PRISM Module Catalog
Complete inventory of all 20 PRISM modules. For each module: what it does,
what APIs it calls, what it outputs, what makes it production-grade.
Use this to write module specs per Module-Spec-Template.
Wave 1: Intelligence Collection (13 Modules, Parallel)
intel-company (Foundation Hub)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Sonnet (structured extraction) |
| APIs |
Perplexity sonar (1 call), Homepage HTML fetch |
| Dependencies |
None (all other modules depend on this) |
| Gate |
If this fails, entire audit aborts |
| Key Output |
legal_name, industry, executives[], competitors[], business_model, has_search_bar |
| Enrichment |
Deterministic JSON parsing (NO LLM). Citation extraction via regex. |
| Validation |
8 checks: legal_name, domain format, business_model depth, executive count, competitor count, source presence, headquarters, no NO_SOURCE |
| Production Edge |
3-attempt retry with smart backoff. Field-level source citations. Homepage automation for search bar detection. Cross-check competitors vs KNOWN_ALGOLIA_CUSTOMERS. |
| Persistence |
Writes to accounts table (all fields normalized) |
intel-techstack (BuiltWith-Powered)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
None (pure API) |
| APIs |
BuiltWith (all 7 endpoints) |
| Dependencies |
None (but reads Account.competitors if populated by intel-company) |
| Key Output |
search_vendor (ACTIVE/TAG_ONLY/REMOVED/UNDETECTED), ecommerce_platform, all_technologies[], competitor_tech_stacks[], golden_angle_competitors[] |
| Enrichment |
None. BuiltWith returns structured data. |
| Validation |
Tech count >= 3, search vendor status valid |
| Production Edge |
Multiple detection sources. Evidence tier VERIFIED for direct tags. Golden Angle identification (competitor using Algolia). Removed tech detection. Competitor fan-out. |
intel-traffic (SimilarWeb + Perplexity)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Haiku (Google Trends assessment) |
| APIs |
SimilarWeb (10 endpoints), Perplexity (Google Trends) |
| Dependencies |
intel-company (competitor domains from Account.competitors) |
| Key Output |
monthly_visits[], traffic_sources[], engagement, device_split, organic_keywords[], google_trends, competitor_traffic[], comparative_summary |
| Enrichment |
1 LLM call for Google Trends momentum |
| Validation |
9 checks: monthly visits present, source distribution, keyword counts |
| Production Edge |
Brand keyword classification. Seasonal pattern detection (>15% deviation). Competitor baseline comparison. |
intel-financial-public (Yahoo Finance + SEC EDGAR)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Sonnet (SEC filing analysis) |
| APIs |
Yahoo Finance (free), SEC EDGAR EFTS (free), Perplexity |
| Dependencies |
intel-company (ticker confirmation) |
| Skip Condition |
is_private=True or ticker=None |
| Key Output |
annual_financials (3-year), market_data, analyst_data, sec_insights, competitor_financials[] |
| Validation |
9 checks: revenue data present, market cap valid, financial years sequential |
| Production Edge |
Multi-year trend analysis. SEC filing link tracking. Public/private branching. Analyst consensus aggregation. |
intel-financial-private (6-Source Waterfall)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Sonnet (Instructor structuring) |
| APIs |
Perplexity (6 waterfall queries) |
| Dependencies |
intel-company |
| Skip Condition |
is_private=False |
| Key Output |
revenue_waterfall (estimates[], best_estimate), funding_data, employee_revenue_model, competitor_estimates[] |
| Enrichment |
10 LLM calls (6 Perplexity + 4 Claude Instructor) |
| Validation |
>= 2 independent estimates |
| Production Edge |
6-source triangulation. Best estimate = median (robust vs outliers). Employee-based cross-check. |
intel-news (90-Day News Sweep)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Sonnet |
| APIs |
Perplexity sonar-pro |
| Dependencies |
intel-company (executives, competitors) |
| Key Output |
prospect_articles[], prospect_exec_quotes[], urgency_signals[], competitor_news[] |
| Validation |
>= 1 prospect article |
| Production Edge |
Urgency signal classification (leadership/product/tech). Executive quote extraction + high-value scoring. Sell signal detection. |
intel-hiring (Apify LinkedIn + Perplexity)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Sonnet (ICP classification) |
| APIs |
Apify LinkedIn Jobs (primary), Perplexity (fallback) |
| Dependencies |
intel-company (executives, competitors) |
| Key Output |
open_roles[] (with icp_tier), hiring_velocity, build_vs_buy signal, buying_committee[], competitor_hiring[] |
| Validation |
>= 1 open role or hiring_summary present |
| Production Edge |
ICP tier classification (Economic Buyer/Technical Buyer/Champion). Buying committee discovery. Build-vs-buy signal. Hiring velocity trends. Dual-source (Apify primary, Perplexity fallback). |
intel-social (Executive Social Intelligence)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Sonnet |
| APIs |
Perplexity, Apify LinkedIn posts (optional) |
| Dependencies |
intel-company (executives, competitors) |
| Key Output |
prospect_posts[], prospect_exec_quotes[], twitter_activity, most_quotable[], competitor_social[] |
| Validation |
>= 1 post or >= 1 exec quote |
| Production Edge |
Multi-platform (LinkedIn + Twitter + public statements). Post engagement scoring. Algolia-relevance classification. Most quotable extraction. |
intel-investor (Said vs Found Engine)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Opus (deep reading + synthesis) |
| APIs |
Perplexity sonar-pro (earnings, YouTube, board), SEC EDGAR |
| Dependencies |
intel-company, intel-financial-public |
| Key Output |
prospect_quotes[], said_vs_found[] (THE core deliverable), competitor_intel[], youtube_appearances[], board_members[], risk_factors[], top_sales_angles[] |
| Enrichment |
10+ LLM calls |
| Timeout |
600 seconds (10 minutes) |
| Validation |
10 checks |
| Production Edge |
TWO execution paths (public vs private). Said vs Found matrix (executive quotes mapped to Algolia value props). Board tech composition analysis. SEC 10-K risk factor parsing. Sales angle mapping. |
intel-partner (Crossbeam + Perplexity)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Sonnet |
| APIs |
Crossbeam (deferred, using Perplexity fallback) |
| Dependencies |
intel-company, intel-techstack |
| Key Output |
partner_overlaps[], co_sell_opportunities[], si_relationships[], vertical_case_studies[], partner_play |
| Production Edge |
Co-sell opportunity identification. SI relationship mapping. Vertical case study matching. Recommended partner play. |
intel-industry (Vertical Benchmarks)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Opus |
| APIs |
Perplexity sonar-pro (5 vertical queries) |
| Dependencies |
intel-company (industry, sub_vertical) |
| Key Output |
vertical_benchmarks[], industry_trends[], pain_points[], algolia_case_studies[], search_vendor_landscape[] |
| Validation |
>= 1 benchmark, trend, pain_point |
| Production Edge |
Vertical-specific research (not generic). Benchmark sourcing (Baymard, Forrester, NRF). Vendor landscape mapping. Pain point to Algolia mapping. |
intel-competitors (Pure Synthesis)
| Aspect |
Detail |
| Layer |
Synthesis |
| LLM Tier |
Sonnet |
| APIs |
None (reads ALL upstream outputs from DB) |
| Dependencies |
intel-company, intel-techstack, intel-traffic, intel-hiring, + optionally intel-investor, intel-social, intel-financial |
| Key Output |
tech_comparisons[], golden_angle_competitors[], traffic_comparisons[], financial_comparisons[], hiring_comparisons[], competitive_position, competitive_scenario |
| Production Edge |
Multi-dimensional comparison (tech, traffic, financial, hiring, sentiment). Golden Angle identification. Competitive scenario generation (GOLDEN/OFFENSIVE/DEFENSIVE/DISPLACEMENT). |
intel-queries (Test Query Generation)
| Aspect |
Detail |
| Layer |
Intelligence |
| LLM Tier |
Opus |
| APIs |
None (Claude only) |
| Dependencies |
intel-company, intel-techstack |
| Key Output |
prospect_queries[] (16 queries, 8 types x 2), competitor_query_sets[], difficulty_distribution |
| Query Types |
Exact match, category, comparison, price-conscious, ambiguous, typo, seasonal, long-tail |
| Validation |
9 checks: query count, type coverage, difficulty spread |
| Production Edge |
Vertically calibrated (industry/category-specific). Difficulty scoring. Competitor-specific queries. Dynamic generation (not hardcoded). |
Wave 2: Browser Testing (1 Module)
audit-browser (Playwright + Claude Vision)
| Aspect |
Detail |
| Layer |
Quality |
| LLM Tier |
Opus (Claude Vision for scoring) |
| APIs |
None (Playwright local browser automation) |
| Dependencies |
intel-company, intel-queries |
| Key Output |
query_results[], search_features[], dimension_scores[] (10 dimensions, 1-10 scale), detected_search_provider |
| 10 Dimensions |
Relevance, Completeness, Ranking, Autocomplete, Faceting, Filtering UX, Sorting, Typo Tolerance, Personalization, Speed |
| Timeout |
600 seconds |
| Production Edge |
Real browser automation (not API testing). Screenshot capture (evidence for AE). 10-dimension scoring via Claude Vision. Automated search provider detection. Rate limiting detection. |
Wave 3: Quality Gate (1 Module)
audit-factcheck (GAN-Inspired Verifier)
| Aspect |
Detail |
| Layer |
Quality |
| LLM Tier |
Opus (evaluation) |
| APIs |
None (reads all module_executions from DB) |
| Dependencies |
All upstream modules |
| Pattern |
Temporal child workflow (isolated context) |
| Key Output |
verified_claims[], conflicting_claims[], gate_verdict (PROCEED/WARN/BLOCKED) |
| Verdict Logic |
>15% contradicted = BLOCKED; >5% contradicted or >30% unverified = WARN; else PROCEED |
| Production Edge |
8-category batched verification. Cross-module contradiction detection. Advisory verdict (does not modify upstream data). Correction manifest for reference. |
Wave 4: Insights (1 Module)
insights-engine (Cross-Audit Benchmarking)
| Aspect |
Detail |
| Layer |
Operations |
| LLM Tier |
Sonnet |
| APIs |
None (DB reads only) |
| Pattern |
Fire-and-forget (does not block audit completion) |
| Key Output |
vertical metrics[], percentiles, sample_size |
| Persistence |
vertical_benchmarks table (idempotent: delete old, insert new) |
| Production Edge |
Historical aggregation across all audits in same vertical. Percentile computation. Anonymized. Idempotent. |
Wave 5: Synthesis (3 Modules, Parallel)
synth-business-case (ROI + Said vs Found)
| Aspect |
Detail |
| Layer |
Synthesis |
| LLM Tier |
Opus |
| APIs |
None (pure synthesis from upstream) |
| Dependencies |
intel-company, intel-investor, intel-industry, intel-competitors |
| Key Output |
said_vs_found[] (4-column matrix), roi_model, displacement_cost, customer_proofs[], timing_signals[] |
| Production Edge |
Said vs Found mapping (executive quotes linked to Algolia value props). ROI quantification. Displacement cost calculation. Industry case study matching. Multi-call synthesis (5 Claude calls). |
synth-sales-plays (MEDDPICC + SPIN Playbook)
| Aspect |
Detail |
| Layer |
Synthesis |
| LLM Tier |
Opus |
| APIs |
None (pure synthesis) |
| Dependencies |
intel-company, intel-hiring, intel-investor, intel-competitors |
| Key Output |
meddpicc[], spin_questions[], objection_handling[], talk_tracks[], power_map, executive_playbook[] |
| Validation |
>= 5 MEDDPICC elements AND >= 5 SPIN questions |
| Production Edge |
MEDDPICC framework. SPIN questioning. Power mapping. Objection library. Role-specific talk tracks. |
campaign-abx (Multi-Touch ABX Campaign)
| Aspect |
Detail |
| Layer |
Delivery |
| LLM Tier |
Opus |
| APIs |
None (pure synthesis) |
| Dependencies |
synth-business-case, synth-sales-plays, intel-hiring |
| Key Output |
email_sequence[] (5 emails), linkedin_strategy, linkedin_messages[], loom_script, collateral_schedule, competitor_messaging |
| Validation |
10 checks |
| Production Edge |
Complete 5-email sequence. LinkedIn strategy. Loom video script. Competitor-specific messaging. Personalization tokens. Collateral calendar. |
Wave 6: Final Report (1 Module)
audit-report (Scored Assessment + Deliverables)
| Aspect |
Detail |
| Layer |
Delivery |
| LLM Tier |
Opus |
| APIs |
None (pure synthesis from ALL upstream) |
| Dependencies |
synth-business-case, synth-sales-plays, intel-competitors |
| Key Output |
overall_score (0-100), dimension_scores[] (10 dimensions), competitor_scores[], pre_call_brief, leave_behind, full_audit_json |
| Validation |
10 checks |
| Production Edge |
10-dimension scoring with competitor benchmarking. AE pre-call brief (60-second read). Prospect-safe leave-behind. Full JSON export. |
Module Communication Patterns
- Database reads. Synthesis modules read upstream outputs from module_executions table.
- Account table. intel-company writes to accounts table; downstream modules read Account.competitors, Account.executives, etc.
- ExecutionContext passing. Shared context (audit_id, domain, company_name, ticker, is_private) flows to all modules.
- No direct module-to-module calls. All coordination via database or shared ExecutionContext.