Sources & the Blend Workbench
The full workflow — record the books you follow, weight them by trust, get an explainable target, and measure your real account against it.
This is the heart of the product: the path from "here are some investors I rate" to "here are the four trades I should make this month, and here's why."
Four pages in the app, in order: Sources → Blend → Target → Reconcile. The Workbench is all four at once, live.
New to the vocabulary? Every bolded term links to its definition; the glossary has them in one place.
Step 1 — Record the books you follow
Most portfolio tools only know about your portfolio. But if you invest by studying people, their books are data too. Sources make them first-class.
A source is any reference portfolio that can feed your plan:
- a tracked manager — someone whose disclosures you follow,
- your own strategy — the book you'd hold if you were starting from scratch today,
- an AI sleeve — a portfolio the AI proposes, kept visibly separate until it earns trust.
Sources are inputs. Your actual brokerage holdings live under Portfolio, and the whole point of keeping them apart is being able to compare the two.
What to record, and what it buys you
For each source: the positions and their weights, the date, and where the information came from. That last part matters more than it sounds.
- Snapshots are dated and append-only. A new disclosure adds to the history; it never overwrites it. So you can compare any two dates and see entries, exits, adds and trims — next to any trades the manager actually disclosed. When a position vanishes with no disclosure to account for it, that gap is visible rather than invisible.
- Got something wrong? File a correction. It supersedes the original and both stay on the record. Nothing silently rewrites history.
- Provenance travels with every number — filing, interview, video, or inference — so a figure from an audited disclosure is never quietly treated as equal to one someone deduced.
- Freshness is computed, not typed. Tell the app how often a manager publishes and it will tell you when they're overdue. A source that never declared a cadence is never labeled fresh.
One company, one line
Managers name things the way they speak — a company's common name, a listing that trades under a different ticker where you live. The instrument aliases panel on the Sources page lets you confirm those mappings once, so the same company stops appearing as two positions. Unresolved names are surfaced for you to confirm; the system won't invent a mapping.
Doing this without the typing
Transcribing a manager's book by hand is exactly the drudgery this product exists to remove. Connect your AI assistant and hand it the disclosure — it records the snapshot, with provenance, under your account, and tells you what it did.
Step 2 — Weight them by trust
Open the Blend designer. A blend is your recipe: how much each source counts, plus the rules the engine follows.
Trust weights are the first half. They normalize to 100%, so 40/20/20/20 means what it looks like. This is the difference between following someone and copying them: hold a new manager at 10% while you make up your mind, keep one at 0% to stay informed without letting them into the plan, and raise a weight when their scorecard has earned it.
Construction rules are the second half — the constraints you'd otherwise re-apply by hand every single time:
| Rule | What it's for |
|---|---|
| Exclusions | Names you'll never hold — with a reason, and an explicit instruction for where the money goes: redistributed, moved to cash, or substituted with another ticker. |
| Position caps | A ceiling on any single position, a floor below which a line isn't worth holding, and a minimum trade size. |
| Overlap amplifier | Count it for extra when several sources independently hold the same name. Off by default, capped when on. |
| Account rules | Placement hints (this belongs in the tax-advantaged account) and holdings to preserve untouched. |
| Accessibility | Tickers you genuinely can't trade, removed with everything renormalized around them. |
| Normalization | How much of the portfolio is invested at all — 96% invested means a deliberate 4% cash reserve. |
The instruction on an exclusion is worth dwelling on. If a manager holds 5% cash as a deliberate defense and you exclude that line without saying "send it to cash", the money lands in equities and you've quietly discarded the defensive part of their strategy. The choice is explicit because getting it wrong is silent.
Every save is versioned with a timestamp and restorable, so your allocation policy is auditable months later.
Step 3 — Read the target
The target is what the blend produces: holdings and weights, a cash reserve, and the names your rules excluded with your reasons attached.
It is explainable by construction. Open any line and you see which sources contributed it and how much each contributed. "Why does the plan hold 9.5% of this?" always has an answer, and the answer is made of your own decisions.
The page distinguishes a live preview (regenerated from your current blend, not saved) from a saved snapshot (kept, named, timestamped). Reconcile against a saved one, so you're comparing against a plan that holds still.
It also shows effective bets: your position count after names that move together are collapsed into one bet, using the correlation groups you declared. Twenty-eight names can be twenty-three bets, with the largest single bet at 30% of the book — a number a position count will never show you.
Step 4 — Measure reality against it
Open Reconcile, pick a saved target and one account.
You get the drift, position by position, and the trade plan that closes it: buys, sells, sizes, and the net cash implication. Export to CSV or print, and run it at your broker. Nothing here places an order.
Reconcile runs against one account at a time, deliberately: a target expressed in percentages only means something against a specific pot of money.
And the questions underneath
The same page answers why do I own this? for every position:
- Conviction custodians — anchor each position to its reason: a source, your own thesis, or the system. Right after an import, everything is unanchored; that's setup, not an alarm.
- Orphans — positions no longer backed by anyone's conviction, e.g. the name a manager exited while you kept holding. The report flags; it never acts.
- Copy fidelity — how faithfully you're actually tracking a manager, measured against what you could realistically have bought.
- Scope — cash and deliberately off-strategy holdings can be declared out of scope, and are named and counted rather than silently skipped.
The Workbench: all of it at once
The Workbench page collapses steps 2–4 into a single live view. Drag a source's weight and the target, the drift against your chosen account, and the trade blotter all redraw together. Nothing is saved until you save a target — so the cost of asking "what if I halved this manager?" is zero.
Use the focused pages when you're editing carefully; use the Workbench when you're exploring.