← HR.com inventory

PRODUCTION OVERNIGHT RUN — Deep Buyer Intelligence for All 63 HR.com Targets

Source: deal-docs/04-prompts/production-overnight-all-63.md

PRODUCTION OVERNIGHT RUN — Deep Buyer Intelligence for All 63 HR.com Targets

MISSION

Run the full buyer intelligence pipeline against all 63 targets in the deal_research table (asset_type = 'Buyer Target'). For each company, build a dossier so detailed that Ewing can cold-call their VP of Corp Dev and reference specific quotes, specific filings, and specific strategic gaps that HR.com fills.

This is NOT a test run. This is production. Run all night. Do not stop until all 63 are complete.

HR.COM ASSETS (what we are selling)

COMPANY CLASSIFICATION & SEARCH STRATEGY

Each company falls into one of four categories. The search strategy changes based on type:

PUBLIC Companies (44 targets — earnings calls available)

Search 8 quarters of earnings call transcripts (Q1 2024 – Q4 2025):
- "{company_name} earnings call transcript Q1 2025" (repeat for Q2, Q3, Q4 2025 and Q1-Q4 2024)
- "{company_name} investor day presentation transcript"
- "{company_name} annual report 10-K SEC filing"
- "{ceo_name} {company_name} strategy interview podcast"
- "{company_name} acquisition announced 2024 2025"
- "{company_name} HR technology partnership"
- "{company_name} community platform ecosystem"

For PUBLIC companies, also retrieve from EDGAR:
- Most recent 10-K or 20-F (for foreign filers)
- Most recent 10-Q or 6-K
- Extract: revenue, revenue growth, EBITDA, strategic priorities, risk factors mentioning "community" or "audience" or "engagement" or "HR", M&A intent language, cash position

PE_BACKED Companies (12 targets — no earnings calls)

PRIVATE Companies (7 targets — CEO interviews and press)

CONSULTING / BIG 4 (3 targets — practice announcements)

Companies That Are ACTUALLY PRIVATE (misclassified as PUBLIC)

These companies are listed as PUBLIC in the database but are actually private. Treat them as PRIVATE:
Anthropic, Cohere, OpenAI, BambooHR, Culture Amp, Deel, Greenhouse, Lattice, Bevy, Mighty Networks, Khoros, Remote, Rippling

DATA ACQUISITION PROTOCOL (8 categories — ALL mandatory per company)

1. GO-TO-MARKET

How do they sell? Revenue model? Sales motion (enterprise/PLG/channel)? ACV range? ICP? Market position in HCM?

2. CHALLENGES (buying signals)

Publicly stated challenges from earnings calls or interviews. Look for: "headwinds," "competitive pressure," "churn," "engagement," "retention," "community." Extract:
- Publicly stated challenges (array)
- Exact earnings call phrases (array)
- Competitive pressures
- Growth constraints

3. VISION & STRATEGY

CEO's stated 2-3 year direction. Strategic priorities. Investment themes. Transformation narrative. Source from earnings calls, keynotes, podcasts, investor presentations.

4. ACQUISITION HISTORY (last 5 years)

Every deal since 2021: target name, deal value, date, stated rationale (exact quote from PR), outcome, source URL. Then characterize: bolt-on? Platform rollup? Audience aggregation? Appetite: active/selective/dormant?

5. SEC FILINGS (PUBLIC companies only)

From most recent 10-K/20-F and 10-Q/6-K:
- Revenue, growth rate, EBITDA
- HCM segment performance (if applicable)
- Strategic priorities from MD&A mentioning community, ecosystem, platform, audience, engagement, HR, workforce, talent
- Risk factors mentioning competitive threats in HR/HCM
- M&A language — capital allocation, pipeline mentions
- Cash position

6. DIRECT QUOTES (minimum 5 per company, 10 for top targets)

Exact verbatim quotes. Never paraphrase. Each quote MUST have:

{
  "speaker": "Full name",
  "title": "Title at company",
  "quote": "EXACT verbatim — do NOT paraphrase",
  "source_url": "URL where published",
  "date": "YYYY-MM-DD",
  "context": "Earnings call Q3 2025 / keynote / interview / etc.",
  "relevance_to_hrcom": "Which HR.com asset this connects to and why"
}

No source URL = don't include the quote.

7. PR & ANALYST COVERAGE

For each acquisition in #4: original press release URL, analyst/media commentary, ratings.

8. FIT NARRATIVE & STRATEGIC THESIS

STRATEGIC LOGIC (2-3 sentences): Why they need what HR.com has, using THEIR OWN WORDS.

ASSET MAP (1 paragraph): Which HR.com assets solve which of their challenges. Every claim references evidence.

BUSINESS MODEL INTEGRATION (2-3 paragraphs): Exactly how they'd deploy each HR.com asset inside their existing products. Name their actual products. Explain the revenue mechanism. Example level of specificity:
"SAP sells HCM to large enterprise customers. HR.com is the ultimate community to embed within SAP SuccessFactors. If the community alone is worth staying within, you get unlimited shots on goal to provide value ahead of selling them your solution."

COMPETITIVE BLOCK (1-2 sentences): What happens if a competitor acquires HR.com instead.

COLD CALL OPENER: One sentence for the corp dev VP's ear in the first 10 seconds. Must reference a specific quote or challenge.

EMAIL HOOK: One sentence for the CEO. Specific enough they think "this person understands our business."

GOLDEN NUGGETS (3-5): The best quotes paired with the HR.com asset that solves the stated pain. Each nugget includes: quote, speaker, quarter/source, why_golden, cold_call_opener.

PAIN MAPPING: For each publicly stated pain point, map it to the specific HR.com asset that solves it, with the source quote and a cold call line.

VISION MAPPING: For each stated strategic priority, map it to the HR.com asset that accelerates it.

CONVERGENCE POINTS: Where pain AND vision intersect — the strongest arguments for acquisition.

FIT SCORE: 1-100 with evidence-based rationale:
- 80-100: Multiple assets directly solve stated challenges. Acquisition pattern consistent. Quotes practically describe needing HR.com.
- 60-79: At least 2 assets address needs. Direction aligns. Some M&A appetite.
- 40-59: General alignment but limited direct evidence.
- 20-39: Tangential.
- 1-19: No meaningful alignment.

TWO-PASS EXECUTION STRATEGY

PASS 1: Light research on all 63 companies

For every company:
1. Run Exa searches (8 queries per company, 5 results each, max_chars=8000)
2. For PUBLIC companies: fetch latest 10-K/20-F from EDGAR
3. Extract with Mistral (researcher_1): 8 categories
4. Cross-check with DeepSeek (researcher_2): same 8 categories
5. Merge with Groq (inspector_1): flag contradictions
6. Synthesize with Claude CLI (synthesizer): fit narrative + golden nuggets + scoring
7. PATCH results to Supabase
8. Save JSON locally
9. Log costs

PASS 2: Deep dive on top 20 by fit_score

After Pass 1 completes for all 63:
1. Sort by fit_score descending
2. For the top 20 scorers, run enhanced research:
- Additional Exa searches (CEO podcast interviews, conference keynotes, blog posts)
- Full pain_mapping (5+ items with source quotes)
- Full vision_mapping (5+ items)
- Convergence points (3+ items)
- Expanded quotes (target 10 per company)
- Business model integration with named products
3. Update Supabase rows with enriched data
4. Update JSON files

LLM ARCHITECTURE (production combo — no benchmarking)

Exa.ai (web search) → raw research
     ↓
EDGAR API (SEC filings) → raw filing text (PUBLIC only)
     ↓
┌─────────────────────────────────────────┐
│ DUAL RESEARCH (parallel)                │
│ Mistral (researcher_1) — 96/100 score  │
│ DeepSeek (researcher_2) — 95/100 score │
│ Both extract same 8 categories          │
└──────────────┬──────────────────────────┘
               ↓
┌─────────────────────────────────────────┐
│ INSPECTION (Groq — inspector_1)         │
│ Merges researcher_1 + researcher_2      │
│ Flags contradictions                    │
│ FREE, fast (3-8s)                       │
└──────────────┬──────────────────────────┘
               ↓
┌─────────────────────────────────────────┐
│ CERTIFICATION (OpenAI — inspector_2)    │
│ Validates all fields                    │
│ Checks for fabrications                 │
│ Assigns confidence levels              │
└──────────────┬──────────────────────────┘
               ↓
┌─────────────────────────────────────────┐
│ SYNTHESIS (Claude CLI — synthesizer)    │
│ Writes fit narrative + scoring          │
│ Generates golden nuggets                │
│ Writes cold call opener + email hook    │
│ If fails: DUAL FALLBACK               │
│   → Run BOTH OpenAI AND DeepSeek       │
│   → Cross-check, use best result       │
│ FREE via subscription                  │
└─────────────────────────────────────────┘

If any two agents disagree on a FACT, include BOTH versions with sources and FLAG it.

API KEYS

WHERE TO STORE RESULTS

Supabase (HISTORICAL — pipeline completed, OLD instance decommissioned)

⚠️ BOTH OLD URLS (asavljgcnresdnadblse and rdnnhxhohwjucvjwbwch) ARE NOW DEAD. Use dwrnfpjcvydhmhnvyzov only.

Table: deal_research

PATCH via /rest/v1/deal_research?company_name=eq.{company_name}&asset_type=eq.Buyer Target

Field Type What goes here
story_narrative TEXT Full multi-section narrative: FIT NARRATIVE, BUSINESS MODEL INTEGRATION, COMPETITIVE MOAT, COLD CALL SCRIPT, EMAIL HOOK, RECENT STRATEGIC MOVES, GOLDEN NUGGETS, PAIN MAPPING, VISION MAPPING, CONVERGENCE POINTS
confidence TEXT HIGH if fit_score >= 70, MEDIUM if >= 40, LOW if < 40
call_opener TEXT Cold call opener sentence
revenue TEXT Latest annual revenue with source
employees TEXT Employee count with source
business_strength TEXT Primary strength category

⚠️ This table feeds the Lovable app LIVE at the Debbie Deal Room. Whatever you write here shows up immediately.

Table: buyer_dossiers

UPSERT via /rest/v1/buyer_dossiers (use slug as unique key):

Field Type What goes here
slug TEXT kebab-case company name (unique)
company_name TEXT Full company name
ticker TEXT Stock ticker (or null for private)
buyer_type TEXT PUBLIC / PE_BACKED / PRIVATE
ceo_name TEXT CEO name
exa_research_raw JSONB All Exa results with URLs
sec_filing_raw JSONB SEC filing extractions
analysis_dual_raw JSONB Both researcher outputs
validation_report JSONB Inspector merge + contradictions
certification JSONB Certified data with confidence
certified_challenges TEXT Stated challenges from evidence
certified_ceo_vision TEXT CEO direction with quotes
certified_key_quotes TEXT All quotes with attribution
certified_acquisition_history TEXT M&A history with outcomes
certified_acquisition_appetite TEXT active/selective/dormant
certified_hr_tech_presence TEXT HCM market position
certified_community_strategy TEXT Community/ecosystem approach
certified_fit_signals TEXT Specific signals HR.com addresses
certified_market_position TEXT Market position detail
certified_revenue TEXT Revenue with source
certified_employees TEXT Employee count with source
quotes JSONB Array of quote objects
quote_count INTEGER Number of verified quotes
fit_narrative TEXT Full narrative
fit_score INTEGER 1-100
fit_rationale TEXT Evidence-based rationale
fit_synthesis_raw JSONB Full synthesis output
pipeline_status TEXT QUEUED → RUNNING → SUCCESS / FAILED
research_cost_usd FLOAT Total API cost
completed_at TIMESTAMP ISO timestamp

Supabase Column Type Rules

Cost Logging

Log every API call to dossier_cost_log table:
- api_name, operation, cost_usd, tokens_in, tokens_out, model, provider

Local Storage

GitHub

RATE LIMITS & DELAYS

RESILIENCE & CHECKPOINTING

  1. Progress file: After each company, write status to data/overnight_progress.json:
    json {"sap-successfactors": "done", "workday": "done", "oracle-hcm": "running", "ukg": "queued"}
  2. Resume capability: If interrupted, read progress file and skip "done" companies
  3. Try/except per company: One failure does not stop the batch
  4. Retry pass: After all 63 attempted, retry any "failed" companies once
  5. Cost tracking: Running total logged after each company
  6. Graceful shutdown: On KeyboardInterrupt, save progress and exit cleanly

CRITICAL RULES

  1. NEVER fabricate quotes. Report how many you found and where you looked.
  2. EVERY quote must have a source_url. No URL = don't include it.
  3. EVERY financial figure must cite its source. No naked numbers.
  4. If two agents disagree, include BOTH with sources and FLAG it.
  5. Use python3 (not python) on this machine.
  6. Log all API costs. Track every call.
  7. Update BOTH deal_research AND buyer_dossiers tables.
  8. Commit and push to GitHub after completion.
  9. Do NOT create new Supabase projects, API accounts, or skills.
  10. Report what you actually found, not what you expected. Mark DeChant will verify.
  11. SAP SuccessFactors is already done. Skip it (or use as template for quality comparison).
  12. Do not stop between companies. This runs overnight unattended.

EXECUTION ORDER

  1. Read all 63 companies from deal_research where asset_type = 'Buyer Target'
  2. Classify each as PUBLIC / PE_BACKED / PRIVATE / CONSULTING
  3. PASS 1 (all 63): For each company sequentially:
    a. Exa research (8 queries)
    b. SEC filing extraction (PUBLIC only)
    c. Dual analysis (Mistral + DeepSeek parallel)
    d. Inspection (Groq merge)
    e. Certification (OpenAI)
    f. Synthesis (Claude CLI → dual fallback)
    g. PATCH deal_research + UPSERT buyer_dossiers
    h. Save local JSON
    i. Log cost
    j. Update progress file
    k. Print: [{n}/63] {company_name} — fit_score: {score} — cost: ${cost} — {status}
  4. PASS 2 (top 20): Sort by fit_score, deep-dive top 20
  5. Retry any failures from Pass 1
  6. Commit and push to GitHub
  7. Print final summary: companies completed, failed, total cost, top 10 by fit_score

Estimated cost: $4-6 total for all 63 companies.
Estimated time: 4-6 hours.

Start now. Begin with Pass 1.