Holdings rows you can read at a glance, run pages that say what each role produced, and an AI budget that grows with your books

What changed for people using Bellwether

• Holdings rows you can read at a glance. Under each holding on Portfolio, why the desk holds it, today's signals, what would make it sell, and the stop and target are a short labeled list instead of a block of gray prose. A long desk note shows two lines and a More control; on a phone each holding is shorter.
• Run pages say what each role produced. A Researcher, daily-review, Critic or Quant run no longer shows an empty "What it decided and why" card. It reads in that role's terms: the brief's headline and calls to check, the review's positions and lessons, the Critic's verdicts, the Quant's measurements, each linked to its source.
• The daily AI budget grows with your books. The desk's daily spending ceiling for model calls now follows the number of active books, under a hard stop the owner sets, so adding a book no longer starves the others by mid-morning. Money for the sessions that protect positions is set aside first; briefs and chat never spend it.
• A session the budget refused says so, everywhere. When the day's AI budget is spent, the Overview says "Desk paused" with since when and until when, the schedule marks the remaining check-ins "won't run" with the reason, and one alert tells the owner in plain words, a stronger one if the end-of-day check itself could not run.
• Admin › Usage opens with today's AI budget. Owners see the day's ceiling and the hard stop, what each book has spent and has left, what is kept aside for end-of-day checks, how many sessions were skipped for budget and each book's state, as of a stated time; the monthly view uses the same ceiling.

What you may notice

• The stop line under a holding leads with the figure ("$412.00 armed, trails up, price 1.2% above it") and the target line no longer repeats its number; "1 order waiting" is the one warm phrase on the row.
• Positions rows on the Overview wrap a long cell onto a second line instead of letting two cells print on top of each other on a narrower laptop screen.
• A key typed while the search field had lost focus no longer opens the Consult drawer: single-letter shortcuts are gone; the two-key shortcuts remain.
• Group labels on run pages are two or three words ("Calls to check", "Watching for", "Positions reviewed", "Lessons taken", "Market read"); an idea the desk has not acted on reads "waiting for you", an outcome not yet known reads "too early to tell".
• The schedule names the 14:48 slot "End-of-day check", and a skipped or refused slot carries its reason as a sentence under the name rather than a fragment after a dot.
• The help article on postures says how a busier cadence raises the day's AI cost and what happens when the budget runs out; the platform estimates each posture's typical daily cost from your own recent sessions, though the posture card does not show that figure yet.
• Browser tabs and bookmarks show the Bellwether mark everywhere; an older icon request no longer comes back empty.

Known gaps and what’s next

• The hard stop over the derived AI budget ships at the figure set on 2026-09-14; a higher one is the owner's call and changes nothing else.
• The Desk timeline's "Later today" line does not yet say in one sentence that the remaining check-ins are paused for budget; the schedule, the desk cards, the mind card and the shell already do.
• Two refinements follow: a permanent record of "Positions reviewed" on older daily reviews, and one settled look for a bare price level inside a quoted sentence.
• The phone Portfolio is shorter per holding but still long on an eight-name book; the numbers grid and the signals strip above the detail are the next places to save height.