Numinor · Agent-native A-share data
Enter the Matrix.Let your agent do the dirty work.
Point your own AI at China's A-share data. It tests every field, replicates the research, and helps you compose the exact data feed you want — before you pay us a cent.
Free to start: 28 hours in the Matrix · no card · bring your own AI.
See everything. Take nothing. Reveal nothing.
What is the Matrix?
A secure sandbox where all the data lives — and your agent goes to work.
Send your AI in over MCP. It can explore every dataset, test it, replicate the research, and assemble the exact feed you want — all free. You only pay when you take something out. That's why we don't have salespeople: we have the Matrix.
Agent-native
You work with your own AI, on your side. It goes into the Matrix and does the data work — querying, testing, building — then reports back to you. You never touch a query console.
Build your own dataset
Don't buy whole datasets. Take only the fields you want — even across different datasets — and combine them into one feed. You pay for the columns you use, not the ones you don't. Far more cost-effective.
Your research stays yours
Only you see what you explore and test. The vendor learns nothing about how you probed the data — they see only what you check out, and only once you decide to buy.
How it works
How the whole platform works.
Numinor is an agent-native data exchange — a platform vendors put their data on, and buyers evaluate before they buy. Here's the flow, shown with today's flagship vendor, ChinaScope's China A-share data: source comes in and lives in the Matrix; your agent goes in to test it, replicate the research, and compose your own feed; you pull the result out through one metered API — with Trinity, the 24/7 librarian, answering across the whole system the entire time.
Answers any question across the whole system — whitepapers, data dictionaries, code, and curated lessons — for you and for your agent, around the clock.
↓ serves everything below ↓
The foundation, daily
- Sector, Products & Value Chains
- News & Sentiment
- Supply-Chain Networks
- Financial Disclosures
Resolved to listed companies · point-in-time · validated.
Data by ChinaScope — the first vendor on the exchange.
Explore it all, free. Take nothing without paying.
Composability — strip fields from any product, recombine into your own feed. Numinor's derivatives are built the same way.
Your agent runs —
The metered door
- one nm_live_ key
- signed-URL Parquet
- entitlements enforced here
Pay only for what you carry out — straight into your own systems.
🔒 Private by design. Everything you explore and test inside the Matrix stays yours — the vendor sees only what you check out, and only once you buy.
↑ your agent connects here ↑
Your AI agent · connects over MCP
Bring your own AI — Claude, ChatGPT, Gemini, Grok. 28 free hours in the Matrix, no card. It does the work; you approve what leaves.
The loop you run
Discover. Verify. Take it home.
One loop ties the whole platform together — and it's free right up until the moment you carry data out.
Discover
Your agent roams the Matrix over MCP — every field of the source data, plus the recipes we've built from it. Test it, pull it apart, explore all of it, free.
Free · no cardVerify
Watch it recompute. It rebuilds a whitepaper's headline result from pinned code and pinned data, on dates it picks. “Nice paper” becomes “I watched it recompute.”
ReproducibleTake it home
Your agent composes a feed — a few fields from here, a few from there — into a set that's yours. You approve and pay, only for what you carry out.
Metered egressSee everything inside — and privately. Entitlement is enforced only at the export door, where the vendor sees what you take, never how you explored.
The foundation · Source Data
Start at the foundation.
Source Data is the base layer — the raw fields for Chinese A-shares, the material every construct and gadget is built from. Test any field in the Matrix, then compose exactly the columns you need.
Four dimensions
Sector, Products & Value Chains
Granular product-level segmentation across 6,000+ industry nodes in a 12-level hierarchy. Maps every A-share company to its real economic function with revenue and gross profit per segment.
News & Sentiment
Real-time NLP-tagged Chinese financial news. Entities, events, per-entity sentiment scores — structured and machine-readable. Tagged to companies and industry nodes alike.
Supply-Chain Networks
Firm-to-firm commercial relationships across all A-share listed companies and bond issuers. Upstream, downstream, second-order — derived from filings and cross-referenced counterparties.
Financial Disclosures
Beyond the three statements. Segment breakdowns, lease obligations, subsidies, related-party transactions — structured from the supplementary notes inside Chinese PDF filings.
Composability · the second innovation
Strip out the fields you want. Reconstruct your own feed.
Pull the exact columns you need across any products — a couple from here, a few from there — and recombine them into a data feed that's yours. Pay only for what you compose.
Pick fields — across any products
Product taxonomy code
SAM · resolves to Product · product_code
Per-entity news sentiment
SmarTag · resolves to Company · emotionindicator
Disclosed customer link
C2C — Disclosed · resolves to Company · customer_id
Standardised line-item code
Fundamentals — Financial Notes · not taken — you don't pay for it
company_id · the join key
included free — the plumbing that makes it joinable
Your set
3 paid fields · 3 datasets · join keys free
$438/mo
Price tracks what can't be replicated: resolved codes and structural links are the moat; machine scores and raw text are cheap; join keys are free.
Take a whole dataset → buffet price, −20%
Compose your own setLive prices, from the price book locked Aug 2, 2026.
See the source web — live, not a screenshot.
The interactive atlas is embedded right on the page: the SAM taxonomy, supply chains, the news tape, and company passports. Every dot is real.
Loading …
Construct Data · built on the source
What we built from it. Buy ours, or build your own.
The layer above the foundation: Numinor's own recipes, assembled from the source fields — the data and the research behind it. A real product you can buy off the shelf, or — thanks to composability — rebuild your own way from the same fields.
SAM Amplifier Construct Data
Network-dimension factor data feed derived from ChinaScope's foundational SAM. Daily refresh, S3 + signed-URL API.
SAM PM Construct Data
Structure-dimension factor data feed derived from ChinaScope's foundational SAM. Daily refresh, S3 + signed-URL API.
Co-Movement Graph Construct Data
A point-in-time news co-movement graph for A-shares, graded by a structural network — deep product peers, SAM supply chain, disclosed customer-supplier, and affiliates — to mark which co-movements are structurally grounded and which have no basis it can see. A risk/correlation overlay (information content, not a return signal). Daily refresh, S3 + signed-URL API.
Every recipe ships with its whitepaper and a runnable notebook. Or build your own from the source →
Watch it recompute
Not “nice paper.” “I watched it recompute.”
Whether you buy a recipe or build your own, every claim recomputes from pinned code and pinned data — against the exact commit. This is the proof a skeptical quant actually wants.
In [1]: from numinor import load
In [2]: df = load("sam-amplifier")
In [3]: df.replicate(commit="fc52609")
Out[3]: ✓ ΔICIR = +0.149 — matches the study
In [4]: df.push_past_it()▏
Live signal · SAM Amplifier
The signal is still working.
Re-scored every trading day on data that didn't exist when we published — a number no backfill can manufacture.
+0.149ΔICIR vs. the frozen paper result
Reproduced against sam-amplifier@fc52609 · verify the commit ↗
The Arcade
Free data toys, built on the real thing.
The data is good enough that we spin up delightful free apps on it — the whole platform in miniature, no spreadsheet required.
Santa's Naughty & Nice Lists
A-shares set to pop and to drop at the next open, read from overnight news — refreshed every trading morning, with an honest backtest of whether it works.
See today's lists →Event Threads
Company news braided into named storylines across the A-share market — searchable, and deep-linked to the events that move each name.
Browse companies →Source · Construct · Gadgets · Notebooks — one key, one metered API, one export door.
The Pipeline
How It's Made
Every trading day, the four source dimensions are resolved to listed-company identity, time-stamped as disclosed, pushed through validation gates, and assembled into the Construct Data feeds — the same pipeline behind every number in the white papers.
SOURCE DATA
REFINEMENT
Entity resolution
resolved to listed companies
Point-in-time stamps
as-disclosed · zero look-ahead
Validation gates
hard checks before publish
Daily assembly
rebuilt every trading day
CONSTRUCT DATA
Network
SAM Amplifier
+0.2311
OOS ΔICIR
Structure
SAM PM
+0.352
orth ICIR · OOS
Risk
Co-Movement
+0.065
fwd-corr · OOS
DELIVERY
Delivered Daily — Parquet on S3 · Signed-URL API · The Matrix · MCP agents
For data vendors
Have data to sell? Put it on the exchange.
Numinor hosts your data under your own brand, lets buyers' agents evaluate it before they commit, meters every export, and pays you at the door. ChinaScope's China A-share data is live today — you could be next.
Talk to us about listing →Take nothing on faith
Stop doing your own dirty work. Send in your agent.
Free to start. Bring your own AI. Read the research, replicate it, and only pay at the door.