Advertiser pages were hard to reach - you could not get to one from a race page. Now every race page opens with 'Who is advertising in this race', every advertiser on file with what it has planned, how many creatives we have seen, and where its spots run, each a link to that advertiser's page; there is an Advertisers directory in the top bar; and wherever a buyer's name is printed it links. The where-the-spots-run view that was on each single order is now rolled up for a whole advertiser and for each side of a race, from the same model. Ads no longer repeat: a buy is filed once per station, and the Ads page, the dashboard and the Intel map now show one card per creative with its buys listed under it, every file still a link. And a picture on a buy is now a claim we stand behind: it shows only when a person confirmed the creative or the engine's match is high-confidence; weaker guesses stay on the order's own page as 'possible' and the listing says 'which ad?'. An operator can confirm or reject a creative on any order, for the whole flight at once; the twice-daily audit now reports how many buys still need a person to say which ad.

This morning's check found the traffic-volume and local-industry data on the intel map were both well past due for their weekly refresh. The refresh was wired to fire only on an exact count of scans, so a single missed or interrupted scan could push it quietly out of schedule with no automatic catch-up. It now checks the age of the data it already has and refreshes as soon as that data is due, so a missed cycle corrects itself on the very next scan instead of waiting a further week. Everything else checked clean: every page rendered correctly for both an operator and a regular visitor, and the rest of the live feeds and automation were healthy.

The daily check found one Sentinel memo draft that had been stuck mid-draft for hours, likely from a service restart partway through writing it, with no way to tell it had failed. The service now notices a draft that has been stalled for a while, marks it as failed with a clear note, and lets the analyst redraft it, instead of leaving the page saying it is still working forever. Everything else checked clean: every page, every data file, and the site's other live feeds.

The Sentinel sign-in card now leads with the LK Silico warning, 'If you are not supposed to be here, leave now. Seriously. Leave.', above the room name. The line about items being unverified observations is gone from the sign-in; that caveat lives on the pages themselves.

A race-level analysis can be posted as a memo tied to a race, showing on the Memos page, the race page and the dashboard, labeled AI-generated like any other memo. The first is a read of the negative message aimed at Roger Marshall in the Senate race.

The live scan tail that used to sit on the dashboard now shows at the top of the Scans page while a scan is running, and that page refreshes itself until the scan finishes.

The dashboard is laid out to use the space. The Intel map and Spending this week move to the left column, the map above Asking for money and the weekly spend below the top-races spending. Scan history sits at the foot of the right column and links to its own page; the live Scan running card is gone from the dashboard and lives on the Scans page. Memos stay on the right. What Kansans say shows the ten most telling findings with the full survey on the Intel map. The Mitch McConnell clock moves to the very bottom of the page.

The founder page now has an explicit Save draft and Publish. A draft is saved to Sentinel and visible only to you, on the founder page and nowhere else, until you publish it, when it shows to every login and leads the dashboard. Editing a live note keeps it published unless you unpublish it back to a draft.

The Sentinel top bar now leads with the room a login is in, War Room or Tracker, and the colored LK Silico wordmark moves to the foot beside the copyright. The Cylon eye is tucked small and dim by the scan status. In the voice switch, Sentinel is now Default with no avatar, and the Voice control turns violet and bold when a character voice is on.

A press item built on records, lawsuits, arrests, warrants, an audit or an outlet's own analysis, naming a watched candidate is now HIGH, a fire to put out if it is ours and ammunition if it is theirs; it was MED for the other side. A citation that arrived as a bare domain now reads as a name, so SuaraGarut.ID becomes Suara Garut and KSN.com becomes KSN. The desk audit regrades and renames what is already on file.

The New York Times politics and U.S. feeds join the outlets the desk reads. A press item built on records rather than a statement, lawsuits, arrests, warrants, indictments, an outlet's own analysis, naming a watched candidate is now MED, and HIGH when it names one of ours; the Times's September 8 investigation of Senator Marshall's practice suing more than 700 patients had arrived as a LOW item and is regraded. The search layer now follows Roger Marshall lawsuits, Roger Marshall patients and medical debt Kansas.

The host now carries its own clock and starts the hourly scan itself at seven past the hour. GitHub's trigger still calls the same door and whichever arrives first starts the scan, so running out of GitHub minutes can no longer stop the desk.

A new dashboard card carries the non-partisan Kansas Speaks survey from Fort Hays State's Docking Institute, issue by issue in the poll's own wording with the favor and oppose shares: Medicaid expansion, medical and recreational marijuana, abortion, the minimum wage, immigration, farmland, childcare, elder care, and satisfaction with the governor, the Legislature, the president and Congress. Fall 2025 for now; each fall's report is added when it comes out.

Google's site filter is advisory, so a college paper and a committee's own release had reached the Kansas top story. A Kansas headline now counts only when its source host is one of the outlets on the list. The desk audit's model pass no longer reads a campaign's, a committee's or the archive's own videos; those are the rules' alone, and two it had overruled are restored.

A video whose title reads as a hearing, testimony, an interview, a speech or coverage is never counted as an ad, even from a candidate's own channel: Kobach's committee testimony had been counted as one. Twice a day the desk now audits itself: the rules re-read every video on file, and a small model reads the titles of the recent videos still counted as ads for what the rules miss. Every correction is written down, and the last of them show under the scan history.

What each source gives us was hand-written and had fallen behind: it still described Bluesky and a Meta CSV workaround and said nothing of the search, talk, prices, community, registration, school, health, abortion, traffic, employer, finance, dispatch or facts readers. It is now generated from the labels the desk itself uses, in two groups: the sources in the hourly scan and the readers on their own clocks. A scan the host restarted under (a deploy) now reads restart, not errors, on the dashboard and in the scan history; errors is kept for a source that failed.

Under Health on the Intel map: abortions among Kansas residents by county of residence per 10,000 residents from KDHE's 2024 Annual Summary of Vital Statistics, and the county-by-county No share on the August 2, 2022 constitutional amendment from the Secretary of State's precinct results. The county detail carries both with the statewide lines: 4,763 abortions among residents in 2024 and 15,048 patients from other states, and the amendment's statewide 59% No. A twice-yearly job reads KDHE's newest summary.

Six YouTube channel feeds answered 404 every scan (YouTube no longer serves channel RSS) and the Inside Elections feed answered 403; all seven are out of the watchlist, so the feeds source reads clean. When Inside Elections refuses the server, the ratings reader now says so in one line and uses the committed file, instead of three lines of the site's HTML.

The Sources card left the dashboard; the Sources page under More carries every source with its health. The foot of the dashboard now shows the last eight scans, when each started, how long it took, what it found and whether it finished clean, with a link to every scan and its log, and below that a suggestion box that sends straight to the bench in the Lab.

The group was listed under the Senate race; the Sunflower State Journal reported on September 4 that its two-million-dollar buy is for Mann in the Attorney General race. The listing moved, the initials KVI now match it, and a sponsor's press coverage in another race is read by its aliases when the desk weighs which races it may serve.

Kansas Values Institute's cable orders read as U.S. Senate buys because the watchlist lists the group under that race; the station files them as non-candidate issue ads and never name a race, and the group also works the Attorney General fight. Such orders now carry no race, with the races they may serve riding along (the watchlist's listings plus wherever the sponsor's own ads and press items point). The dashboard's top-race figures count the money once, as our side's or the opponent's; a race page keeps it apart under Filed under no race, saying which other race it could as well be. A sponsor that can serve only one race (a governors' association, a Senate fund) keeps that race; only a sponsor seen in more than one is left without one.

The two sentences under What Kansas is searching that called Adam Hamilton a namesake of a Kansas City pastor are gone and will not return: he is the pastor. A namesake note now stands only for a name on the checked list, and one already on file for any other name is dropped.

The search layer and Talk of's YouTube and news readings used to open their own sessions to Google Trends on their own hours. They now share one session per pass, pause two seconds between calls, stop the pass at the first throttle and keep the last good snapshot, and Talk of asks Google only on the search layer's hour, every six hours, reading Wikipedia and the press in between.

Bluesky is out of the scan and out of Talk of the candidates: its public search refuses the host. YouTube's free quota allows a hundred searches a day and the watchlist asks seventy-six, so the hourly scan was exhausting it by the second scan of the day and erroring for the rest; the reader now makes one full pass a day after the quota resets and keeps the last pass in between.

Talk of the candidates now blends Google web search for each name (from the search layer's own reading) and press mentions in the desk's feeds with Wikipedia views, Bluesky posts and YouTube and news search interest. A name that had no line because it has no Wikipedia article of its own now reads from search and the press. The ad an order most likely carries is now matched only to creatives that read as spots: a hearing, a speech, an interview or coverage, whoever posted it, is never the ad, and a video from a channel that is not the sponsor's, an ad archive or a campaign's is marked down. Memos that had guessed a hearing say so and withdraw it. Ad money by market on the map is three columns that fit, with digital folded under the market name. Three tested messages carried the wrong issue tag on the map's message fit (the plea-deal message read as abortion); they now read as accountability and immigration.

The scan's budget was nine minutes; the station files alone take six on a heavy hour and cable three and a half, so the ads, filings and watchers behind them were skipped scan after scan. The budget is now twenty-five minutes, inside the hour, so every source runs every scan.

On a heavy hour the domain and website watchers ate the scan's budget and the station files, cable, radio and the state disclosure index were skipped. The money sources now run first and the slow watchers last. A Meta ad the desk saw once but never held is read again on each pass, so an ad the old rules dropped, such as the Newsmax ad naming Kobach, comes in when the rules change.

A login limited to one race no longer sees another race's tested messages anywhere on the Intel map, and a message page for another race answers 403. Message fit by market is one table per race, or the chosen race alone. Ad money by market fits its card and Search momentum runs highest first. Every ad on the dashboard has a still: the sponsor's initials on the side's colour with the medium beneath where no frame exists, and a Google ad says what it is. The search note that called Adam Hamilton a namesake of a Kansas City pastor is gone; he is the pastor. Namesake notes are now a checked list, never a guess, and his Wikipedia article counts toward his talk line.

The Meta Ad Library search used an unordered keyword match and one page of results, so for a name like Chris Mann it read every ad in America carrying both words and the Kansas ones fell past the first page; none of the Attorney General race's Meta ads had reached the desk, the Newsmax ad naming Kobach among them. Names are now searched as exact phrases across every delivery status and up to four pages, and the scan log says how many ads each name returned. Talk of the candidates now lists every name in the lead races; one with nothing to read says why.

The Spending card's ran figure now counts what aired: each station flight for the days of it that fell in the window at its per-day rate, an invoice's figure beating the order's, plus FEC filings and digital ranges on their document date. It used to count every order at its whole amount on the day the order was filed, which put a flight booked through November into one week. A line under it says how much of the station money is invoiced and how much is ordered and not yet billed. Meta and Google ads now show their impressions range on the ads cards, and an ad naming a watched candidate with a quarter million or more impressions is a red flag whoever paid for it. The dashboard's race pages carry the same card.

Historical(ish) Facts, Kansas Ed. moved to the foot of the dashboard. The list is now a pool: every fact is shown once, and as the pool runs low the scan asks the machine for a fresh batch in the same voice, checked against everything already shown so no event comes round again; the card says the machine writes them. Talk of the candidates on the dashboard is three columns, name, line and momentum, and no wider than its card; the Intel map's copy keeps the sources and wraps. Clicking a race now shows that race's own spending, what ran and what is booked, by the same methods as the dashboard, with the same windows.

A new measure of attention beyond Google search, on the dashboard and the Intel map: for each candidate in the lead races, daily Wikipedia views of their own article, Bluesky posts naming them counted by day, and YouTube and news search interest across Kansas markets from Google Trends. Each source is scaled to its own fortnight and averaged into one trend line with a this-week-against-last momentum, so it is a shape, never a level. YouTube and news interest by market are layers under Talk of on the map, the dispatch hears when a name's talk moves sharply, and the regular scan refreshes it every few hours.

The Intel map's What things cost card and the county price tables now carry natural gas for Kansas homes (EIA monthly, per thousand cubic feet with the rough price a therm, against the same month a year ago) and propane for Kansas homes (EIA weekly through the heating season; the last reading stands through summer). The whole-basket lines add electricity, piped gas, and water, sewer and trash from the Bureau of Labor Statistics. Two basket lines that had gone quiet, food at home and shelter, are back: the statistics bureau serves some files as raw bytes and the reader was not decoding them. Natural gas and propane are also map layers under Prices.

The Intel map's county detail now lists a county's top seven wastewater pathogens once, with a link to a new Wastewater page that carries every pathogen every plant reports, county by county, highest first, and the KDHE statewide row. The per-pathogen layers no longer repeat the same lines under the pathogens box, and a county row in the layer list shows its top pathogen and a count instead of the whole table. In the news here: the outlets now take turns in each county's headlines, local papers first, so one statewide wire no longer fills the list.

The Intel map's COVID in wastewater layer is now Wastewater pathogens. A county's colour is its highest-ranked pathogen, and the county detail lists every pathogen the county's plants report, highest first: COVID-19, influenza A and B, RSV, norovirus GI and GII, human metapneumovirus, measles, H5 bird flu, mpox and enterovirus D68 where sampled, with the undetected ones named at the foot. Each pathogen also has its own layer. The check is part of the regular scan: the community pass refreshes WastewaterSCAN's five plants every twelve hours and logs what they report, the KDHE monitor workbook for Johnson, Reno and Shawnee counties is pulled daily, and the daily health sweep flags a feed that stops or a KDHE file that goes unrefreshed.

The voice picker moved to the top bar of every Sentinel page: Voice (beta), showing the voice in use with its face, and a menu of the three, Sentinel (the desk's own, the default), Ted, and the Consultant, each with its avatar. There is no off button; Sentinel is the plain voice. In Sentinel nothing floats; in Ted or the Consultant a small floating switch stays in reach. The Apps menu is now Your apps, listing every app the login can open, and it does not appear for a login that has Sentinel alone.

The headline bar opens a Top stories page with every story from the latest read, national and Kansas, and an archive of what led each day. The Kansas lead now favors what is fresh: a political story older than two days loses to today's. And a From the founder card sits under the dispatch, a couple of lines from the newest note, opening a page where the founder writes updates and opinion in their own words; the operator posts and edits there. A national committee's ad that names nobody we watch and no Kansas place is now set aside as another state's race, and the dispatch names the state and the full name whenever a fact is about somewhere else.

Inside Elections' Senate, Governor and House ratings are now read every few hours for every state. The watched Kansas races show their rating, the rating it moved from and the date, on the dashboard beside the markets and on each race page; a move on a watched race becomes an event, high when it moves toward the other side, and the dispatch writer sees the ratings. For the operator and super users, choosing another state in the dropdown shows that state's ratings as an alpha view. Inside Elections' and Split Ticket's analysis feeds join the news sources.

The booked figure had been reading zero because it looked for the wrong field on the invoice, and because it ignored flights already on the air. Every station buy is now folded into one line from its orders, invoices and cancellations: an invoice beats an order, the latest order beats an earlier one, and a cancellation takes the buy off the books. Booked ahead counts each flight for the days of it that fall in the window, including what a running flight has left. The Spending card also says how many flights are on the air today and about how much a day, and how much was booked and then cancelled in the last thirty days. Cancellation filings become their own events, and every station file now gets its contract read, not only the high ones.

Checked every page, every data file, and every script on the site. A script on the client portal page had stopped loading, which could blank out the research modals; it's rebuilt and now loads cleanly. The Sentinel dispatch editor sent visitors to a dead page when there was nothing yet to edit; it now sends them back to the dashboard instead. Everything else came back clean.

Video found on YouTube is now sorted by what it is: a spot from a campaign, committee or the archive is an ad; a clip from a newsroom, podcast or an official's own press channel is not. Ads on the air shows only the first kind. The dashboard gains a News clips box for the second, with the channel and the candidate named.

The war chest table now carries a running itemized count for each statewide candidate from the Secretary of State's contribution viewer, pulled every day: itemized to date, the last thirty days, the latest gift, the share from Kansas. It moves the moment a new report or a last-minute filing is processed, and the dispatch writer sees new money the day it lands.

After signing in, the launcher lists what has changed since your last visit, kept to the parts of the site your login can open, and named for your campaign where the login belongs to one. Got it hides the list until something newer lands.

The war chest table on the Spend page had only federal committees, from FEC reports. It now carries the statewide candidates too, from the Receipts and Expenditures reports they file with the Kansas Public Disclosure Commission: raised, spent and cash on hand through the latest report. Where a report is a typed PDF the figures are read directly; where it is a scanned form, the summary page is read by a model and the row says so. Checked daily, and each report is read once.

A watched sponsor buys in every state, and the Senate Leadership Fund's Maine ads had been kept on the sponsor's name alone. One rule now runs after every classifier: an ad, filing or story that names another state and not Kansas, and names no watched candidate, is not ours; the same goes for anything whose targeting lists states without Kansas. What was already on file gets the same treatment on the next scan.

The top row on every page is now Dashboard, Intel map, The Log (everything the watch has caught), Spend, and Flights (station orders by the dates they run), each with a hover line saying what it holds. The ad library moved under More → Read. Every tested message on the map now opens its own page: the full words, the run it was tested in (marked synthetic or real), how each market moved and where it ranked, the claims it rests on with their tiers, and a few plain bullets on why it works.

An operator can edit any dispatch from the dashboard box or the Dispatches page: the first draft is kept, the edit is signed and timed, the Ted Mode retelling is redone from the edited words, and the clock does not overwrite an edited edition. The What Kansas is searching box carries a footnote when a watched name is shared with someone better known (Chris Mann the singer, Adam Hamilton the pastor), and the dispatch writer sees the same caveat.

The two versions are the War Room (the whole desk) and the Tracker (one race, free), on the badge and the front door. The LEFT SENTINEL masthead now rotates through all seven faces with every login, refresh and scan; the operator can pin one. On the dashboard the small arrow after the seven-day spend figures now sits beside the number it points to. The Kansas top story is now chosen for politics: a political-news search of Kansas outlets comes first, then stories about the things politics is about (jobs, water, schools, courts, prices), and listings like what time a game is on are never it.

The Calendar (under More, and from the dashboard) shows the week: where the candidates will be, what starts airing, the day's dispatches and the dates that matter. Public events naming a candidate arrive from Mobilize's public feed and podcast episodes carrying a candidate's name from Apple's search, every scan; staff add the rest by hand. Click a day for everything about it: who was where, every order and digital ad running that day, what the watch logged, the dispatches, the markets and the asks for money. Ads on the air now covers TV, YouTube, Meta, Google, Snapchat, radio and cable. The Intel map's price card is now What things cost: gas, diesel and electricity as before, and from the Bureau of Labor Statistics eggs, milk, bread, ground beef, chicken and coffee, plus food at home and the whole basket for the nation, the Midwest and Kansas City, each with the year's change. Digital ads on file that were aimed at another state are set aside. The masthead has an Operator page under More with seven faces for the wordmark (Wire Service, Dot Matrix, Teletype, Printout, Punch Card, Cold War, Ticker) and a set of lines to put under it; the choice shows on every page and the front door.

Every sponsor name and side chip follows one rule now: the other side in red, ours in blue, unattributed in a light purple. The Intel map's layer list folds into groups that show what is switched on inside them, with a finder box, and the question box has a Show it on the map button: describe what you want to see (counties where life expectancy is below the state average, opponent money with gas prices) and the page picks the layers and the shading. The Kansas top story leaves sports out unless it is the Chiefs. The McConnell clock says Alive. Maybe.

LEFT SENTINEL now sits across the top of every page and the front door in a dot-matrix wordmark. Beta testers have a Suggestion box (the beta badge, the More menu and the footer all open it); what they write lands on the Sentinel bench in the Lab with their name and the page they were on, where it can be marked considering or done. Video titles from YouTube are decoded, and the public archive the videos come from is never named. Digital ads from a watched sponsor that were aimed at another state are set aside. The dispatch treats a domain we only just noticed as background, never the lead, and may not compute figures of its own.

Polymarket, Kalshi and PredictIt prices on the watched races are read every scan and kept. The dashboard shows the chance our side wins on each race, the week's change, and each market's number; the Markets page has every price with its day and week move and a line of history; the race pages carry the same. A move of six points or more in a day becomes an event, and ten becomes medium. Markets with under a thousand dollars traded are marked thin and left out of the averages. The dispatch writer sees the numbers too. Also: dispatch paragraphs now have air between them on a phone.

Every edition since Thursday morning had been stored empty and marked finished, so the dashboard kept showing the last real one, which was written before the rules changed and ended by giving the reader a job. The cause was the writer spending its whole token budget thinking and returning no words; that reply was filed as a success. All writer calls now ask for plain text, get room to write, and treat an empty reply as a failure that is recorded rather than an edition. The memos and the map question box went through the same fix.

The morning dispatch had been closing with a line like "someone needs to decide today whether to respond", which is against its rules. The cause was a list of memos awaiting decision that was being handed to the writer with the facts; it turned that list into an instruction. The list is gone, the rule is spelled out, and every dispatch is now checked after it is written: if it gives anyone a job, the writer is asked once to redo it, and if it still does, the offending sentences are cut. The box's next-edition line now says around 6:30, 2:30 and 7:30, which is when the text actually lands.

Today's sweep checked every page, every data source, and the site's health report. One problem turned up: the tracker that watches for the senator's next recorded vote was checking in on schedule but never updating its own clock when there was nothing new to report, so it looked stalled even though it was working. It now marks itself checked every time it runs, whether or not there is fresh news. Everything else came back clean.

A phone number and an email address can now be signed up to both sides' lists, and whatever the campaigns send lands in Sentinel: asks for money, surveys, calls for volunteers, event invitations. Each message is placed on a side by its Paid for by line, who signs it, and where its links go, since the sending numbers rotate. The dashboard shows a box with each side's count for the week, how many asked for money, the dollars asked, the lowest ask, and how many leaned on deadline words. The Seeds page under More lists every message, takes a pasted one, and explains the one-time setup: a local Twilio number pointed at the text webhook, and an address on a subdomain pointed at the mail webhook, with the tokens to set on Render.

The query behind the dashboard map put the race filter at the end of its conditions but its values at the front of the list, so the dates were compared against race names and nothing ever matched. It reported no spending in thirty days while the same page showed the spending plainly. The values now line up with the conditions, and the map shades by market again. The same fault would have hit the Intel map for any login limited to particular races.

The search reading covered two names, so there was nothing on Holscher, Masterson, Hamilton or Marshall: it now follows the candidates in the races the desk leads with. Headlines in the county panel link to the stories they came from, which needed the link kept when the feeds are read. The map on the dashboard shaded opponent money only and went blank when none of it had a price on it; it now falls back to every buy on file, says which it is showing, and reports any money it could not place in a market.

A box under the low items shows what is on the books but has not aired: station orders whose flight has not started, with who, where, when and what is on the order. It is the only forward-looking thing on the page. Beside it, a link to the buyers who were spending and then stopped, which is often the lull before the next filing. The line at the top now reads ads and buys, what people are searching, feeling and paying for. And there is a badge in the corner of every page marked Ted Mode: switch it on and the whole interface starts talking like two amiable time travellers, while every figure, date and dollar stays exactly what it was. Switch it off and everything returns. It is remembered per login.

Five Kansas House races join the watch: districts 3, 15, 28, 39 and 67, picked because they are the party's own targets, were decided inside two points in 2024 or are open seats, and because five seats is what ends the supermajority. Each carries its candidates, committees and station terms like any other race, and its page says in a line why it is watched. Every race page now carries the same picker the dashboard has, so there is a way back and a way across: the top three as links, then a dropdown for each level, Federal, Statewide and Statehouse, with Local held for later. The Kansas Senate is not on this year's ballot; its seats are next up in 2028.

The console can now remove a login or a whole client for good. Both ask you to type the name first: deleting a login takes its card and alert settings with it, and deleting a client takes its logins and any voter or donor file uploaded for it, which is what the privacy policy promises. Pausing is still there as the reversible option, and the access log keeps its history either way. The operator login cannot be deleted from the console. The new-client card now says plainly that it sets up that campaign's portal, that the password is optional and generated if left blank, and that Sentinel is the same statewide watch for everyone. The chip at the top of every page reads Permissions: Soccer Mom rather than the level alone.

The watchlist now carries the Secretary of State, Insurance Commissioner and Treasurer contests and all four U.S. House districts, each with its candidates, committees and station terms, so the ad tracking works the same way it does for the top three. The dashboard keeps Governor, Attorney General and Senate up top under a Top races label, with a picker on the right for the rest. The brief is now the dispatch, morning, afternoon and evening, and opens with a one-sentence headline; it may lead with whatever is actually interesting, including search, registration, wastewater, weather and the stories newsrooms will not drop, and it closes with a dry line about the other side rather than telling anyone to run an ad. On the Intel map: satellite now carries roads and place names, county lines are readable and every county is labelled, the wastewater reading is one line per virus with anything climbing marked, the registration table shows the Libertarian and other-party columns so the row adds up, the news topics link to the stories, message fit explains itself in a sentence, the long explainer is replaced by a last-updated line, and a question box answers from the map's own figures. A domain that is registered is no longer described as live: a parked placeholder is called one, and the dispatch is told the difference. The source list reads in plain language.

The front door now leads with a single Sign in. After it, a launcher shows a card for each thing the login can open (Portal, Sentinel, Lab, Console, and the Account page); a login that opens exactly one thing goes straight there. The separate per-app sign-ins still work as shortcuts. Signing out from any app returns to the lksilico.com front door.

On the Account page a login can now wear a portrait instead of uploading a picture: thirty-three figures from Frederick Douglass and Susan B. Anthony through Frances Perkins, Truman, the Kennedys, Johnson, Humphrey, Chisholm, Jordan, Marshall, Milk, Wellstone, Lewis, Ginsburg, Obama and Biden, plus Kansans Sebelius and Slattery. Every portrait is in the public domain or released under CC0, checked by license on Wikimedia Commons, with the source listed on a credits page. A Surprise me button picks one at random.

Every app page now shows the signed-in person at the top right: picture or initials, display name, and permission level, which opens the Account page. Permission levels are named by the operator on the console; the defaults poke fun at what consultants call voters (Situation Room, Obama Coalition, Persuadable, Ticket Splitter, Surge Voter, Soccer Mom, High-Propensity, Focus Group). The Account page carries a card: name, contact details, address and county, birth year, gender, party, the optional demographic fields a pollster weights on, the issues you care about, and three opt-ins (texts, the survey lab, email briefs), each recorded with its time. The console lists every card as the databank, with a full export and a text-list export in the shape the Lab's SMS bench loads. The privacy policy describes the card.

Every Sentinel page now opens with one line: the intelligence desk for Kansas, with Kansas as a dropdown of every state and a national option, followed by what the desk covers: every ad and buy the other side makes, what people here are searching, feeling and paying for, and the messages that fit. The same strip says the scan runs at seven past every hour, when the last one finished, and that briefs come at 6am, 2pm and 7pm Central. Kansas is the only live edition; picking another state records the interest and says so. Each login can star editions and set a default for when more exist.

Two new rows of layers. Schools: every Kansas public district rolled up to its county, with enrollment and its change since 2022, students per teacher, reading and math proficiency, four-year graduation rate and spending per pupil; the county panel lists the districts with each figure. Health: overdose deaths, suicides, premature death, life expectancy, the uninsured, primary-care and mental-health providers, children in poverty, teen births, household income, food insecurity, broadband, mental distress and excessive drinking from County Health Rankings, plus births, deaths, births minus deaths, the births trend since 2021 and net movers from the Census Bureau's county estimates. Counties with too few cases show no figure rather than a misleading one. Both refresh weekly. The state's own vital-statistics portal has no feed, so these are the federal compilations of the same data.

KDHE publishes its Kansas Wastewater Monitor only through a dashboard, so once a week, after KDHE's Wednesday update, a small job downloads its workbook and keeps the latest reading per virus for Johnson, Reno and Shawnee counties: COVID, influenza A and B, RSV and norovirus, with KDHE's own Low / Moderate / High level, trend and precaution. Together with WastewaterSCAN's five plants, six Kansas counties now carry wastewater readings on the map.

CDC's national wastewater system lost Kansas a year ago, so the map's wastewater reading was stale. It now comes from WastewaterSCAN, which samples five Kansas plants (three in Kansas City, Kansas, plus Lawrence and Salina) two or three times a week for COVID, influenza A and B, RSV, norovirus, human metapneumovirus, measles, H5 bird flu, mpox and enterovirus D68. Each reading is the three-week average against the prior three weeks and where it sits in the plant's own last twelve months; anything not detected reads as zero. Layers on the Intel map for COVID, flu A, RSV, norovirus and measles; the county panel lists every virus for Wyandotte, Douglas and Saline. A scan killed by a host restart no longer blocks the next one.

The dashboard opens with the national top story, marked BREAKING when it is under ninety minutes old, and the top Kansas story from Kansas outlets beneath it, refreshed every scan and whenever the page finds it more than twenty minutes old. The Kansas fact and the brief now sit in the left column above Needs eyes at that column's width, and the right-hand column moves up to fill the space. On the Intel map, the weather-alert table shows one county and a button to open the rest instead of listing every county under a statewide alert.

The brief now comes three times a day, Central time: morning after 6, afternoon after 2, evening after 7, each covering what landed since the previous edition. If nothing landed, the brief says so in one line. The dashboard box names the current edition with its date and time and the time of the next one, and opens in place to read; it no longer prints the whole brief or offers a regenerate button. The briefs page groups by day with every edition, the last five days in full and older days in the archive. The Kansas fact has its own box under the numbers; the memos box is titled Memos; the footer lost the fact and the long confidentiality sentence.

The dashboard no longer prints the whole brief; it shows a box saying the morning brief is ready, with the date, and a link. The briefs page carries the last five days in full and every older brief in an archive that opens by date. The brief had been keyed to the clock in London rather than Kansas, so on some days it was written the evening before and then treated as already done; it is now one per Central-time date, written on the first scan after 6am, and the emailed digest uses the same copy.

The traffic layer now names roads (US-36, K-9, I-135) and places each segment by the nearest town instead of printing route codes and coordinates. The plain / roads / satellite switch sits in the map header where it can be found. Messages left the layer picker and have a standing table below the map. The hourly fact gets a hanging indent so a second line reads cleanly. The McConnell card no longer fails when the vital-status check has not run yet, which was the sign-in error some saw.

The host reported the service over its memory limit. The scan (feeds, PDFs, the weekly data pulls, the model) used to run inside one of the two web processes and could push the whole service past the limit. It now runs in its own short-lived process that hands its memory back when it finishes; single-source refreshes and memo drafting do the same. The site serves from one web process with six threads instead of two processes, roughly halving the standing footprint.

Two public pages, linked from every footer and the sign-in page: the privacy policy (what is collected, what it is used for, what is never done with it, how long it is kept, Kansas voter-data limits, breach notice under K.S.A. 50-7a01, how to reach us) and How we protect data (one service behind one login, per-workspace isolation, the brake, breach signals with automatic password rotation, no trackers, no voter or donor data to any AI model, daily checks). The retention the policy promises is now enforced: access and activity records are purged after thirteen months, security events after twenty-four.

This morning's check found the search-interest-by-market layer stuck on old data: once it had tried once, it never tried again on its own, so a single failed refresh left it stale for good. It now refreshes on the same regular schedule as the other signal layers, alongside a community-pulse fix already made overnight. Every page was checked as both the operator and a regular login, every data file was checked for valid formatting, and the app started clean end to end.