Trader UX review — the user is watching a live tape, not your feature

A trader's attention is the scarcest resource in this product. Every review question reduces to: does this cost attention, clicks, or trust at a moment the market is moving?

Judge from screenshots and flows produced via visual-qa / ui-harness, plus reading the interaction code. Review flows, not screens: entering a trade, adjusting a stop, switching timeframes — walk each affected flow start to finish as each persona.

The personas

Review as all three; they disagree, and the disagreements are the findings.

Rafa — order-flow scalper (WIN, tick/volume bars, seconds matter). Eyes locked on tape + book + bubbles. Decides in under a second, acts by hotkey, never by menu. Anything that interrupts the stream mid-decision — a popup stealing focus, a confirm dialog on a flatten, a layout jump on bar close — is money lost. Judges every control by "can I hit it without looking?" and every piece of information by "is it true now?".

Marina — swing/context trader (BTC + index futures, multiple timeframes). Works across tabs and split panes; compares, measures, draws. Cares about consistency (the same gesture must mean the same thing in every pane), persistence (her drawings, presets and layouts must survive a restart), and honest history (backfilled or inferred data clearly marked). Tolerates menus; does not tolerate relearning.

Duda — newcomer on paper trading (learning the platform). First week. Doesn't know the hotkeys or the jargon. Needs discoverability (can she find the action without the manual?), safe defaults (a mis-click must not do something irreversible), and states that explain themselves (a disabled button with no reason reads as a bug). If Duda misreads a label as financial fact — a simulated P&L that looks real, an inferred side shown as truth — that is the worst finding in this file.

Heuristics (order-flow specific)

• Interruption budget: zero. Nothing steals keyboard focus or covers price/tape/forming bar uninvited. Confirmations only for destructive, irreversible acts — and never on the exit path of a losing trade (flatten must be instant; precedent: flatten-on-reset, Shift+X).
• Action cost. Count clicks/keys from intent to done for the critical actions (enter, stop, flatten, switch symbol/timeframe). Critical = 1 gesture. Compare against the flow before the change: any regression in gesture count is a finding.
• Glance cost. The key number of the feature must be readable from the chart without leaning in, at the moment it matters. Where does the eye have to travel?
• Trust. Data honesty is UX: inferred, simulated, delayed or incomplete data is labelled at the point of reading, not in a tooltip. Sim results say sim; tape-proven fills say how they filled.
• Stability under fire. Fast tape, bar close, feed reconnect: does the layout hold still? Do controls stay where muscle memory expects them?
• Latency is UX. A frame hitch during a fast tape is an interruption — Rafa reads flow from motion, and a stutter breaks the read exactly when the information peaks. A feature that looks perfect in a screenshot but drops frames under a dense tape fails Rafa even with zero visual findings. Repo priority order applies here too: never trade runtime for prettiness.
• One language. Same chips, same button pairs, same placement grammar as existing surfaces. A trader configures once and expects the dialect everywhere.

Severity — in trader terms

• Blocker — can cost money or the moment: focus theft mid-tape, an extra gesture on flatten/stop, simulated or inferred data readable as real, an irreversible act one mis-click away.
• Should-fix — costs attention or trust: layout jump, inconsistent gesture, unexplained disabled state, key number needing a menu.
• Consider — polish that would make a persona faster but blocks no one.

Output

Per affected flow: the persona walk (what each persona hits, in one or two sentences each — only where they diverge), then findings ranked by severity, each with the flow, the persona who suffers, the evidence (screenshot or file:line), and the concrete fix. Close with one line per persona: can Rafa trade through it, can Marina keep her workspace, can Duda figure it out alone? A clean review says so briefly — never pad.

Blockers and Should-fixes feed the same gate as arch-review: resolve or explicitly defer in the PR body.