
3. The biggest single codebases by size are the web apps. Top by repo size: mavin (799 MB, TypeScript), kofweb (704 MB, TypeScript), DIY-POC (407 MB, JS), ocb-web (399 MB, TypeScript), DIY-frontend (256 MB). The largest customer-build is Brand-26_backend at 60 MB plus Brand-26-webapp at 57 MB — material but well below the platform's flagship apps.

Vendor-embedded engineering is exact: Vendor-01 7.8%, Vendor-02 6.3%, Vendor-03 1.1%, B&W Consulting 1.1%, Vendor-05 1.0%. Peak commit day 2023-08-02 (233 commits) — Aug-Sep-Oct 2023 was the most active engineering period.

1. Vendor-embedded engineering is measurable. Author-domain breakdown: Contigo internal 62.9%, gmail.com 18.3% (likely contractors with personal addresses), vendor-01.com 7.8%, vendor-02.com 6.3%, vendor-04.com 1.1%, vendor-03.com 1.1%, vendor-05.com 1.0%. Excluding the gmail share, named external partners contribute ~17% of all commits — a substantial fraction.
2. Six years of commit history with files-changed and line counts. The single biggest commit day was 2023-08-02 with 233 commits, far ahead of any later day. Aug-Sep-Oct 2023 was the most active engineering period in the dataset, predating the Brand-09 launch wave by nearly two years.
3. Top repos by commit volume reveal where the platform's engineering effort actually went. crowdsource (8,658 commits), diy-react-frontend (8,647), ocb-web (7,332), dags (7,225), creators-app (4,798), datamagnet (3,992). Customer-specific repos like Brand-26_backend (513) and Brand-26-webapp (312) are real but moderate efforts compared to the platform core.

Four coordinated Release/<label> waves — Release/Brand-09, Release/airflow, Release/Contigo, Release/Brand-09-2 — each a separate deployment moment. 94.7% merge rate.

1. Four coordinated Release/<label> waves are visible in the PR data — not just the Brand-09 one. Release/Brand-09 (13 PRs across 8 repos, 2025-06-10 → 2025-07-08) is the biggest. Release/airflow (5 PRs across 3 repos, 2026-04). Release/Contigo (4 PRs, 2025-09 → 2025-10). Release/Brand-09-2 (3 PRs, follow-up to Brand-09). Each cluster names a specific deployment moment — these are the choreographed release events.
2. A 94.7% merge rate is a strong code-review hygiene signal. MERGED 10,238 / DECLINED 413 / OPEN 158. Title-prefix breakdown: Feature 1,074, Bugfix 198, Release 30, Hotfix 26 — a clean, disciplined branching workflow.
3. The crowdsource and diy-react-frontend repos are the busiest by PR count. Top repos by PRs: crowdsource (2,222), diy-react-frontend (1,533), dags (1,344), datamagnet (1,127), creators-app (1,119). Customer builds (Brand-26_backend 170, Brand-26-webapp 120) are real but secondary.

Code-review discussion that captures architectural decisions never written down anywhere else.

1. Code-review discussion captures architectural decisions never written down anywhere else. 'Should we cache this at the edge?', 'who owns the auth migration?', 'is this back-compatible?' — senior-engineering conversations that don't make it into commit messages or Confluence. The PR-comment archive is the most candid engineering-debate dataset in the corpus.
2. Per-PR comment density correlates with risk. PRs with many comments are typically schema migrations, security-sensitive changes, or contentious refactors — a small share of PRs but most of the risk-bearing changes. A built-in 'which changes warranted scrutiny' filter.
3. Author breakdown identifies de-facto code reviewers. Authors commenting heavily across many repos are senior reviewers (often architecture-level); those commenting in one repo are domain experts. The reviewer network is visible without HR data.

Deploy cadence, test pass rate, and pipeline health per repo.

1. Real deploy cadence per repo. State, result, and duration per run. CI health (pass/fail rate), deploy cadence (runs per week), and pipeline fragility (high-fail repos) are direct metrics — not modelled, observed.
2. Many pipelines are defined but never executed — a surprising signal. Often a CI-config-drift artifact, useful for engineering-hygiene audits — reveals where CI investment was made but never operationalized.
3. Trigger types show automation maturity per repo. Push-triggered pipelines are mature CI; manual-triggered pipelines are gated workflows; schedule-triggered are nightly batches. The trigger mix per repo is an automation fingerprint.

Branch and tag inventory revealing the release-tagging convention (Release/Brand-09, semver tags) per repo.

1. Release/Brand-09 is a coordinated-release fingerprint. Tag-naming conventions per repo show how releases are gated — semver tags (mature), date-based tags (coarser), or no tags at all (continuous deployment). A measurable release-management discipline signal.
2. Long-lived integration branches reveal trunk-based vs. branch-based development. Repos with persistent develop and release/* branches use a more formal release process; repos with only main use trunk-based.
3. Tagger identity and date per tag. Who actually cuts releases — and how often. A clean signal for release-management roles independent of commit volume.

Author-domain identifies vendor staff vs. internal — clean filter for any internal-only engineering analysis.

1. Author-domain identifies vendor staff vs. internal. External-domain users are vendor staff (Vendor-01, Vendor-02, Vendor-03); contigo-domain users are internal. Slicing analyses by 'internal-only', 'vendor-only', or 'mixed-team' is straightforward.
2. Account type separates bots from humans. Platform-10 security scanner is the only bot author detected — clean filter for human-only productivity analyses.
3. Display name to account-id mapping is the canonical identity link. Anchors all commits, PRs, and review comments to a stable user identity across all six years of history.

Five files spanning the entire visual identity of three products: Brands App, Verve Studio creator app, and the Brand-26 customer build.

1. Five .fig files cover the entire visual identity of three products. The Brands App (the brand-side flagship), the creator-side platform (Verve Studio), and the Brand-26 customer build. There's nowhere else in the corpus where you can see the design source-of-truth for everything Contigo built — these few files cover it.
2. The Brand-26-extra.fig sits inside a complete cross-tool customer build. It pairs with Brand-26_backend, Brand-26-webapp, Brand-26_crm_service repos and the ET-4440 [Brand-26]: Brand-26 App Jira epic — design + backend + CRM + epic for one customer engagement form an aligned set.
3. The capture is a single-day snapshot, not a version history. Useful as a static design-spec reference; for the engineering history that produced these designs, the corresponding Bitbucket repo timestamps tell the story over time.

Django auth — every brand-side user, creator, and admin who ever logged into Contigo's platform.

1. Every user who ever logged into Contigo's platform is here. Brand-side users, creator-side users, internal staff — Django's auth_user is the foundational identity layer. Every other application table refers back to it.
2. Permission flags partition roles cleanly. is_superuser = platform admin; is_staff = internal support team; is_active = currently logged in. Active-user share and admin-to-user ratio are visible directly.
3. date_joined and last_login give the signup-to-login funnel and login-retention shape. Per-cohort retention — fraction of users who joined in month M and logged back in 30 days later — falls out cleanly. The cohort-retention curve a SaaS investor expects.

Cross-product user record distinguishing creator-side from brand-side accounts.

1. Distinguishes creator-side from brand-side at the identity layer. profile_type separates the two halves of the platform; user_type further categorizes. The cross-side identity surface — fundamental for any creator-vs-brand cohort analysis.
2. Payment-permission flag separates active commercial users. Users authorized for payouts are revenue-generating creators; non-payment users are sign-ups or pilots. A clean filter for any payout-and-margin study.
3. Primary-contact flag identifies the key contact at each brand. When multiple users belong to a single brand, this flag picks out the decision-maker — the right person for buyer-committee modelling.

Platform-side brand registry — find brands engaged in product but not yet in CRM (and vice versa).

1. Platform-side brand registry that lines up with Zoho. The application's brand list and the sales team's brand list aren't always the same. Lining them up surfaces gaps: brands engaged in product but not yet in CRM, or in CRM but never activated. Both are interesting.
2. Active vs. archived brands. Currently-engageable brands are a strict subset; the archive captures churned or paused relationships.
3. Free-text keywords per brand support fuzzy matching. Tags and aliases for matching against social-media mentions, hashtags, and external scrape data — useful for cross-source brand-resolution.

Per-brand seat usage — multi-seat brands signal committed adoption; single-seat brands signal trials.

1. Per-brand seat usage signals adoption depth. Brands with many active users typically have committed adoption (real workflow embedded); brands with one user are trials or pilots. The seat distribution per brand is a measurable adoption-stage indicator.
2. 'Show me all users at brand X who logged in last 30 days' is a direct lookup. Per-brand DAU/MAU is computable.
3. Disabled rows mark deprovisioning events. Useful for brand-side seat-churn analysis distinct from full-account churn.

Repeat-collaboration graph — which creators have worked with which brand multiple times, and how the relationship started.

1. Repeat-collaboration graph between brand and creator. Brand-creator pairs that have multiple campaigns together signal genuine creator-brand fit — repeat collaborations are a strong revenue signal beyond one-off deals.
2. Tier and status fields expose relationship maturity. First-time pairs vs. multi-campaign pairs vs. retainer-style relationships are all distinguishable. Direct input for predicting which pairs will convert into the next campaign.
3. Origin of the relationship is preserved. Whether it started from a brand-side discovery query, an inbound creator pitch, or a manual override — useful for measuring the platform's actual matchmaking value vs. just being a payment rail.

Every campaign brief ever logged — budget, brand, target audience, deliverable spec, platform mix.

1. Every campaign brief ever logged on the platform — the unit of brand-side work. Captures campaign name, brand, status, start/end dates, primary platform, final budget, target audience demographics, geo-targeting, deliverable spec. The structured representation of a brief, which most influencer-marketing data lacks entirely.
2. Creator-targeting filters expose how briefs actually match creators. Influencer age range, gender, location, genre, max engagement rate — the filter set is rich enough to drive automated discovery without re-prompting. A real, working brief-to-creator-match input set.
3. A test-campaign flag separates real spend from internal QA. Without that filter, ad-hoc QA campaigns inflate apparent activity. Critical for any revenue-cohort analysis.

Per-deliverable contract spec with a multi-stage moderation lifecycle: draft → ops review → brand review → AI moderation → live.

1. The full moderation lifecycle, preserved. Draft submission → ops review → brand review → AI moderation → live. Each stage has its own timestamp and reviewer; a complete content-moderation pipeline is observable per deliverable. Rare to see this end-to-end outside one company's internal data.
2. Compliance flags reveal contractual fulfilment. Brand hashtag, brand tagged, brand mentioned, link in bio, link in caption, link in story — direct signals on whether the creator credited the brand as required. Per-deliverable compliance rate is the contractual-discipline metric.
3. Rejection-reason history preserves moderation rationale. When deliverables are rejected, the reason is captured (off-brief, low-quality, too late, brand-style violation). Distribution by creator tier is a creator-quality signal.

Bid-vs-final-payment trail — the brokerage economics of every (campaign, creator) negotiation the platform handled.

1. The bid-vs-final-payment trail. Each row captures the offered bid, currency, payment terms, brand-side approval, ops-side approval. With payment-status fields alongside, you see which deals closed at the asked bid vs. negotiated down — the brokerage economics of every (campaign, creator) pair.
2. Performance-marketing campaigns are fully supported. Shipping address, payment mode, vanity URLs, coupon codes per opt-in. The schema isn't just 'creator agreed to post' — it's 'creator agreed, was sent product to address X, is tracked via UTM Y, paid via UPI Z'.
3. Reasons for declined opt-ins are preserved. Why a creator declined (price, brand-fit, schedule) is captured as a structured value — direct input for understanding what makes briefs convert.

The mapping from internal campaign to public Instagram/YouTube post ID — the bridge to public engagement data.

1. The bridge from internal campaign to public social engagement. Each row says 'campaign C produced post P on platform X' with the platform-native ID. Without this, the internal data is disconnected from the public data; with it, post-engagement on public platforms ties back to the internal funnel.
2. Posted-at timestamps tie campaign delivery to actual go-live. Plan vs. actual-publish gap per deliverable is the primary measure of delivery discipline.
3. Shortcodes and platform IDs are preserved for re-fetchability. Public posts can be re-scraped or compliance-audited months after the campaign ran.

Per-campaign top-creator highlight reels — the post-campaign output a brand client actually pays to receive.

1. Per-campaign top-creator highlight reels. The brand-client-facing output that the post-campaign Master Roster Sheet pulls from — top performers by reach, engagement, conversions. The data the brand actually pays to receive, persisted at point of generation.
2. Wordcloud URLs preserve content-theme summarization. Each campaign has a wordcloud built from post captions and comments — theme distribution per campaign as a derived artifact.
3. Soft-deletes preserve report history. Past report snapshots are recoverable; revisions can be re-derived. Audit-friendly.

The cross-platform creator profile (IG + YT + FB + TW + LI) with engagement, genre, and predicted-reach features.

1. The cross-platform creator profile, indexed by influencer. Each (influencer, platform) row carries follower count, follows count, media count, engagement rate, genre, city/state, verification, fake-follower percentage. Instagram, YouTube, Facebook, Twitter, LinkedIn handles all in one place — the multi-platform creator portfolio is plain to read.
2. Predicted reach, predicted impressions, predicted views are pre-computed. Model predictions per handle, ready for use as features for downstream models or as labels for re-training. Pricing-tier validation is built in.
3. Three layers of genre classification — platform-claimed, model-predicted, manually-verified — are kept separate. The data needed to evaluate accuracy of the platform's own genre model.

Cached Instagram profiles with model-predicted reach, engagement rate, and fake-follower percentage.

1. Cached Instagram profile per influencer with the full Insights surface. Followers, follows, media, business category, verification, account type — refreshed via the scrape pipeline at known intervals.
2. EPE (engagement-per-engagement) and EPM (engagement-per-million-views) are pre-computed. Direct features for any reach or engagement-rate model — no need to derive these from raw post stats.
3. fake_follower_percentage is a critical anti-fraud signal. Combined with predicted reach, you can detect creators with inflated audiences before brand contracts get signed. Real-world platform-trust data.

Per-post engagement (likes, comments, reach, impressions, saves) attached to the campaign that paid for the post.

1. The influencer-marketing ROI table. Per-Instagram-post: likes, comments, video views, reach, impressions, saves, exits, replies, taps-forward, taps-back, engagement, sentiment score. The full Instagram Insights API surface, captured per post.
2. Sponsored posts carry the campaign and brand directly. Per-campaign post-engagement aggregation falls out naturally — ROI per campaign, by brand, by industry.
3. Sponsored vs. organic posts are clearly partitioned. Scraped organic posts (the creator's regular content) provide a baseline against which sponsored-post performance can be compared — direct input for 'did the campaign work?' questions.

Cached YouTube channels with predicted long-form and Shorts performance.

1. Discovery cache with predicted performance for both long-form and Shorts. Each channel carries expected views and expected engagement for both formats — pricing-tier validation built in.
2. Three layers of genre classification preserved. Platform-claimed, model-predicted, manually-verified — kept separate. Useful for classification-model evaluation.
3. Country and default language support geo-targeting matching. Brief filters by region, language, and vertical are direct lookups.

Per-video engagement attached to campaign IDs when sponsored — sponsored vs. organic separable cleanly.

1. Per-video engagement attached to campaigns when sponsored. View, like, favorite, comment counts; duration; published-at. Sponsored vs. organic separable cleanly.
2. Hashtags and tags arrays preserved. Useful for content analysis or keyword-trend modelling per channel.
3. Duration in seconds enables Shorts vs. long-form distinction. Combine with view counts and you have the basis for Shorts-vs-long-form performance comparison.

Facebook profile cache — lower-volume than Instagram, reflecting the platform shift over the data window.

1. Lower-volume than Instagram — reflects platform shift over the data window. Facebook profile cache is smaller and less actively scraped, mirroring the broader shift away from Facebook for influencer marketing.
2. connected_at captures when each profile was first platform-tracked. Useful for understanding scrape-coverage age — older connections may have stale data.
3. Enabled flag distinguishes active from churned profiles. Active subset is the discovery-eligible cohort.

Facebook Pages tracked with cross-link to Instagram via ig_profile_id — unified Meta-graph for brand analytics.

1. Tracked Facebook Pages with cross-link to Instagram via ig_profile_id. Brand pages on Facebook and Instagram are linked directly — unified Meta-graph for brand analytics.
2. Category, fan count, country preserved. Useful for Page-targeting filters when building campaigns.
3. Insights-update timestamp shows refresh cadence. Recently-updated pages are scrape-fresh; long-stale pages may need re-fetch — a data-quality signal.

Twitter / X creator coverage — secondary channel, distinct cohort from Instagram and YouTube.

1. Smaller volume than Instagram or YouTube — reflects platform focus. Twitter / X is a secondary channel for the platform; data here is the cohort active in that channel.
2. Verified and protected flags partition the cohort. Public-content creators with public accounts are eligible for campaigns; protected accounts aren't.
3. Account-creation timestamp captures account age. Useful for filtering out brand-new (potentially low-trust) accounts in discovery.

LinkedIn-side B2B-style influencer surface — historically a smaller channel for influencer marketing.

1. B2B-style influencer surface — historically a smaller channel. Country, language, location, headline preserved per profile.
2. Per-row enabled flag. Active LinkedIn-creator cohort is a clean filter.
3. Cross-platform identity preserved via influencer_id. Linkable to the same influencer's Instagram, YouTube, Facebook, Twitter profiles.

Per-influencer affinity scores across 13 genres — the raw material for the platform's discovery ranker.

1. Per-influencer affinity scores across 13 genres — gaming, fashion, food, technology, lifestyle, fitness, travel, beauty, entertainment, comedy, parenting, education, health. The discovery-ranker's input matrix.
2. The same scoring model is applied at multiple grains. Per-influencer, per-IG-profile, per-YT-profile, per-IG-media, per-YT-channel, per-YT-video — letting you understand which level of granularity the platform classifies at.
3. An overall synthetic score combines the per-genre scores. Directly usable as a discovery-ranker feature.

The taxonomy that connects brand-side industry to creator-side genre — central to discovery matching.

1. Industry → genre → sub-genre taxonomy used across the platform. FMCG → Food → Snacks; BFSI → Insurance → Health Insurance. The map that connects brand-side industry to creator-side genre — central to discovery matching.
2. Keyword arrays per industry support free-text classification. Useful for fuzzy classifier input when matching campaign briefs or brand-mention text to canonical industries.
3. Active flag and timestamps per row. Taxonomy evolution over time is visible — when new categories (e.g., crypto / Web3) were added is recoverable.

The keyword dictionary that classifies creator content into genres.

1. Genre-keyword dictionary used by the persona classification pipeline. One row per (genre, keyword) — the lexical layer that scores creator content into a content genre.
2. Status flag per keyword (active vs. deprecated). Classifier-rule lifecycle is preserved — useful for understanding why historical classifications differ from current.
3. Anchors the keyword-vector / genre-vector pipeline outputs. The lexical foundation of the platform's persona model.

Per-creator lifetime earnings — the direct payout history that drives gamification and tier classification.

1. Per-creator lifetime earnings on the platform. The most direct payout-history signal in the corpus — visible in the influencer app, drives gamification and retention.
2. Earnings thresholds map to creator tiers (silver / gold / platinum). Direct input for differentiated outreach, pricing, and account management.
3. Average payout per accepted brief per creator over their lifetime falls out naturally. The creator's-eye-view of platform value.

Time-series of post engagement — the engagement curve from first hour to plateau.

1. Time-series of post engagement. A post can have multiple stat rows over time — the engagement curve from first hour to plateau is reconstructable. Useful for predicting final reach from early signal.
2. Combine snapshot performance with the post's content metadata for cross-feature modelling. Caption length × engagement rate, hashtag count × reach, etc.
3. Snapshot timestamps preserve refresh cadence. Per-post, the snapshot history reveals how the platform's scrape pipeline operates and how engagement curves stabilize.

Performance-marketing campaign rollups — clicks, installs, registrations, acquisitions per offer.

1. Performance-marketing rollups per offer. Total clicks, unique clicks, installs, registrations, acquisitions, revshare — the data behind any 'how is this performance campaign doing' dashboard.
2. A custom-KPI column allows non-standard metrics per offer. Engagement campaigns and awareness campaigns plug their custom KPIs into a single column — the schema stays stable while accommodating campaign-specific success criteria.
3. Soft-deletes preserve historical reports. Past report snapshots are recoverable — supports any 'what did we report at the time' compliance question.

In-app notifications dispatched to creators — the platform-to-creator engagement channel beyond email.

1. In-app notifications dispatched to creators — engagement channel beyond email. Per-influencer, per-campaign, with an archived flag tracking whether the creator acknowledged.
2. Notification-to-engagement ratio per campaign. A high notification volume with low engagement is a noise-spam indicator — useful for tuning notification policy.
3. Tied back to opt-ins for full notification-funnel analysis. 'Notification dispatched → opt-in completed' is a measurable conversion.

Outbound transactional email log — every platform-to-creator email send.

1. Outbound transactional email log — every platform-to-creator send. SendGrid send-id linked to recipient influencer plus internal message ID.
2. Top-volume creator recipients are typically high-engagement / high-tier. Email-channel intensity correlates with creator-tier classification.
3. Failed sends are separately attributable. When SendGrid bounces are linked back, you have the cohort of unreachable creators — important data-quality / contact-list-hygiene input.

Telesales call log — direct evidence of phone-channel outreach.

1. Telesales call log — direct evidence of phone-channel touch alongside email and meeting touches. Per-call duration, contact, outcome.
2. Outbound calls per lead × lead conversion = direct measure of telesales effectiveness. The most honest version of that metric in the dataset.
3. Time-of-day and day-of-week patterns. Call timestamps reveal which hours / days produce contact vs. no-answer — a measurable productivity signal for inside-sales operations.

M365-side calendars — half the org's meeting footprint (the other half on Workspace).

1. The other half of the org's meeting footprint. Half of Contigo runs on M365 (Microsoft); the other half on Google Workspace. To see the full meeting footprint, both calendar services need to be combined.
2. Default calendars distinguish working schedules from auxiliary subscriptions. Holidays, company-calendar, project-calendars are subscriptions; only primary calendar reflects an individual's actual schedule.
3. Owner and access flags per calendar. Useful for calendar-permission-leakage audits, especially when meeting invites cross internal/external boundaries.

Subject lines that name customers and initiatives — meeting cadence per brand is plain to read.

1. Subject lines name customers and initiatives. 'Brand-09 launch readiness', 'Brand-26 sync', 'Brand-10 QBR' — when a brand name shows up as a meeting subject, that brand has scheduled stakeholder commitment. Meeting cadence per customer is plain to read.
2. Online-meeting flag separates Teams-hosted meetings from external/in-person. Online-meeting share is a remote-work signal you can compute exactly per period.
3. Recurring meetings are tied together. Distinguishes one-off meetings from recurring rituals (weekly stand-ups, monthly business reviews) — different operational signals, both visible.

Per-event attendance including external brand-side participation — the strongest customer-collab signal outside email.

1. Per-event RSVP graph including external participation. Brand-side participants from brand-07.com, brand-10.com, etc. show up here when they attend internal meetings — the strongest evidence of customer collaboration outside email.
2. Required vs. optional attendance is distinguished. Required-only attendance is a different graph than required-or-optional; both views are useful for capacity planning.
3. Response status (accepted, declined, tentative). Decline patterns reveal calendar conflicts and de-facto availability constraints.

M365 users with job title and department populated — combine with Teams users for the most complete staff directory.

1. M365 users with job title and department populated. Same profile-richness as the Teams users; combine the two and you have the most complete employee directory in the corpus.
2. Account-enabled flag distinguishes active from deprovisioned. Useful for restricting analyses to current employees while preserving history.
3. Identity is consistent across M365 services. No resolution needed — the same person shows up everywhere with the same email.

Workspace-side calendars including resource calendars (rooms, equipment) — facility-utilization data baked in.

1. Includes resource calendars for rooms and equipment. Office-meeting-room utilization is observable directly — a real-estate utilization signal baked into the data.
2. Primary calendar distinguished from auxiliary subscriptions. Only primary calendar reflects an individual's actual schedule.
3. Time-zone per calendar. Mostly Asia/Kolkata; divergent zones flag dedicated cross-timezone collaborators or remote staff.

Brand-03 has the most meeting cadence — 881 events with Brand-03 in the subject. Then Brand-15 (730), Brand-08 (729), Brand-26 (601), Brand-09 (567). Meeting-graph customer ranking is very different from deal-value ranking.

1. Brand-03 is the brand with the most meeting cadence — 881 events with Brand-03 in the subject. Then Brand-15 (730), Brand-08 (729), Brand-26 (601), Brand-09 (567), Brand-09 (331), Brand-23 (263), Brand-14 (241), Brand-18 (181), Vendor-11 (180), Brand-17 (177), Brand-07 (130), Brand-25 (72), Brand-02 (60). The meeting-graph customer ranking is very different from the deal-value ranking — Brand-03 has 8x more meetings than Brand-10 (107) despite Brand-10 being the larger account by some measures.
2. External-domain organizers reveal sustained customer collaboration. Brand-side participants from vendor-05.com (733 events organized!), brand-09.com (521), brand-10.com (107), meta.com (74), vendor-12.com (129), platform-11.com (86) host meetings on the calendar — the strongest evidence of customer collaboration outside email.
3. Recurring rituals are reconstructable verbatim. The recurrence rules preserve weekly all-hands, monthly QBRs, quarterly business reviews — recurring meeting cadence is observable end-to-end. Plus attachment URLs that tie meeting artifacts in Drive to the calendar event that produced them.

External-domain participation from brand-side organizers (brand-07.com, brand-23.com, etc.) — measurable customer collaboration.

1. External-domain participation visible. Brand-side organizers from brand-07.com, brand-23.com joining internal meetings — measurable external-collaboration signal independent of email.
2. RSVP status tells you who actually showed up. Decline rate per organizer measures convening effectiveness.
3. Per-event attendance is full-fidelity. Every (event, attendee) pair is preserved for full meeting-graph analysis.

Workspace users with org_unit_path exposing the OU tree — department-level rollups become trivial.

1. org_unit_path exposes the OU tree. Mirrors the org chart to whatever depth is configured in Workspace — department-level rollups are immediate.
2. Admin flag identifies Workspace administrators. Security-relevant cohort — admins have elevated access to other users' Drive content.
3. Suspended-account flag captures deactivated users. Useful when restricting analyses to current employees.