Glossary

A first-time purchase of a stock you don't hold, proposed by the Opportunity Finder after a qualified name dips and the dip's quality clears the bar.

Buying more of something you already hold. The Control Tower proposes it only through gates — trend strength, a short-term health check, the position cap, and a healthy market.

Sell a quarter of the position. The ladder's lightest sell — scale back, keep most of the exposure.

Sell half. A decisive de-risk that still keeps a runner in case the position recovers.

Close the whole position — the ladder's verdict that the original reason for owning it is gone.

An exit triggered by age, not by loss: a dip-buy on the agent book still open after its maximum holding period is asked out, profitable or not. Labelled separately so discipline is never confused with a risk breach.

A pre-set exit slightly below the 200-day line, attached to anything you keep after a trim. If the trend truly breaks, the position leaves without a new debate.

A deliberately small top-up on a market dip — sized by rule, never by enthusiasm.

You said yes. On the agent account that executes the trade; on the main account it records your commitment to trade by hand. Graded from your actual fill forward.

You said no to the whole ask. Nothing trades. The recommendation is still graded — if it would have lost, your rejection counts in your favour.

You approved a batch but left this item out — a per-item no inside an overall yes. Graded like a rejection.

You didn't respond and the approval link expired. Nothing trades. The scorecard remembers it as silence rather than a decision — and still shows what the ask would have been worth.

Main-account only: you approved in the app, but no matching trade ever appeared at the broker. Held apart from both agreement and refusal, because it is neither.

An internal state: the engine generated this recommendation but you were never asked — typically an off-schedule test run. Hidden from your ledger and your metrics; you'd only meet it on a deep-linked trace.

How Judex treats the main account: it never places orders there. Your approvals are recorded, the trading pen stays in your hand, and the reconciler verifies what you did.

The recommendation itself, graded forward from the moment it was made: what following it exactly at the suggested price would be worth today. This — not the closed trade's profit — is how the system is judged.

Your verdict graded from your actual fill forward: what agreeing was worth compared to not acting. If you skipped the recommendation instead, your grade is its mirror — skipping a right recommendation cost you, skipping a wrong one saved you. Moves with the price — the trace labels it "so far".

The difference between the price the engine saw and the price you actually got. Manual delay on the main account, and even automatic fills, both create one — and it's recorded, because it's real money.

The whole trade's result, buy to sell, in booked money. It mostly grades the original purchase — which is why it lives under "what it meant to your portfolio", not under the recommendation's verdict.

For things that didn't happen — a rejected buy, an ignored trim — the honest what-if, computed from the suggested price to now. Always labelled; never real money.

Of the recommendations actually delivered to you, how many are currently right, judged forward from their suggested price. Recommendations still too close to their suggestion to grade sit out — of both sides of the ratio. The system grading itself, in public.

A call whose price hasn't yet left a narrow band around the suggestion — most often one made today, still priced off its own bar. Graded neither right nor wrong (the scorecard shows —), and left out of the hit rate until it moves.

Of everything you were asked, how much you agreed to. Not a target — a mirror. The scorecard pairs it with what following (or not) actually earned.

The value of winning calls you skipped — counted once per idea, however many times it re-asked. The price of your no.

One day's asking of one idea. A persistent signal re-asks daily, and each ask is its own row — "if it repeats, it repeats" — because each was a real decision moment.

Finder buy ideas nobody acted on, shown from your side: what skipping each one saved or cost. Each idea appears once, dated by its latest signal, with its full signal span.

On an untaken idea: when it first asked, when it last asked, and how many times in between. Persistence you can read at a glance.

Five pre-agreed rungs — Watch, Caution, Reduce & review, Urgent, Hard breach — by how far a holding is under water. Each rung escalates what you're asked. The thresholds were chosen before the loss, when nobody was upset; the live values are on the Home strategy card.

The average closing price over roughly the last ten trading months — the market's most-watched boundary between a healthy long-term trend and a broken one. Closing below it escalates a holding to Urgent, and protective stops anchor to it.

The medium-term trend line. A Finder name must be above it on the day it first qualifies (enter on strength); after that, dipping below it is tolerated — often that is the buyable dip. Qualification is lost only when the moving averages un-stack or the price closes below the 200-day line.

The market's overall weather, read from the big indexes: healthy, cautionary, or broken. When it's broken, all new buying stands down — no exceptions for good stories.

The Tower's strength bar for adding to a holding. A weak trend blocks the add, and the block names the gate.

No single name may grow past a set share of the account. An overweight position gets its trims bumped one tier heavier — concentration is de-risked first.

A position too small to be worth managing — tiny both as a share of the book and in absolute money — is exited whole rather than trimmed into crumbs.

Gate one: strong enough to deserve a dip. Scored on trend, business quality, strength versus the market, and momentum; only high composite scores make the watchlist. Weakness alone never qualifies anything.

Gate two: a qualified name dipped, and the dip was examined — market-wide, sector-wide, or company damage? Damage is rejected outright. Every evaluation writes a score and a one-sentence reason you can read in the trace.

Gate three: the day's best-quality dips — at most a handful — sized and placed in the approval email. The only Finder output that asks for your verdict.

How far a stock has fallen from its recent high — the Finder's trigger and its sizing input: deeper dips commit more capital, inside hard caps.

Whether the stock has beaten the market itself over recent months. The "strength" the Finder insists on before it will buy weakness.

A 0–100 momentum gauge. Healthy momentum qualifies; overbought extremes block buying even in strong names; collapse-grade oversold readings are treated as panic, not opportunity.

No buying into an imminent earnings announcement, however attractive the dip — announcement night is a coin flip, and the Finder doesn't flip coins.

The daily ceiling on new buying: a maximum number of buys and a maximum share of available cash, however many signals fire. Deferred names re-signal on their next dip.

The pool the Finder scans daily — roughly a hundred large, liquid quality companies. Everything starts here; almost nothing survives the gates on any given day.

Whether the broker currently allows the trade at all. A candidate that isn't tradable is flagged rather than offered.

The afternoon state of your world: what changed, portfolio health, and any positions on the ladder — each with its reason and a sized action. "No action required" is a real sentence it sends.

The hunt's daily report in funnel order — Selected, Evaluated, Qualified — each name with its reason. Reading, not homework: the actions live in the approval email.

The one email that asks. A single-use link to one screen with every proposed action, its size, and its reason. Approve, reject, or let it lapse — all three are recorded outcomes.

After you approve a main-account action, gentle nudges follow until the reconciler sees your trade — because approved-but-never-traded is a real state, and it's better caught early.

Once a month, the whole loop closed in one email: what was asked, what you did, what both were worth — the same honest arithmetic as the Scorecard page.

The count of things currently awaiting your verdict, shown on Home and in the briefs. Zero is a perfectly good number.

Your primary eToro account. Judex watches it and asks about it but never trades it — every action there is record-only, executed by you.

The account Judex may trade with your approval — where Finder buys execute and the time-stop applies. The experiment runs on the experimental book.

The account filter every page shares. All is the default everywhere; Main and Agent narrow the view. Owner-level things (approvals, emails) only appear under All.

The live queue: everything currently awaiting your verdict, with the same single-use token the email carries. Once decided or expired, items leave this screen — history lives in Records.

The history hub: Recommendations (every ask, episode by episode), Trades (your buys and sells), Approvals (your responses), and Emails (everything sent). Each row opens its trace.

All record types interleaved into one timeline — the story of each idea in order: recommended, you acted, emailed.

The drill-down behind any row: what was recommended, the data and rules it used, the run that produced it, the emails that disclosed it, your response, and what it all meant. The audit trail, one click deep.

The built-in assistant, running on your own AI key. It answers from your data — why a signal fired, what a trace means, what's waiting for you — and any action it stages still needs your explicit confirmation.

The price the engine saw at the moment it made the recommendation — the anchor for grading the recommendation, and the reference your execution is compared against.

The current market quote, fetched live when you open a trace — so "so far" verdicts are as fresh as the screen you're reading.

One end-to-end execution of the daily machinery, with one id threading it all — fetch, judge, brief, ask. Every recommendation and email carries its run id, which is why everything is traceable.

The exact strategy file a recommendation was computed under, fingerprinted and stored. Rules change over time; every rec remembers which rules judged it. Live values: the strategy card on Home.