Recent Bug Fixes

• When a Stripe Capital payout ran into a problem, it could leave the payout record stuck or show a deposit amount in QuickBooks higher than what was actually deposited. Skimmer now handles these failures cleanly, so payout records and QuickBooks deposit amounts stay accurate.
• When Skimmer had trouble loading a user's account details, an owner opening the Capital page could see a "Page not found" message instead of the page itself. Skimmer now handles this correctly, so owners see the Capital page as expected.
• Navigating straight to the Capital page, instead of clicking into it from elsewhere in Skimmer, could result in a blank page. Skimmer now loads the Capital page correctly no matter how it's reached.
• If the financing information on the Capital page had trouble loading, the whole page could come up empty. Skimmer now displays the rest of the page even if financing details have trouble loading.

• When totaling chemical usage, AI Insights could add up quantities measured in different units as if they were the same, producing a meaningless total. Skimmer now keeps units of measure separate when totaling, so chemical usage totals are meaningful.
• Asked how many different chemicals were used, AI Insights could report only the first page shown on screen as if that were the complete list. Skimmer now reports the true count of chemicals used, not just the first page shown.
• Asking which pools would be serviced today could return zero results even when pools were scheduled, because routes that hadn't started yet weren't counted. Skimmer now includes scheduled-but-not-yet-started routes in this count, so the answer matches what's actually on the schedule.
• Questions comparing past service to what's scheduled next could compare completed visits against a projected schedule rather than like-for-like data, producing a misleading comparison. Skimmer now compares consistent data on both sides, so these comparisons are accurate.
• Asked for profit and labor cost together, AI Insights reported a combined labor-plus-chemical figure as if it were labor cost alone. Skimmer now separates labor cost from chemical cost, so each figure reflects only what it's supposed to.
• At a service location with more than one pool, AI Insights could count and cost the same chemical dose multiple times — once for each pool at that address. Skimmer now matches each dose to its own pool, so dose counts and costs are accurate at multi-pool properties.
• When asked a reading threshold question, AI Insights could ignore which reading type was actually asked about and count every reading type over that threshold instead. Skimmer now keeps the specific reading type in mind, so threshold questions return results for what was actually asked.
• When a question combined two reading conditions, AI Insights could silently exclude pools where the two readings were logged by different technicians, undercounting the true result. Skimmer now correctly includes these pools, so combined-reading questions return the full, accurate count.
• Certain reading questions were processed in a way that made them take roughly three times longer than expected, close to 70 seconds in some cases. Skimmer now processes these questions more efficiently, bringing response time back to normal.
• Asked how many pools matched a question, AI Insights could report the count of just the first page of results shown on screen instead of the true total. Skimmer now reports the true total count, not just what's visible on the first page.
• Asked about a monthly reading trend, AI Insights could apply a "latest reading only" filter across the whole time range instead of each month individually, making a declining trend appear to be rising. Skimmer now calculates each month's trend correctly, so the direction shown matches what's actually happening.
• When a question returned no results, AI Insights gave the same generic "no data" message regardless of the reason. Skimmer now explains what it searched for when a question comes back empty.
• The Revenue figure in AI Insights profit answers reflected service revenue only, leaving out billable chemical charges, so it didn't match the Profit figure or the Sigma Profit report. Skimmer now includes chemical revenue in this figure, so it reconciles with Profit and the Sigma Profit report.
• Asking AI Insights to compare this year to last year could return $0 for every prior-year value, making a real comparison impossible. Skimmer now pulls both years' data correctly, so year-over-year comparisons return real numbers.
• Asking AI Insights about a specific technician's profit required typing their exact full name; a partial name or how someone would naturally refer to them returned a false "no data" result. Skimmer now matches technician names more flexibly, so a natural way of referring to someone returns their results.

• Asking AI Insights about a customer name with an apostrophe, like "O'Brien," returned wrong or empty results because the search stopped at the apostrophe. AI Insights now reads the full name correctly, so these searches return accurate results.
• Weekly average comparisons in AI Insights could come out slightly off when the date range didn't line up with a Sunday, and asking the same question twice sometimes produced different answers. AI Insights now calculates weekly averages consistently regardless of how the date range falls, so the same question gets an accurate answer every time.
• When asked about scheduled service stops, AI Insights pulled from actual completed stops for a single date but from a different, projected data source for date ranges, so answers could contradict each other depending on how the question was phrased. AI Insights now uses one consistent data source for scheduled-stops questions, whether the question covers a single day or a range.
• When asked about something like "the last Monday of each month," the supporting detail behind AI Insights' answer counted every stop for the whole month instead of just the stops on that specific weekday, so the numbers didn't match the answer. AI Insights now filters correctly down to the target weekday, so the detail lines up with the reported total.
• When comparing "this quarter" numbers, AI Insights sometimes showed last quarter's date range alongside this quarter's figures, making the answer confusing or misleading. AI Insights now labels each quarter's figures with the correct date range.
• When asked how many stops a tech would still service today, AI Insights counted stops that had already been skipped, overstating the number. AI Insights now excludes skipped stops from that count.
• When a question included a specific number, like a dollar amount or count, AI Insights sometimes only picked up the first digit, leading to wrong answers. AI Insights now reads the full number correctly, so answers reflect the actual value asked about.
• Certain last-month questions could get labeled as "month to date" instead of "last month," creating confusion about what period the answer covered. AI Insights now labels the time period correctly for these questions.
• Asking AI Insights the same scheduled-stops question more than once could return different, sometimes contradictory answers. AI Insights now generates the same, correct answer every time for this type of question.
• When asked about labor cost in profit reports, AI Insights included chemical and dosage costs in the total but labeled the whole amount as labor. AI Insights now reports labor cost on its own, separate from chemical and dosage costs.

• Asking AI Insights to compare the next four weeks against the last four weeks returned a per-week average that was too low. Skimmer now calculates and displays the correct per-week average for that comparison.
• When AI Insights searched for a specific product or text term, it built the search in a way that didn't follow Skimmer's standard method for handling user-provided text. Skimmer now uses that standard method for these searches.
• AI Insights sometimes returned incorrect profit figures when a question spanned a date range that had to be split into multiple periods. Skimmer now calculates profit correctly across split date periods every time.
• When AI Insights answered a profit question, the header sometimes displayed the wrong date range for the answer. Skimmer now shows the correct date range on profit answers.
• AI Insights occasionally showed a technician's route assignment start date instead of the actual service date when listing daily route stops. Skimmer now consistently shows the correct service date for each route stop.
• AI Insights displayed route-stop times in UTC instead of local time, putting them five hours off from Service History. Skimmer now displays route-stop times in local time in AI Insights Reports, matching Service History.
• AI Insights treated scheduled, completed, and skipped service stops inconsistently, which could produce misleading counts. Skimmer now applies consistent logic across scheduled, completed, and skipped stops.
• Asking AI Insights for a sales total could return very different numbers depending on whether it counted invoiced amounts or actual payments collected, sometimes even showing opposite year-over-year trends for the same question. Skimmer now applies one consistent, clearly defined sales total so results are reliable no matter how the question is phrased.

Note: These issues only affected pros signed up for the Embedded Bookkeeping Beta.

• When Layer calculated sales tax on an invoice, rounding each line item separately could push the total a few cents above the tax total shown on the Skimmer invoice. Skimmer now calculates Layer's sales tax so it matches the invoice's tax total exactly.
• When enrolling a batch of accounts in Layer, a single failed item could stop the whole batch without reporting the failure or retrying it. Skimmer now isolates failures so one problem account no longer stops the rest of the batch from enrolling.
• A customer's payment was recorded, but the matching account credit was never applied, leaving the account balance incorrect. Skimmer now applies the account credit correctly whenever a payment is recorded.
• A data export crashed whenever a customer activity log entry was missing an update timestamp. Skimmer now handles missing timestamps gracefully so the export completes successfully.

• AI Insights built its underlying queries with incorrect start and end date values. Skimmer now generates the correct dates.
• Start and complete times for route stops in AI Insights tables displayed in UTC rather than the user's local timezone. Skimmer now shows these times in local time.
• When a pro asked AI Insights a question about a specific day of the week, like "how many stops did I do on the last Monday of each month," the answer reflected the wrong day. Skimmer now rejects an invalid weekday calculation instead of quietly correcting it to the wrong day, so "last weekday" questions return the day that was asked about.
• Because service dates were stored with a time component, questions about a specific day (like "last Monday") returned zero matching results. Skimmer now matches on the date correctly, regardless of the stored time.
• Questions about installed items pulled results from a customer tag with a similar name instead of actual installed-item records, returning payment data instead of item data. Skimmer now pulls from the correct installed-items data.
• Filters like "stops after 3pm" compared against UTC time while the results displayed in local time, so the wrong stops came back. Skimmer now compares filters in local time so the results match what's shown.
• AI Insights displayed a placeholder date of January 1, 2010 as the completion time for service stops that hadn't finished yet. Skimmer now leaves the completion time blank for stops that aren't done.
• Questions about time spent on work orders pulled from an empty data source and always returned zero. Skimmer now pulls from the correct source and returns accurate durations.

• Some customers' invoices appeared correctly in the Pool Service tab but not in the Invoices Preview section. We corrected the display logic so invoices now appear consistently in both locations.

• The AI Insights column detector missed several money columns and incorrectly flagged some count columns as currency fields. We updated the detection algorithm to recognize additional money column patterns and reduce false positives.

• When asking for data about a specific date, AI Insights sometimes ignored that date and returned results for a broader default range instead. AI Insights now respects the date specified when the data table is built.
• In the expanded view of the AI Insights panel, the data table would jitter while text was typed in the prompt field. The layout no longer jitters when you type in the prompt field.
• When an AI Insights query returned no matching data, an empty table was rendered instead of a clear no results message. The view now displays a no results message instead of an empty table.

• When you searched by customer name in the Quotes or Invoices pages and then deleted the text from the search field, the filter was not cleared and the results list stayed filtered instead of resetting. These lists now refresh when the search field is emptied.

• When importing products from a spreadsheet, any row with a markup percentage above 999.99 or an oversized cost or price value caused the entire import to fail with only a generic "Failed to import" message. The importer now validates those fields before saving and returns a clear message explaining which value is out of range.
• When importing customers from a spreadsheet that included a Customer Status column, that column was ignored. Every imported customer was set to Lead, regardless of the value specified in the spreadsheet. The import now reads and applies the status value from the spreadsheet correctly.

• The selection fields for reminder days and suspension days in Invoicing Settings didn't respond to taps on Safari for iPhone and iPad, making it impossible to update those settings on mobile. The fields are now fully usable on iOS Safari.

• Admin users could not approve quotes on behalf of customers, even when payment was not required at the time of approval. Admins can now approve quotes without a credit card being required. 

• When a work order was linked to a customer registered as a business rather than an individual, the customer name displayed as "null null" instead of the company name. Company names now appear correctly on work orders.
• When work orders on a job were moved to Scheduled, the parent job's status did not update and stayed stuck at "Not Started." The job status now recalculates correctly when work orders are scheduled.