AI portfolio intelligence
Own a portfolio you can explain.
Stock Rankings keeps your real holdings and the disclosed books of investors you trust in one place, blends them into a target you control, and turns the gap into trade proposals. You approve, refine, or deny — nothing trades on its own.
Early access is free and currently limited to invited accounts — sync, targets, and trade plans included. No card required.
You already do this work. Your tools just don't keep the receipts.
If you invest seriously, you follow a handful of investors whose judgment you rate, you keep a spreadsheet of what you hold, and a few times a year you try to reconcile the two. The spreadsheet forgets where every number came from. This doesn't — it was built so that every position, weight, and trade can answer for itself.
“Why do I own this?”
Anchor every position to the manager whose conviction put it there — its conviction custodian. A position with no reason left behind it is flagged as an orphan. Flagged, never sold for you.
Conviction custodians“Am I actually following them?”
Copy fidelity measures your book against each investor you meant to mirror, position by position — including what they hold that you don’t.
Copy fidelity“What did they quietly sell?”
Manager books are kept as dated, append-only snapshots. When a position disappears with no disclosed trade, the diff names it: the exit they never mentioned.
Snapshots“How many bets am I really making?”
Names that move together collapse into effective bets. A 28-stock portfolio can be 23 real bets — or far fewer. Diversification you can count, not guess.
Effective betsFrom raw accounts to an approved trade
One loop, always in this order. The system prepares everything; the only irreversible step is yours.
Connect
Sync your portfolio tracker, import a CSV, or type holdings in. Then record the disclosed books of investors you follow as sources — dated, append-only, with the origin of every number kept.
Blend
Weight those sources by how much you trust them, add your rules — exclusions, position caps, correlation — and the blend produces one explainable target. Every line traces back to who believes it, and how strongly.
Approve
Reconcile compares the target with what you actually hold and turns the drift into a concrete trade plan. Approve, refine, or deny each proposal. Nothing executes on its own.
Your assistant · connected over MCP
Here's the Q2 disclosure from the manager I follow — record it.
Recorded as snapshot #4 — append-only, #3 stays. Three changes: added CPNG at 2.1%, trimmed AAPL 6.1% → 4.8%, and SHOP is gone with no disclosed sale — flagged as an undisclosed exit.
Anchor my ASML position to their book.
Done — ASML's custodian is now that manager. Your copy fidelity against them is 91.4%.
What just happened
Snapshot #4 recorded
append-only · #3 kept
Diff vs #3
1 add · 1 trim · 1 exit
Undisclosed exit flagged
SHOP · no stated trade
ASML anchored
copy fidelity 91.4%
What's inside
Six pillars, one goal: higher intelligence on fresher, more accurate data. Each card links into its full documentation. Tier-marked pillars activate at public launch — in early access, every account runs on Free.
One true book, with receipts
Tracker sync, CSV import, or manual entry — every account converges on one versioned portfolio. Every number carries provenance (who said it, and when) and freshness that admits its age: a source that has gone quiet reads stale, never fresh-by-default.
Managers, on the record
Record what the investors you follow disclose. Snapshots never overwrite — corrections supersede, history stays. Diffs show every add, trim, and exit, with their stated reasoning attached verbatim.
Predictions, scored without flattery
Log a manager’s public claims verbatim, each with a falsifiable resolution condition. Scorecards refuse to show a hit rate below eight resolved calls — a small sample reads as a small sample, not a track record.
An AI Council on your real book
Chat 1:1 with AI portfolio managers, or put the whole Council on a trade idea and let them debate it — grounded in your actual holdings, not generic takes.
An assistant that does the typing
Connect Claude or any assistant that speaks MCP — the open protocol AI assistants use to work other software — to your portfolio. It files snapshots, fixes ticker mappings, anchors positions: executing what you state, never inferring your judgment. And it can never place a trade.
Execution that waits for you
The automated execution manager — built on the Alpaca brokerage — can only ever get a trade to proposed. Four hard guarantees stand between any idea and real money:
- Your approval required
- Hard-coded risk limits
- Kill switch
- Paper-first prove-out
Built to be checked, not believed
Software that sits next to your money should not ask for faith. These are structural properties of the system — the docs describe exactly how each one works, and the product shows them to you as you use it. Credentials you connect are encrypted at rest (AES-256-GCM) and used only on your behalf — the privacy policy spells it out.
Approval is a hard boundary
There is no path from proposal to execution that skips you. Real-money execution adds four more locks on top, including a kill switch.
The execution lifecycleHonesty is the default output
Unknown reads unknown. Unrecorded reads not recorded. Stale reads stale. The system admits a gap rather than fabricate a number to fill it.
How freshness is computedEverything is versioned
Portfolios, rules, agents — every change keeps its history. Corrections supersede; nothing is silently overwritten. You can always see what changed, and when.
Append-only snapshotsThe AI has hands, not judgment
Your assistant executes the instructions you give it. What you believe, and what you are willing to risk, stay yours to say — it never guesses at either.
What it can never doThe system does the doing. You do the deciding.
That contract is enforced in code, not promised in copy. Start with the docs and check it yourself — then bring your book.