LENS

Context/LENS-BRIEF.md

LENS Product Brief

Authoritative synthesis of all product decisions for LENS. Produced 2026-04-07 by merging two independent digests of ~14,700 lines of conversation, cross-referenced against the Obsidian vault PRD, three deep research reports, and the live codebase.

This document supersedes digest-final.md, digest-part1.md, digest-part2.md, and all files in docs/files/. Any future session should read this first.

1. WHAT LENS IS

LENS = Leading Engagement & Narrative Synthesis

A social content intelligence platform that gives marketing teams an empirical, sourced understanding of how their content performs on LinkedIn versus competitors. It finds the people, patterns, and gaps that exist in the SPACE BETWEEN entities, not inside any single entity's data.

LENS is NOT a content recommendation tool. Content recommendations are the downstream output, not the purpose.

Sibling to PRISM (Prospect Intelligence Platform). LENS watches the market surface continuously; PRISM audits depth on demand. LENS feeds PRISM, PRISM feeds sales.

Subject company for v1: Algolia. v1 = the demo. v2 = the product. Do not conflate.

2. FIVE LOCKED PRINCIPLES (NEVER VIOLATE)

Entity Symmetry -- Every entity runs through identical pipelines. No special casing. Algolia is just entity_id: E01. The is_subject: true flag is the ONLY difference.
The Pyramid is the Processing Order -- L1 -> L2 -> L3 -> L4. Each feeds the next. No shortcuts.
Layer 4 is the Product -- Layers 1-3 are infrastructure. Layer 4 (meta-intelligence) is what humans read and act on.
One Person Record -- A person engaging with multiple entities = ONE enriched record cross-referenced to all entities. Never duplicated.
Industry-Agnostic by Design -- Change the is_subject flag = new customer. This is the business model.

3. THE FOUR-LAYER PYRAMID

Layer 1: Raw Data Engine

Identical scraping pipeline for ALL entities
Company pages + exec profiles + engager profiles
Posts + likes + comments + shares + follower counts
Tools: Apify (supreme_coder/linkedin-post actor, tested and confirmed)
Enrichment: Apollo/Clay (deferred to post-MVP)
Storage: PostgreSQL

Layer 1A (SAFE, build now): Post content, comment text (no identity), follower counts, exec posts and engagement counts.

Layer 1B (RADIOACTIVE, deferred): Named individuals who liked/commented, their job titles/companies, cross-entity overlap.

Layer 2: Entity Synthesis Engine (per-entity, identical template)

Five modules per entity, run weekly: - A: Content Performance -- format leaderboard, cadence, top posts, engagement trends, optimal posting window (Stage 1) - B: Audience Profile -- cohort breakdown seniority x industry, named engager list (Stage 2+, needs Layer 1B) - C: Narrative Intelligence -- message pillars, topic clusters, tone, CTA intent, narrative evolution (Stage 1) - D: Exec Voice -- what leadership amplifies personally, divergence from company narrative (Stage 1) - E: Engager Universe -- company breakdown, seniority distribution, geographic spread (Stage 2+, needs Layer 1B)

Layer 3: Comparison Engine (cross-entity)

Content comparison: engagement rates, format distribution, acceleration
Audience comparison: cohort ownership, engager overlap, unclaimed cohorts
Narrative comparison: topic ownership, convergence, white space, tone
Engager flow: switchers, migration patterns, dormant, new arrivals

Layer 4: Meta-Intelligence Engine (the product)

v1 outputs: - Weekly Issue (Monday, Editorial mode) - Explore Workbench (always-on investigation surface)

v2 outputs (deferred): - Monthly Battlecard (full competitive narrative landscape) - Real-Time Alerts (Slack: competitor spikes, switchers, narrative shifts) - Entity Deep-Dive (on demand for any entity)

4. THE TWO-MODE PRODUCT (CONCEPT C, LOCKED)

READ MODE (Editorial) = The Weekly Issue

Default entry point for execs + content team
Cadence: weekly (publishes Monday 08:00)
Verb: consume (passive, paced, narrative)
Format: ONE scrollable page at /editorial with floating TOC sidebar
ALGOLIA ONLY. No competitors. Period.
Competitors were removed from Editorial entirely (April 6 directive)
9 fixed sections (see Section 7 below)

EXPLORE MODE (Workbench) = Investigation Surface

Cadence: always-on, user picks time range
Verb: investigate (active, free, interrogative)
Audience: analysts, content team, social team
This is where ALL competitive analysis lives
4 top-level asymmetry lenses (see Section 8 below)
7 entity lenses as nested drill-downs inside asymmetry lenses

Bridge Between Modes

READ -> EXPLORE: Click any claim's arrow icon to open Explore pre-filtered to supporting data (preserves competitive context)
EXPLORE -> READ: "Pin to next issue" queues findings for next Monday's draft. Pinned findings are TRANSLATED into Algolia-centric framing before appearing in the issue (keeps editorial voice clean, never says "Constructor is beating us")

5. FOUR PRODUCT PRINCIPLES

Sunday Sync -- Weekly background job (Sunday 23:00 UTC). Verifies executive roster for all monitored entities via web search/Perplexity. Versioned JSON per week (immutable snapshots). Drift detection: new joiner, departed, promoted = story leads for the Issue. Runs BEFORE anything else in the weekly pipeline.
Source Links Everywhere -- Every claim has a URL. Every post -> LinkedIn URL. Every engager -> profile URL. Every exec -> LinkedIn + company website. Every community signal -> thread URL. No naked numbers. No unattributed claims.
Show The Work -- Every Content Brief has 4 labeled boxes: - COHORT: which audience persona is growing/engaging - PATTERN: what content pattern works for them - SIGNAL: the trigger (competitive move, white space, community thread) - WHO TO REACH: 5 named engagers who would receive it well - Minimum 3 of 4 confidence markers needed to publish. If AI cannot fill all 4, brief is rejected before editor sees it. - Confidence markers: (1) signal fresh within 7 days, (2) pattern well-established, (3) direct contrast/opening exists, (4) engagers identified.
Audience Tracking -- 5 personas tracked across 4 rolling time windows (this week, last month, 3 months, 6 months). Drift detection feeds Content Brief targeting.

Five Personas:

Ecommerce Builder, AI/ML Practitioner, GTM Leader, CS Leader, AI Founder

6. EDITORIAL WORKFLOW (Sunday Night Pipeline)

22:00  Sunday Sync runs -> drift flags generated
22:30  Census complete -> entity config updated -> drift queued as story leads
23:00  Data pipeline runs -> all LinkedIn posts, community signals, partnerships, events
23:30  Issue draft generation -> AI drafts every section using templates
[gap]
07:00  Monday: Editor reviews -> approves/edits Opener and Cover claim
08:00  Monday: Issue published -> Slack digest sent

Editor: Arijit in MVP
Tue-Sat: Team investigates in Explore, pins findings for next issue
Pinned findings get prioritized in next Sunday's AI draft

7. WEEKLY ISSUE STRUCTURE (9 SECTIONS, LOCKED)

The issue is a single scrollable page at /editorial. All sections are Algolia-only. Individual URLs are anchor deep-links (#cover, #opener, etc.). Floating Table of Contents sidebar for navigation.

The Nine-Section Arc (same order every issue):

Cover -- Issue number + two parallel sentences (max 10 words each, serif, large) + date + stat line (posts, engagers, entities tracked). Left rail = issue TOC. "Past issues" link at bottom.
Opener -- Editor's note, 80-150 words, three beats: what week showed / why it matters / reader guidance. First-person editorial "we". Sign-off: "-- The Editor"
S1: The Thesis -- One claim, one simple visual (e.g., two bars: customer proof 4.8% vs vision 2.1%), one supporting sentence (60-100 words).
S2: What Performed -- RANKED LIST, always exactly 3 posts. Each: large numeral + post title + "view on LinkedIn" link + "Why it worked" paragraph (40-60 words) + named reposters + engagement stat.
S3: Cohort Comparison -- 4 columns (THIS WEEK / LAST MONTH / LAST 3 MONTHS / LAST 6 MONTHS). Five personas with % and direction arrows. Total engager count per period. Editorial takeaway. Two cross-links.
S4: Who Was Watching -- Always FIVE named engagers (not more, not less). Each: Name + Role + Company + Why-they-matter paragraph + Role tag [Customer/Alumni/Tech voice/Partner/Prospect].
S5: Network Movements -- Three sub-feeds, ALL Algolia-side: EXECUTIVE MOVES (from Sunday Sync), PARTNERSHIP MOVES, EVENT FOOTPRINT.
Closer: Content Briefs -- Three briefs always. Each: Title + HOOK (dashed border, copy-paste ready, in quotes) + Format/Voice/Persona/ Ship by + SHOW THE WORK (4 boxes) + Confidence (4 markers, min 3/4) + cross-link to Explore.
Colophon -- Issue number + date, data stats, editor/draft/approved/ published timestamps, legal disclaimer, end-of-issue statement.

What a Weekly Issue NEVER Does:

"Thank you for reading"
Apologize for missing data
Hype itself
Address reader as "you"
Break editorial voice
Use em-dashes or exclamation points

8. EXPLORE MODE: 4 ASYMMETRY LENSES

Left rail shows only 4 top-level asymmetry lenses in v1. Entity lenses (Content Performance, Audience, Narrative, Exec Voice, etc.) appear as "DRILL DEEPER" expandable sections inside each asymmetry lens.

Voice Gap was folded into Narrative Gap for v1 (deferred as standalone).

Screen 10: Explore Landing

Left rail = 4 lenses listed + time selector + entity checkboxes. Main = brief description of each lens + "where to start" guidance.

Screen 11: Cohort Gap

Table: Persona / ALGOLIA / FIELD AVG / GAP. "WHERE THE GAP IS GROWING" line chart. Drill-deeper expandable links. "WHO WE'RE LOSING" named people section. Action buttons: [Pin to next issue] [Export] [Ask LENS].

Screen 12: Switcher Signal

Summary count + breakdown (leaning toward competitors / equal / leaning toward us). AT-RISK SWITCHERS list with expandable rows showing per-entity engagement counts and actual posts engaged with.

Note: Switcher Signal requires Layer 1B data. In v1 (Layer 1A only), this lens will show limited/mock data or be gated.

Screen 13: Content Gap

FORMAT PERFORMANCE table (Format / ALGOLIA / FIELD / WHO WINS). "THE GAP" plain English explanation. "THEIR TOP VIDEO POSTS" with links.

Screen 14: Narrative Gap

TOPIC OWNERSHIP heatmap grid (4 states). "THE GAP" plain English. "THEIR TOP POSTS" on that topic. Includes Voice Gap analysis (folded in for v1).

9. SCREEN INVENTORY (17 total for v1)

#	Screen	Mode	Status
1	Cover	Editorial	Built
2	Opener	Editorial	In progress
3	Thesis (S1)	Editorial	In progress
4	What Performed	Editorial	Pending
5	Cohort Comparison	Editorial	Pending
6	Who Was Watching	Editorial	Pending
7	Network Movements	Editorial	Pending
8	Closer	Editorial	Pending
9	Colophon	Editorial	Pending
10	Explore Landing	Explore	Pending
11	Cohort Gap	Explore	Pending
12	Switcher Signal	Explore	Pending (L1B)
13	Content Gap	Explore	Pending
14	Narrative Gap	Explore	Pending
15	Editor Review	Workflow	Pending
16	Archive	Utility	Pending
17	Settings	Utility	Pending

App Shell

Top bar: LENS logo | [EDITORIAL] [EXPLORE] tabs | Entity selector | Profile
Nothing else. Mode tab swaps entire body.

Route Structure (Next.js 16 App Router)

apps/web/app/
  layout.tsx          (app shell)
  page.tsx            (redirect to /editorial)
  editorial/
    layout.tsx        (editorial layout with left rail TOC)
    page.tsx          (single scroll: all 9 sections)
  explore/
    layout.tsx
    page.tsx          (landing)
    cohort-gap/page.tsx
    switcher/page.tsx
    content-gap/page.tsx
    narrative-gap/page.tsx
  editor/page.tsx
  archive/page.tsx
  settings/page.tsx

10. VISUAL IDENTITY

LENS is a publication, not a product surface.

Typography: Serif (Fraunces, from Google Fonts) for ALL content: headlines, body, dates, sign-offs. Mono for ALL metadata: issue numbers, stat lines, timestamps, labels. No sans-serif except shadcn/ui system UI. Currently using Georgia as fallback; Fraunces not yet added.
Color: Black on near-white (or near-white on black for dark mode later). One accent color, used sparingly. NO gradients. NO brand colors.
Whitespace: Generous. py-24, max-w-4xl standard. "Breathe like newspapers, not dashboards."
Hierarchy: Big serif for the one thing the page is about. Small mono for everything else.
Logo: Colophon (prominent) + app footer (tiny, persistent). Needs Algolia logo SVG (not yet provided). No Algolia branding on most pages.

11. LENS PUBLICATION VOICE (from Voice/LENS.md)

Identity: Editorial publication voice. Quiet, confident, restrained. Trade weekly editor.
Narrator: Third-person observer. Never "I" (except Opener). Rare "we". Never "you".
Register: Formal enough to trust, informal enough to read.
Verbs: observe, track, identify, surface, note, record, measure, report.
Banned from reader-facing copy: cohort, segment, normalized, regression, A/B, funnel, stack rank, amazing, incredible, game-changing, leverage, synergy, thought leadership.
Replacements: audience, persona, pattern, story, voice, trend.
Zero em-dashes. Zero exclamation points. Zero hedging.

Note: "cohort" is banned from reader-facing output but used internally (Cohort Gap lens name, cohort analysis module).

12. DATA COLLECTION POLICY

What LENS Collects:

Company-owned pages on the open web (About/Leadership/Team, no auth)
Company posts on social platforms (brand handles + tagged employees, via official APIs only)
Engager identity from public engagement on public posts (public fields only)
Public community discussions (HN, Reddit, GitHub public repos, conference websites, press releases)

What LENS Does NOT Collect:

Private profiles, posts, accounts
Content behind login walls
DMs, private groups, paid newsletters
Email, phone, contact details (even if technically public)
Third-party enrichment / identity resolution
Sentiment analysis or psychological profiling
Geographic/demographic data not explicitly published

How LENS Collects:

Official platform APIs (LinkedIn MDP, X API, Reddit API, etc.)
Licensed third-party providers (Apify, Perplexity) where THEY handle ToS compliance
Public web pages respecting robots.txt
NEVER by circumventing authentication, rate limits, or access controls

Three Enforcement Layers:

Source Catalog (gatekeeper): all sources documented before use
Ingestion Modules (enforcement): every record stored with source URL + timestamp + method + basis. Records without source URL rejected.
Rendering Layer (display): cannot render any claim without source URL. Missing source = omit, not render.

13. LEGAL RESEARCH FINDINGS

Three independent deep research reports (Claude, Gemini, OpenAI, April 2026) stored in Obsidian vault Research/ all reach the same conclusion:

Layer 1B (engager identity capture from competitor pages) is legally untenable.

Evidence:

Proxycurl ($10M ARR): destroyed by LinkedIn lawsuit (Jan 2025)
KASPR: EUR 240K fine, deleted 160M records (Dec 2024)
Common Room: removed competitor engagement feature (Dec 2024)
hiQ v. LinkedIn: won CFAA, lost on contract law (Dec 2023)
ProAPIs: sued for fake accounts (Oct 2025)

The API Wall is Absolute:

LinkedIn MDP: aggregate stats only
Reactions API: individual URNs only for admins of YOUR OWN content
No official access to competitor engagement data at individual level

Key Variable: Amplemarket

Operating contact-level competitor engagement intelligence without apparent enforcement. Investigate their methodology before committing to Layer 1B.

Three Defensible Paths:

Own-page engagement analytics (Reactions API, authorized, zero risk)
Public signals CI (competitor content + jobs + website + G2 + press)
Multi-signal intent layer (partner with Bombora/6sense + client data)

Impact on v1:

v1 operates on Layer 1A only. Viable Explore lenses: Content Gap, Narrative Gap. Partially viable: Cohort Gap (own-page only). Blocked: Switcher Signal (requires cross-entity engager identity).

14. ENTITY REGISTRY

Entity	Tier	In DB
Algolia	Tier 1	Yes
Constructor.io	Tier 1	Yes
Coveo	Tier 1	Yes
Bloomreach	Tier 1	Yes
Elastic	Tier 1	Yes
Typesense	Tier 2	Yes
Meilisearch	Tier 2	Yes

7 entities in database, 28 executives stored, all LinkedIn URLs validated.

Each entity gets 3-5 exec profiles monitored.

15. TECH STACK (LOCKED)

Layer	Choice
Frontend	Next.js 16.2.2 + TypeScript strict + Tailwind + shadcn/ui
React	19.2.4
Monorepo	pnpm + Turborepo
Testing	Vitest + Playwright + React Testing Library
Logging	Pino
Deploy	Vercel (apps/web as Root Directory)
Backend	Python (deferred)
DB	PostgreSQL
Scraping	Apify (supreme_coder/linkedin-post)
Enrichment	Apollo/Clay (deferred)
Workflows	Temporal.io (deferred)
Backend deploy	Modal/Railway/Fly.io (deferred)

Monorepo Structure:

LENS/
  apps/web/              <- Next.js frontend (Vercel Root Directory)
  packages/
    ui/                  <- shared React components (@lens/ui)
    config/              <- shared eslint/tsconfig/tailwind (@lens/config)
    types/               <- shared TypeScript types (@lens/types)
    mock-data/           <- mock data for frontend (@lens/mock-data)
  services/              <- Python backend (deferred)
  shared-py/             <- shared Python libs (deferred)
  workflows/             <- Temporal workflow definitions (deferred)

16. DATABASE SCHEMA (from Part 1 conversations)

Core Tables:

entities (18 fields): entity_id, name, linkedin_id, website_url, is_subject, industry, color, etc.
entity_snapshots (10 fields): weekly snapshots with employee_count, engagement_trend enum
exec_registry (10 fields): discovery_method enum, relevance_score
posts (24 fields): velocity_score, AI fields with enum types, period_id
comments (8 fields): comment_sentiment enum
persons (26 fields): gravity_score, switcher_flag, migration_vector, CRM fields
engagements (9 fields)
narratives (6 fields): module_type enum
comparisons (5 fields)
insights (5 fields): insight_type enum
verticals (5 fields)

Editorial Tables (new for v1):

issues: number, publish_date, cover_claim, status, editor_note
content_briefs: 3 per issue, title, hook, format, voice, persona, ship_by, 4-box provenance
pins: pinned findings from Explore for next issue
sync_history: Sunday Sync versioned snapshots, immutable per week
personas: classification records per engager

17. SUNDAY SYNC JSON SCHEMA

{
  "week": "2026-04-07",
  "last_run": "2026-04-06T23:00:00Z",
  "entities": [
    {
      "company": "Algolia",
      "domain": "algolia.com",
      "execs": [
        {
          "name": "Bernadette Nixon",
          "title": "CEO",
          "linkedin_url": "linkedin.com/in/bernadette-nixon",
          "source_url": "algolia.com/about/leadership",
          "verified_at": "2026-04-06T23:00:00Z",
          "status": "confirmed",
          "previous_status": null
        }
      ]
    }
  ]
}

Drift flags: "new joiner", "departed", "promoted/moved". Both linkedin_url AND source_url mandatory. Missing = "unverified".

18. 8 BACKEND MODULES

#	Module	Layer	Stage	Status
1	Sunday-Sync	L1	1	Placeholder
2	Post-Ingestion	L1	1	Placeholder
3	Engager-Capture	L1	2 (1B)	Blocked
4	Cohort-Analysis	L2	2 (1B)	Blocked
5	Asymmetry-Engine	L3	1 (partial)	Placeholder
6	Issue-Drafter	L2	1	Placeholder
7	Editorial-Review	L3	1	Placeholder
8	Reader-Frontend	L4	1	In progress

All modules implement the Manifesto contract: execute(), validate(), persist(), health_check(). All return typed Pydantic ModuleOutput.

19. VERTICAL CONFIG FORMAT

vertical: Enterprise Search
subject: Algolia
entities: [Constructor.io, Coveo, Bloomreach, Elastic, Typesense]
exec_discovery:
  min_followers: 1000
  min_posts_per_month: 2
reporting:
  slack_webhook: [url]
  email_recipients: [list]
  timezone: America/New_York
colors:
  Algolia: "#6B3FA0"
  Constructor: "#00B4D8"
  Coveo: "#FF6B6B"
  Bloomreach: "#0077B6"
  Elastic: "#6B7280"
  Typesense: "#F59E0B"

Planned verticals: 1. enterprise-search.yaml (LIVE) 2. ecommerce-platform.yaml (Spryker vs Commercetools et al.) 3. headless-cms.yaml (Amplience vs Contentstack et al.)

20. NOTEBOOKLM DECK BENCHMARK

Arijit built a manual LinkedIn intelligence report using Claude Chrome extension + NotebookLM. The 15-page deck became the design benchmark for LENS Read mode.

Transferable elements:

Funnel narrative (23 posts -> 3 deep dives -> 110+ profiles)
Micro-persona decomposition (5 personas, %, color-coded, proportional)
"Player Cards" = Engager Registry (searchable, filterable)
Content Performance 2x2 Matrix (Stars / Hidden Gems / High Noise / Needs Work)
Alumni Effect, Competitor Intel (N x N matrix), Exec Reach (personal vs brand amplification ratio)
Engagement-to-value-prop mapping = Narrative Divergence Map

What the deck doesn't do that LENS adds:

No temporal dimension (deck = 30-day snapshot; LENS = time series)
No evidence tiering (LENS: every claim has confidence grade + drill)
No comparative baselining (LENS compares against monitored-set average)
No engager-to-pipeline linking (LENS + PRISM closes to CRM)

21. KEY PEOPLE AT ALGOLIA

Name	Role	LENS Usage
Laura	CMO	Reads cover+opener+closer in 90 sec
Rod	Social media content	Full issue reader, ships briefs
Ode	Marketing	Full issue reader, ships briefs
Nicole	Content development marketing	Editorial mode primary
Piyush	Chief Ecosystem Dev Officer	Arijit's manager
Jillian	RevOps/Marketing	CoE co-founder
Katie	Head of Sales	Sales consumer

22. WORKING AGREEMENT (at LENS root)

TDD strictly: write failing tests first, RED, implement, GREEN
Never claim done without pnpm lint + pnpm typecheck + pnpm test
Never commit without verification passing
Scope discipline: do NOT pre-build types, mock data, or code for future screens
If pre-existing bug blocks verification: smallest fix, report explicitly
Honest reporting: "Noticed but not fixed" section required
No speculative work: flag as SUGGESTIONS only

Speculative Work Incident:

Claude Code added types/mock data for unbuilt screens (CohortRow, Engager, LinkedInPost, etc.). Task 01.5 rolled back all speculative additions. 75 tests -> ~40 tests. Rule: "When retrofitting with tests, only test what currently exists. Flag additions as SUGGESTIONS."

23. WHAT'S BUILT SO FAR

Monorepo scaffold (apps/web, packages/ui, config, types, mock-data)
Editorial Cover screen at localhost:3000/editorial: - "LENS . ISSUE 14" masthead (tracked uppercase) - "Proof beats vision. / Luxury beats everything." in serif - Thin horizontal rule - "Week of April 7, 2026" - "23 posts . 147 engagers . 6 entities tracked" - "scroll to read" cue - Font: currently Georgia (Fraunces not yet added)
Some additional editorial sections in progress
Phase 0 backend: 7 entities + 28 executives in PostgreSQL
Obsidian vault with PRD complete, specs as placeholders

24. v1 SCOPE

In Scope:

Both modes with bidirectional bridge
9 editorial sections + 4 explore lenses + landing
Editor review screen
Sunday Sync + drift detection
Source links + Show The Work enforcement
Audience tracking (4 time windows)
Content Gap + Narrative Gap lenses (Layer 1A)
LinkedIn only, single-tenant (Algolia)
Data Collection Policy enforced
Export to PDF only

Explicitly Deferred (v1.5/v2):

X as second platform
Multi-tenancy (workspace_id on every table)
Hiring/pricing signals as asymmetry dimensions
Voice Gap as separate lens (folded into Narrative Gap)
Standalone entity dossier views
Multi-editor workflow
PRISM integration
Mobile/responsive
Slack and email delivery
BYOK (Perplexity, Apify, Anthropic, OpenAI, Gemini)
Stripe integration, user accounts/auth/roles
Self-service onboarding, admin screen

Explicitly OUT OF SCOPE:

Private data scraping
Sales battlecards
Real-time alerts
Arbitrary user-defined charts

Reversibility Principle:

Every v1 decision must be reversible toward v2. Mock data shapes must map to future DB schema with workspace_id. "Subject company" = parameter, never hardcoded constant.

25. PRD SUCCESS CRITERIA

Editor reviews and approves issue in <30 minutes Sunday night
CMO reads cover + opener + closer in 90 seconds and gets full picture
Content marketer ships at least one brief as-drafted per week
Analyst identifies real gap/signal from Explore not visible in editorial
Every claim links to verifiable public source within 2 clicks
Sunday Sync runs 4+ consecutive weeks, detects >= 1 real exec transition
Zero orphaned records (all have source URL, timestamp, licensing basis)

26. SOLOPRENEUR VISION

LENS is one project in a solopreneur operating system (MyOS):

Project	Purpose	Status
LENS	LinkedIn competitive intelligence	In progress
PRISM	Prospect intelligence (account deep-dives)	Placeholder
CoE	AI Innovation Center of Excellence	Placeholder
CurioQuest	Personalized children's storybook	Placeholder
MyOS	Meta-orchestration for all projects	Placeholder

LinkedIn goal: 1,440 -> 14,000 followers (10x) from April 7, 2026
Platform is vertical-agnostic: swap entity registry for any industry
LENS -> PRISM triggers: gravity_score > 75, switcher detected, competitor mention, exec narrative shift
Base/MyOS architecture: extract from LENS, don't pre-build. Build two LENS services first, find common patterns, THEN extract.

LENS -> PRISM Handoff Triggers:

gravity_score > 75     AND not in CRM    -> auto-queue in PRISM
switcher_flag = true   AND tier_1_account -> immediate PRISM alert
competitor_mention     in comment thread  -> flag for PRISM
exec_narrative_shift   detected           -> PRISM competitive audit

27. WHAT WENT WRONG (LESSONS LEARNED)

Context amnesia -- Previous Claude session lost all decisions
Over-engineering editorial -- Built editorial before investigation surface
Wrong build priority -- Should have built Explore first
Repeated product discovery -- Claude kept re-asking answered questions
No persistence of decisions -- Key decisions lost to compaction
Speculative work -- Claude added code for unbuilt screens
Early/intermediate decisions persisted as final -- e.g., competitors in Editorial was an intermediate decision that got locked in digest-final.md despite being reversed

28. ARIJIT'S WRITING VOICE (from Claude analysis of Medium articles)

Pattern-first thinking: leads with underlying pattern before instance
Acceleration structure: builds momentum (slow -> fast -> climax)
Earned complexity: starts simple, earns complexity through specificity
Honest confrontation with failure: names mistakes directly
Practitioner's eye: notices what others miss, names it
Signature move: "Here's what nobody talks about..."
Voice/Arijit.md identified but not yet written in vault

END OF PRODUCT BRIEF

This document is the single source of truth for LENS. It will be synced to Obsidian at Projects/LENS/Context/LENS-BRIEF.md.