How AlphaPod Works
Deterministic Data.
Evidence-Graded Everywhere.
Every number on AlphaPod is computed by versioned SQL and Python from source rows — never by a language model. Models only write prose, and only from numbers that are already computed, validated, and gated for publish.
Get the Map FreeEvery U.S. AI datacenter project — tracked from permits, grid filings, and emissions data
The Process
What Happens Behind Every Number
Ingestion, entity resolution, and derivation are deterministic — zero LLM cost. Language models enter only at the last step, and only to write prose.
1. Ingest
A scheduled Lambda fleet pulls permits, grid filings, WARN/layoff notices, npm/PyPI downloads, token-price series, and SEC filings daily to quarterly. Every row is stamped with a source_tier (T1–T4) and, for capacity rows, a capacity_basis.
2. Resolve
An entity-resolution graph maps project→parent→ticker, employer→ticker, and package→ticker, with human-review provenance. Plausibility rules quarantine implausible rows — a parent tracking more GW than the largest hyperscaler, a labor event hitting >20% of headcount — before they can render.
3. Derive
Versioned SQL and Python compute capacity rollups, the labor EPS bridge, and the developer-commoditization triangle — zero LLM cost, fully reproducible, and backtested against subsequently reported outcomes once enough quarters exist.
4. Publish
Language models write prose only — Brief sections and dossier narratives — built strictly from the numbers already computed. Every run passes an evidence-sufficiency check, schema validation, numeric grounding, and an entity whitelist before it reaches you.
Coverage Universe
600+ U.S. IT & Communications Stocks
One entity graph and one covered-ticker universe across all three surfaces — from mega-cap platforms to emerging growth names.
Software & Cloud
MSFT, CRM, NOW, ADBE
Semiconductors
NVDA, AMD, AVGO, INTC
Internet & Media
GOOG, META, NFLX, SNAP
Hardware & Devices
AAPL, DELL, HPQ, WDC
Cybersecurity
CRWD, PANW, ZS, FTNT
Fintech
SQ, PYPL, COIN, AFRM
Telecom
T, VZ, TMUS, LUMN
IT Services
ACN, IBM, CTSH, EPAM
One Data Spine, Three Surfaces
Who Wins the AI Buildout
Demand proven by permits, cost proven by labor filings, durability proven by adoption data — sharing one entity graph and one covered-ticker universe.
AI Infrastructure Intel
Every tracked project, evidence-graded T1 (permits, grid filings) through T4 (claims), mapped to the tickers that own the campuses plus the pull-through layers. Confirmed (T1+T2) capacity always leads; T3/T4 is labeled “Est. pipeline” and never rolled into a dollar headline.
Project map, exposure map, beneficiaries, capacity↔revenue, grid stress.
AI Disruption / Labor Exposure
WARN and layoff evidence — filing links, employees affected, hiring-velocity context — scored into a modeled per-ticker EPS impact (low/base/high) with an uncertainty grade and the assumptions shown. AI-related attribution only where the source discloses it.
Rolling backtest scoreboard once ≥2 quarters exist; a “calibration period” chip until then.
Developer Commoditization
A per-ticker triangle score, always shown with its three component series: de-noised package-adoption momentum, same-model token-price trend, and FinOps/quantization fingerprint counts. Concentration warnings render when a score rests on too few packages or repos.
Own backtest scoreboard against subsequently reported revenue and pricing.
Ticker Evidence Dossier
Cross-cutting: buildout exposure, labor exposure, and developer signals for one covered ticker on one page, every number linking to its primary source rows, with the evidence-sufficiency state shown next to it.
Weekly Buildout Brief
Cross-cutting: new confirmed projects, status and power milestones, labor events, developer signal changes, and entity-resolution changes — structured deltas across all three surfaces, every week. Free teaser; full issue at Pro.
Data Foundation
The Corpus Is the Moat
Point-in-time, evidence-tiered, and accumulating daily via the Lambda fleet. A competitor with a frontier model can't prompt this into existence — they have to run the pipes for a year.
The Corpus
- 618K+ source items
- 76,714-filing SEC manifest
- ~90 dc_* datacenter tables
- WARN + layoff history
AI Infrastructure
- Permits & grid filings
- EPA & emissions records
- Power-market readiness
- Capacity restatement history
Labor & Disruption
- WARN notices
- Layoff event filings
- Hiring-velocity signals
- Sector wage series
Developer Adoption
- npm / PyPI download series
- Same-model token-price history
- FinOps config fingerprints
- Concentration warnings
Evidence Architecture
Graph-Backed, Not Source-Exposing
Every figure on AlphaPod is tied back to evidence in our entity-resolution graph — for cross-source corroboration and audit-grade traceability. The graph is the differentiator, so the product surfaces evidence categories and graph-level signals, not the underlying providers, raw URLs, or parser internals.
Sanitized Evidence Categories
Permits, grid filings, WARN/layoff notices, package-download series, token-price history, and FinOps fingerprints — surfaced as labeled categories, not raw provider feeds.
No Source Exposure
Provider names, raw URLs, document IDs, parser names, model identifiers, prompt text, and pipeline internals never leave the platform. What you see is the graph-level conclusion and the evidence category — never the cookbook.
Audit-Grade Traceability
Every number in the Ticker Evidence Dossier and Weekly Brief links to its source rows and the derivation `version_id` used — enough to re-derive the figure yourself, without revealing how the graph was built.
Methodology Transparency
Honesty About Model Maturity, Built In
Transparency is easy for us — the spine is deterministic — and structurally hard for chatbot competitors to match.
Evidence Tiers, Defined
T1 (permits, grid filings) down to T4 (unverified claims). Operating capacity must be earned by site-specific evidence before it counts toward a confirmed figure.
Corrections Ledger
Every entity-resolution change, capacity restatement, dedupe merge, and tier reclassification is recorded append-only — who, when, what, why, and the source — and summarized in the Weekly Brief.
Backtest Scoreboards
The labor EPS bridge and the developer triangle both carry a rolling scoreboard against subsequently reported outcomes. Until enough quarters exist, the surface shows a "calibration period" chip — not false confidence.
The same evidence layer is also served where analysts already work — via MCP into Claude and ChatGPT — with free snapshot tools for anyone and full ticker-level access on Institutional.
Start With the Map — Free
Pin-level project detail, evidence-tier chips, and the public Weekly Brief teaser. No credit card required.
Get the Map Free