The desk's own numbers agree with each other before it trades

What changed for people using Bellwether

• The desk's own readings agree with each other. The blend score the portfolio manager sees in its decision table and the one its signal tool returns are now the same reading of the same finished bars, including for the names it is offered when the book is under-invested; a sentence above the table says so, so the desk stops spending its turns re-checking the platform's arithmetic. A finished day's reading no longer changes within a session, and when the basis really is different — a refreshed market ranking, a new release — the reading's identifier says which part changed instead of looking identical.
• One 52-week high everywhere. The factor read, the price-history tool, the nightly screen and the symbol page now compute the 52-week high and low the same way: the highest traded price of the last year of completed sessions. For dividend payers that is the actual print rather than a dividend-adjusted figure, so exits set "off the 52-week high" sit where the tape says they should; a name with less than a year of history says how many sessions its high covers instead of calling it a 52-week figure.
• The Quant's nightly check is dated correctly again. Each nightly run is stamped with the day it ran, reads the book's own marks, trades and orders through that day, and says plainly how far behind the bundled price history is, instead of dating itself to the last price bar and reporting an empty book. When the price history lags, the Quant is told which of its evidence refreshes only with a release and that any market-regime call must name the date of the tape it read.
• Three research ideas tested and retired, a fourth killed on review. Residual momentum, quiet-name weekly reversal and gross-profitability quality were each pre-registered, built point in time on the 500 most liquid US names and measured after costs; none showed an edge the platform would trade, and an independent kill review confirmed all three. A fourth idea, institutional-ownership breadth from quarterly holdings filings, read weak and was killed on review as momentum arriving seven weeks late. The Evidence page's registry now counts 146 configurations tested and no survivors; nothing in the live blend changed.
• The "restated" note stays, on evidence. The one sentence the signal tool leads with for a day and a half after the platform changes how it reads finished bars was trialled live against the same sessions without it: no session escalated a restated bar as a data problem in either arm, and the sentence's only measurable effect was that the desk named the change correctly more often. It stays as shipped.

What you may notice

• "Near the 52-week high" readings become a little rarer in the first sessions: distance from the high is now measured against the highest traded price, not the highest close, so most names read slightly further below it.
• Take-profits and exits the desk re-derives from the 52-week high on dividend-paying names can move up to the actual print; they fill later, not earlier.
• Recently listed names drop out of the screen's 52-week-high and deepest-below lists and show no 52-week figure on their page until they have a year of history.
• Asking the signal tool twice for the same name in one session returns the same finished-bar number by design; only the forming intraday bar is fresh.
• The nightly Quant summary shows the run's own date and a "price data N days behind" line until the bundled history is refreshed with a release.

Known gaps / coming next

• The instruction sentences that ship with this release (the note above the decision table, the Quant's data-lag line) were accepted on the scripted no-regression suite with a dated waiver each; the live checks that close them are the first three portfolio-manager sessions after the release and a live Quant scenario in the next evaluation stream.
• The desk's saved notes that still say "trust the tool, not the table" are closed by a one-time memory pass right after the release; until it runs the desk may re-verify a defect that no longer exists.
• A "no change tonight" card from the Quant, so a held verdict is visible instead of silent, is the next build in that lane.
• The price history bundled with each release still ages between releases; a scheduled refresh is proposed, not built.