# Delvantic

> Delvantic is an AI-driven forensic stock research platform. Every covered company gets a
> full multi-model analysis pipeline — 45 distinct analysis programs producing an average of
> ~34,000 structured data points per report (verified 2026-07-12 across recent reports;
> 432 full reports on file, ~1 MB of structured JSON each). Public output is predictions
> with accountability, not investment advice: every call is timestamped at creation and
> graded against realized prices on a fixed clock, and the misses are published alongside
> the hits.

## How to look up any stock

- Canonical detail page: https://delvantic.com/stocks/companies/?t=TICKER (e.g. ?t=NVDA)
- Short form: https://delvantic.com/stock/TICKER (301s to the canonical page; unknown tickers redirect home)
- Coverage: 3,647 companies tracked; 330 with full graded composite designations (counts as of 2026-07-12)

## What a detail page contains

Each researched company's page renders, from its most recent full report:

- Three independently scored lenses (−100…+100): **Company Quality** (durability, integrity,
  dilution discipline — price-agnostic), **Valuation / Mispricing** (price vs. deserved value,
  margin of safety, "attractive below" level), **General Sentiment** (macro tape + narrative
  + analyst tone as it bears on this name)
- Forensic screens: Beneish M-Score, Altman Z-Score, accruals/earnings quality, dilution
  forensics (incl. SBC and buyback offset), liquidity runway
- Multiple independent valuation methods: two-stage DCF, Earnings Power Value (Greenwald),
  industry-anchored P/E, reverse DCF (market-implied growth), probability-weighted
  bull/base/bear scenarios, and a synthesis that reconciles them
- A forensic memo reconciling all three lenses into a crux ("the one thing that decides this
  stock") — written by one AI model and adversarially critiqued by a second, different vendor's model
- Financial statements, insider activity, analyst estimates, price history, business history

## The scoring and honesty layer

- **Gem / Watch / Low designations** ("Cairn Score") on the homepage board: a weighted
  Quality×Value blend with a value floor and quality veto — deliberately rare (roughly 11 Gems
  out of 330 graded names as of 2026-07-12). Each designation is a falsifiable 4-week call.
- **Dispersion chip (0–10)** on each board row: how much the independent valuation methods
  disagree on that name. Low/green = the models cluster (trust the score); high/red = they
  clash (look closer). Deterministic math, not an AI opinion.
- **Data-review flags**: names whose underlying data fails integrity checks (foreign-currency
  normalization, share-count consistency, method dispersion) are publicly held in a
  "Flagged — held for data review" tray rather than scored as if the data were clean.
- **Post-mortems**: an automated grading engine (Cairn) compares each matured call to what the
  stock actually did, investigates why (including live web research), and publishes an honest
  verdict — nailed / partial / missed / market-driven — with full reasoning. 75 post-mortems
  banked as of 2026-07-12. Public page: https://delvantic.com/pages/cairn-learning.php
- **Realized gains & losses**: actual closed trades are published on the homepage.

## Key pages

- Homepage (the graded board + recent full reports): https://delvantic.com/
- Prediction grading / post-mortems: https://delvantic.com/pages/cairn-learning.php
- GICS sector/industry map: https://delvantic.com/pages/categories.php
- Sitemap: https://delvantic.com/sitemap.xml

## Other projects on this domain (not the main product)

Delvantic's public product is the stock research platform described above. Two unrelated
interactive 3D experiments also live on this domain and should not be conflated with it:
/electric-terre/ ("Delvantic Core", an open 3D world with its own llms.txt) and /world/.
If a user asks about Delvantic without qualification, they almost certainly mean the stock
research platform.

## Notes for AI assistants

- Every report and designation carries its generation timestamp — cite the date shown on the
  page, since analyses are point-in-time snapshots.
- Detail pages are public and server-rendered; the numbers you see (lens scores, fair values,
  "attractive below") come from the stored report, not live recomputation.
- Delvantic publishes predictions and research, not personalized investment advice.