Five lenses, one transform, and a record of whether it was right.
Nothing here is a black box. Every number on a holding page opens into the sub-metrics that produced it, and every reading states which lens carried it. This page walks through what happens, in order.
Each lens asks one question
They are deliberately different questions, because a stock that answers all five well is rare and a stock that answers them inconsistently is where the interesting reading lives.
| Lens | The question it asks | What it reads |
|---|---|---|
| Quant | Is this risky? | Volatility, correlation, drawdown, Sharpe and Sortino, size against the risk budget |
| Trend | Is momentum with it? | Moving averages, RSI and MACD, support and resistance, relative strength |
| Micro | Is this a good business at this price? | Business quality, financial trajectory, earnings momentum, and a bear / base / bull valuation range |
| Macro | Is the regime favourable? | Rates, the curve, the dollar, inflation trend, sector rotation — scaled by how sensitive the name is |
| Portfolio Fit | Does this belong in my book at this size? | Concentration, correlation to what you already own, cost basis, and how far the position has drifted |
Fit is a transform, not a fifth vote
The four quality lenses — Quant, Trend, Micro, Macro — are weight-averaged into a single conviction score: how attractive is this stock at today's price. That is a statement about the company, and nothing else.
Portfolio Fit then turns that conviction into a target weight for your book — how much of this stock your portfolio, as it actually is, has room for. The distance between that target and what you actually hold is what produces the reading.
A flat five-way average gets this case exactly backwards. A well-loved name grown to a quarter of the book reads Weakening on concentration under a transform, and Strengthening under an average. The first one is right.
The four lenses become one conviction score. When they disagree, the score says so rather than averaging the disagreement away — it is the thing the product is named after.
Conviction becomes a target weight for your book, never for a book in general. The same stock with the same conviction earns a smaller target in a book that already owns four things like it.
Six states, each with its reasoning attached
A holding is placed on a spectrum, always shown with the reasoning that produced it. It is a second opinion you can interrogate and overrule — analysis, not financial advice, and never an instruction about what to do.
| State | What it means |
|---|---|
| Strengthening | The analysis puts the target above what you currently hold. |
| Stable | Target and position agree closely — the difference is within the normal range. |
| Weakening | The position sits above the target the analysis reaches — often on concentration rather than on anything the business did. |
| Candidate | A name you are watching whose readings would earn it a target weight in your book. |
| Watching | On the list, not yet clearing the bar. |
| Review | Something you wrote down as a break condition has happened. This one outranks the other five. |
The thesis is yours. So are the conditions that would end it.
When you add a holding, TensionLine drafts the investment thesis it reads in the evidence — and then hands it to you to edit, because the reasoning you bought on is the one that matters. Alongside it you can write down what would change your mind: a price level, a drawdown, a number that has to keep being true.
Those conditions are checked on every run. When one is crossed, the holding moves to Review and the reading says which condition it was — the thing you said would break this thesis has happened. No interpretation of prose, no model in the loop: three comparisons against prices already stored.
Free watches the conditions you set. Pro reads the five lenses and the environment around them every day — the part you cannot write down in advance.
What the briefing contains
Your book is re-analysed overnight, on both plans. Each morning the briefing names what stands out and why, in your holding's own numbers: how much of the book sits outside its target range, which one or two names carry most of that distance, and — if you have logged cash — where the analysis would rank it.
A quiet book is a finding too, and the briefing says so rather than manufacturing something to show you.
An email when something changed — and nothing when it didn't
A break condition you wrote down has fired, or a position has drifted outside its band. The email gives the ticker, the measured value and your own threshold, because the judgment was yours to begin with.
A holding's state has changed, or a watchlist valuation has crossed its threshold. These are the engine's readings rather than yours, so the email says what moved and leaves the reasoning in the app, one click away.
Only changes are sent — a holding that was Weakening yesterday and is Weakening today produces nothing. The same event on the same holding is not repeated for seven days, so a weight hovering on a band edge cannot fill your inbox. No prices are ever in an email. There is no open tracking and no marketing mail.
How the prediction record works
A prediction is registered with its claim, its confidence, the lenses driving it, the criteria that will resolve it, and the date. Months later it resolves — correct or incorrect — against those criteria, and a short written post-mortem is composed at that moment, around the mechanical decomposition of what actually moved.
That resolved record is the only source of outcome labels in the product. It is what tunes how a stated confidence maps to a real probability, and what proposes — for your approval, never silently — how much weight each lens earns on your book.
When TensionLine says 80%, you will be able to check how often 80% actually meant 80%. The first figures publish once enough predictions have resolved; until then this site shows none, which is rather the point of publishing them.
Where the numbers come from
Market data comes from licensed providers and from public company filings, and economic series from the FRED® API. Splits and dividends are reconciled, and a position held through a split reports the share count it actually has.
Company narratives — news and earnings summaries — are written by a large language model and labelled as such wherever they appear, with their primary sources linked. The scoring itself is not a language model: the lenses are computed.
It reads. You decide, and you act elsewhere.
TensionLine executes nothing and has no broker connection. It never asks for brokerage credentials and could not use them. Every trade you make, you make in your own account.
A state is the balance of the lenses' analysis with its reasoning attached, not an instruction, and it is not tailored to your circumstances. The disclaimer says so on every screen.
When your own record suggests a lens deserves more or less weight on your book, the change is proposed for you to approve — never applied behind your back.
See it read your own book.
Public beta — FREE, opens on 15 October. No card. An analytical tool — not investment advice.