Every weeknight after the market closes, an AI analyst reviews the day's price data for a fixed list of stocks and publishes what you see here: its read on the market's overall mood, a few trade ideas it would consider, and the names it's deliberately skipping. No real money is involved. Every idea is logged the night it's made and later graded against what prices actually did — the scorecard is the honest tally. Only a strong track record moves the project to its next phase.

Market regime

The market's overall mood. Risk-on: investors are buying confidently. Risk-off: they're playing defense. Mixed or neutral: no clear direction.

Confidence

How strongly the analyst believes in a call, from 0–100%. It's tracked over time to see whether high-confidence calls actually do better (see Calibration).

Long / Short

Long (green) means the analyst expects the price to rise. Short (red) means it expects the price to fall.

Entry

The price the analyst wants to get in at. If the stock never touches that price within a few days, the idea simply expires unused — counted as "never filled."

Stop

The get-out price if the trade goes the wrong way. It caps how much a losing idea can lose.

Target

The price where the idea pays off. Reaching the target before the stop counts as a win.

R : R

Reward-to-risk ratio: potential gain divided by potential loss. 2.0 means the distance to the target is twice the distance to the stop — risking $1 to make $2.

Suggested cash

The share of a hypothetical portfolio the analyst would keep in cash tonight rather than in trades. High cash means caution.

Risk / Horizon

Risk is the analyst's 1–10 read of how volatile or fragile the setup is. Horizon is how many days the idea has to play out before it expires.

Standing aside

Stocks the analyst looked at and chose to avoid tonight, with its reasons. Skipping a trade is a decision too — and it gets logged like one.

Hit rate

Of all the ideas graded so far, the percentage that reached their target before their stop.

Profit factor

Total gains divided by total losses across graded ideas. Above 1.0, the wins outweigh the losses; below it, they don't.

Avg win / loss

The average percentage gained on winning ideas and lost on losing ones. A few big wins can outweigh several small losses — that's why hit rate alone doesn't tell the story.

Calibration

A self-check: do the analyst's confidence numbers mean anything? If 70%-confidence calls win no more often than 55% ones, the confidence is noise — and that's worth knowing.

Brier score

The standard yardstick for confidence claims (borrowed from weather forecasting, 1950). Each call's stated confidence is scored against what actually happened; the average error is the Brier score. Lower is better — always saying 50% scores 0.25, and a perfect forecaster scores 0. "Skill" compares against the laziest honest strategy: always stating the era's own win rate. Positive skill means the confidence numbers carry real information; negative means they mislead.

Graded

After enough time passes, each idea is scored automatically against real market prices. Nothing is cherry-picked, and misses stay on the record.

Stand down / Report filed

Nights with no calls still appear in the ledger. An amber STAND DOWN row means the system chose not to trade into a binary event (CPI, FOMC) per its rules — discipline, on the record. A gray REPORT FILED row means the analyst looked and found nothing worth proposing. A date missing entirely means something failed — the staleness warning at the top of the page exists for exactly that case. Silence is never ambiguous.

Lanes (v2, v3, ...)

More than one analyst version can run at once, each as its own sealed lane: its own nightly report, its own ledger from zero, its own gate. Lanes never share or blend records — each is judged alone, against the same rules and the same frozen v1 benchmark. Every lane ever run stays on this page, including the ones that wash out. A new lane exists only after its design and its predicted results are written down in advance.

v1 / v2

Analyst versions. v1 (Jul 18–31) traded with a broken yield feed and no macro awareness — its final numbers are frozen as the benchmark. v2 (Aug 3 onward) sees real macro data and stands down before major events. The headline tiles and the Phase 2 gate count v2 only; the ledger shows both, labeled.

Decision ledger

The complete list of every call ever made — wins, losses, expirations, and ideas that never filled — newest first. It never gets edited or trimmed. This running record is the whole point of Phase 1.