Markets intelligence on the AI, defense, and energy build-out.

🌐 Live now at hardpowerintel.com

The site is up and running, and today's briefings are free to read right now — no account required to see them. Paid subscriptions are on the way, but the checkout isn't wired to a payment processor yet, so for the moment everything on the site is open. If you're an investor, analyst, or just curious, go read a live brief and come back.

Imagine hiring a sharp research analyst whose entire job is to read the boring-but-important stuff — thousands of pages of government contracts, corporate filings, regulatory notices, and industry news — every single day, across three industries: artificial intelligence, defense, and energy. Instead of handing you a pile of raw documents, this analyst hands you a single short briefing that leads with the point, tells you which companies each development actually affects, and shows you exactly where every fact came from so you can check the work yourself.

That analyst is what Hard Power Intelligence automates.

Most people who follow these markets don't have too little information — they have far too much of it, scattered across dozens of government websites and news feeds, written in dense bureaucratic language, and disconnected from what it means for the companies and stocks involved. A defense contract award, an energy regulator's licensing decision, a chipmaker's filing — each is public, but reading them all, connecting them, and figuring out what changed and why it matters is a full-time job. HPI does that work on a schedule and delivers the result as a clean briefing.

Three ideas make it trustworthy rather than just another news summarizer:

• It leads with the bottom line. Each briefing is written "BLUF" — Bottom Line Up Front, a style borrowed from military and intelligence reporting. You get the conclusion first, then the supporting detail, so you're not hunting for the point.
• It shows its sources. Every item links back to the original document — the actual contract, filing, or regulatory notice — so nothing has to be taken on faith. You can always click through and verify it yourself.
• It labels how sure it is. Every item is graded by confidence: a confirmed fact from a primary record looks different from reported news, which looks different from HPI's own analysis or speculation. You always know whether you're reading a hard fact or an interpretation.

The name is a play on three kinds of power: hard power (defense), electrical power (energy), and compute power (AI) — the three forces this product tracks, and increasingly the three forces that overlap. A single company can sit at the intersection of all three, and spotting those overlaps is where HPI aims to be most valuable.

HPI is a research and reporting tool, not investment advice and not a stock-picker. It tells you what happened and why it might matter; what you do with that is up to you.

Hard Power Intelligence (HPI) turns scattered public data — SEC filings, federal contract awards, energy and macro releases, regulatory actions, and news — into source-attributed, confidence-graded BLUF intelligence briefings for investors tracking the three strategic build-outs reshaping global markets:

• AI Infrastructure — the multi-year compute build-out: chips, data centers, hyperscaler capex, and the power demand it creates.
• Defense — next-generation systems: drones and autonomy, space, directed energy, electronic warfare, and the rebuilding defense-industrial base.
• Energy — new-tech and dominance trends: nuclear (including SMRs), alternatives, the grid, R&D, and energy-security themes.

The name is a triple play on power — hard power (defense), electrical power (energy), and compute power (AI) — the three forces this product tracks.

HPI is a recurring intelligence-production engine, not a chatbot and not a stock picker. On a cadence, it ingests disparate public sources, resolves every mention to the investable entity behind it, and synthesizes a professional Bottom-Line-Up-Front (BLUF) brief where every item shows its basis and confidence — from confirmed primary records to attributed reporting and clearly-labeled HPI analysis. One shared brief per sector desk serves all subscribers, so the marginal cost of another reader is near zero.

The product is positioned as source-attributed research and reporting — an informational publication, not personalized investment advice.

"An intelligence officer for the industries rebuilding national power."

The data that moves these sectors — DoD contract awards, EDGAR filings, NRC licensing, hyperscaler capex, export-control actions, energy data — is public but fragmented, slow to read, and disconnected from its market meaning. Serious investors don't lack data; they lack the time to aggregate it and the tooling to connect a raw event to the security it affects. HPI does that aggregation, connection, and synthesis on a cadence — and shows its work.

See docs/PHILOSOPHY.md for the full thesis.

1. The transmission layer. HPI's core asset is an entity-resolution graph that links a raw event to the investable security and the thesis — e.g., a contract award to "Lockheed Martin Rotary and Mission Systems" resolves to LMT, its segment, its program, and (via supply-chain edges) the second-order beneficiaries most tools miss.
2. Provenance by construction. Every ingested record carries a source URL and timestamp, so every item can show its basis. Grounding is transparency the reader can check — not a filter that withholds important-but-not-airtight signal.
3. A published accuracy bar. A built-in eval harness flags fabricated specifics before publish and grades each item's confidence/attribution (confirmed → reported → HPI analysis → speculative), turning hallucination — the industry's weakness — into a visible trust signal.
4. "What changed" focus. Change detection over the graph means each brief is signal — what's new since yesterday — not a re-summary of the same headlines.
5. Wide-net intake, free-data backbone. HPI takes any on-thesis development with a linkable source; public primary records (USAspending, EDGAR, EIA, NRC, FRED, Congress.gov, BIS) are the free, credible backbone, not the only admissible source. The moat is judgment and the entity graph over a wide intake.

Pipeline shape: a per-source scheduler fetches each source at its own cadence → adapters emit immutable, hashed, cited records → entity resolution writes provenanced edges into a bitemporal graph → change-detection + graph-grounded RAG generate a cached BLUF brief per desk → an eval gate checks it for fabrication and grades each item's attribution before publish → the brief renders to web cards and PDF.

Full design in docs/ARCHITECTURE.md.

Web-first subscriptions billed through Lemon Squeezy as Merchant of Record (reader-app model — subscriptions are sold and managed on the web to avoid app-store commissions). Tiered: a free daily brief (current day), a Pro tier ($19/mo or $179/yr) with archive, entity 360, PDF export, and follows, and premium deep-dive reports in later cycles. Designed to run cheaply (~$80–230/mo infrastructure) and reach break-even at a small number of subscribers.

Deployed to production 2026-06-12 — the full stack is live end-to-end: web reader (Vercel) → API (Fly.io) → Supabase, with auth and a cited Defense brief at /desk/defense. Gates 1–8 closed. This is an active, early-stage build.

Since launch (2026-06-14): the product is organized around the Defense Tech ∩ AI ∩ Energy convergence thesis — its north-star (DECISIONS.md D060). The engine moved off fixtures onto live data and became genuinely multi-desk:

• Production ingestion runner (scripts/run_ingest.py, D057) — pulls fresh data through a retry/backoff fetcher with DB-level dedup, per-source cursors + accounting, a circuit breaker, and hot-window retention. The daily-brief.yml cron is live (06:00 UTC, ingest once → publish all three desks), made safe to run unattended by the reliability work below.
• Layered analyst brief (2026-06-16, D071/D073) — each item pairs the cited fact with a grounded read (why it's material) and watch (forward catalyst), plus a brief-level convergence_read, rendered as an "Analysis — HPI interpretation" drill-down (P3). The analysis is held to a grounding gate (regenerate-then-omit) so it adds interpretation without fabricating — depth that keeps the trust model. Reliability + freshness gates make daily publishing trustworthy: provable-claim publish floor (D070), regenerate-on-failure (D072), and a novelty/anti-rehash gate so tomorrow isn't a re-summary of today (D074).
• Four live adapters — USAspending (defense-tech awards, thematically scoped cross-agency not DoD-only, D059), SEC EDGAR (full-text search, the first cross-desk source — one adapter feeds all three desks, D061), arXiv (the technology-advancement leg, D066), and NRC (via the Federal Register API — the Energy desk's regulatory leg: nuclear/SMR licensing and HALEU fuel events, D095).
• Convergence-aware, multi-desk briefs — generation is desk-scoped across Defense / AI / Energy (D062); a record touching ≥2 desks is boosted as the convergence signal (D060). All three desks publish from live data with the layered analyst voice (BLUF → at-a-glance ledger → convergence → cited facts → grounded analysis → GDELT signal). A strategic-significance gate (D085) drops true-but-trivial items so a desk skips cleanly rather than publishing filler.
• Entity-resolution graph (live, D091/D092) — the moat (the "transmission layer"): a SEC-seeded reference entity set (~8k companies, de-duped by CIK) and a precision-first resolver held to an accuracy eval gate (precision 1.000 / recall 1.000 / zero false links on the golden set). Each brief item is resolved to the investable entities behind it and rendered as entity chips (ticker for public companies; a "private" chip for closely-held/venture firms, which are minted from authoritative ingest identifiers (UEI/CIK), not a curated list). Verified end-to-end on a live brief (every link correct, zero false positives). Chips click through to Entity 360 pages (/entity/[id] — identifiers, the desks an entity spans, recent appearances), and an entity seen on ≥2 desks is flagged as a cross-desk convergence signal — the moat, live end-to-end.
• Interactive Convergence Graph (live, D146–D153) — the entity graph made visible at /graph: an interactive, on-brand force-directed map where companies recurring together across desks cluster, with gold convergence nodes (spanning ≥2 desks) and gradient edges that blend the two sectors they bridge. Built as: cross-desk co-appearance edges (CONVERGES_WITH), a name-gazetteer that lifts linking past the identifier-only ceiling, a filtered graph API, the interactive viz, edge-hover that surfaces the cited stories behind each connection, and a toggleable federal-funding overlay (AWARDED agency→company edges from USAspending — the first structured semantic layer). Behavior is instrumented so demo reactions are never mistaken for demand.

Remaining before public launch: Lemon Squeezy go-live (the checkout/webhook code is built; needs account creds + variant IDs in Vercel/Fly — config, not engineering), a live-validation pass on the AI/Energy desks plus dedicated sources for depth (EIA/NRC, interconnection queues, more AI-infra), and the flagship cross-domain convergence brief. Auth config (Supabase Site URL) is fixed. Tracked in DEPLOYMENT_CONFIG.md §6 and CHANGELOG.md.

Scope boundaries in docs/SCOPE.md.

This repository is built under Meridian, an agent-harness framework that enforces a gate-by-gate workflow with schema-validated memory and independent evaluation. HPI also serves as a Meridian dogfooding exercise: its hybrid nature (a full-stack web app whose core is an ML/RAG eval pipeline) exercises Meridian's composable gate DAG. Build progress is tracked through .meridian/gates.yaml.

Hard Power Intelligence produces informational research, not investment advice. Nothing it generates is a recommendation to buy or sell any security. Always do your own research.

Source-available under the PolyForm Noncommercial License 1.0.0. You're welcome to read, study, fork, and build on this code for any noncommercial purpose. Commercial use requires a separate license from the author — please get in touch. This is not an OSI "open source" license; commercial rights are reserved.

Paul Christopher Schmidt — @PCSchmidt