Version 1.2.1 - nine languages, and a quieter start. NINE LANGUAGES INSTEAD OF TWO - Spanish, French, Italian, Japanese, Korean, Portuguese and Chinese join English and German. The whole app, not a menu bar over English screens. - Japanese, Korean and Chinese survive the PDF export: the fonts are embedded, so the proposal reads on paper the way it reads on screen. - Numbers, dates and quotation marks follow the language, not the machine. THE START SEQUENCE - The app opens with the animation from the website, inside the window. - It plays once per version. It does not greet you again every morning. LANGUAGE, CURRENCY AND APPEARANCE ARE THREE DIFFERENT THINGS - Appearance follows the machine: light or dark, as the system says. - Language and currency belong to the document and travel with it. - The currency symbol follows the tab in front through every dialog. A file open in a background tab no longer relabels the money in the one you are looking at. THE TOUR AND THE OPTIONS GRID - Seven tour cards became five, and the tour no longer walks on by itself. - The grid stops cutting its verdicts short, and no longer recommends a model nobody has filled in. WHERE A NUMBER CAME FROM - Every box says which numbers you set and which ones the app set for you. - A save that fails now says so, instead of failing quietly. - Settings are reachable from the keyboard.

Version 1.2 — the system around the process. THE ARCHITECTURE VIEW - A fifth view: which systems run, in which version, which tables and views they hold, and where every column comes from. - Pick a node and read what a change to it reaches, what it provably leaves alone, in which order the work has to be built, and what that is in person-days — with the assumptions printed next to the figure. - Context, Systems, Data, Lineage: four distances through one model, not four drawings that drift apart. - Problems it finds on its own: an arrow that lands nowhere, a derivation whose source is gone, a key that changes type on the way, personal data reaching the outside, a system past its end of support. - Needs: what a KPI, report or export already has, what carries a caveat, and which column is missing. READ A FOLDER INSTEAD OF RETYPING IT - SQL schemas and compose files are parsed exactly. Prose goes to your own AI endpoint under a citation rule: what cannot be traced back to a passage is not offered at all. - Two ways through the same lines. A hundred columns out of a schema file are a list, with a filter, "only changes" and select all. A dozen readings out of a handover note are a walk: one line at a time, with its quote, the file it came from, and what stands in the model today. - Take it, Adjust or Discard. A value you correct takes the same path in as one that was read — there is no second, untested route. - A number outside any plausible range is dropped, never bent into shape. A step that already exists produces one line per value, not a second copy. - Everything applies in one step, and one Undo takes the whole walk back. CARRY IT OUT OF THE APP - Change memo (PDF): one decision, what it touches, what it provably leaves alone, the effort. - Documentation as an editable Word file. - The architecture drawing as draw.io or SVG, still editable. NUMBERS THAT SAY WHICH DAY THEY MEAN - Every load figure now names its basis: the average day, or the busy day that nine in ten working days stay under. - The comparison flags its own contradictions — money saved while cycle time or throughput moves the wrong way, a range widened at one end only. - Presentation mode for the client meeting. Esc leaves it. THE PROPOSAL PDF - IBM Plex with tabular figures, a Before/After table that reconciles with the headline number, the measure on the cover, the assumptions register inside the offer. Options that are not this offer are called "further levers", not results. - Process documentation without placeholders, with busy-day and average-day load per step. SIGNAL COLOURS MEAN ONE THING - Teal, amber and red mean utilisation and nothing else. Selection, structure, delete, licence and hints have their own colour language, and a letterhead accent colour that comes too close to a signal colour is flagged. ALSO - A draw.io drawing the client already has can be read straight in — as a proposal like any other, every line naming the cell it came from. - All demos and templates recalibrated: exactly one bottleneck per process on the busy day. WHAT IS FREE CHANGED - Saving a file, the PDFs and data export are free now; the free PDF carries a trial mark and the default letterhead. - Pro removes the mark, puts your logo, colour and letterhead on every page, and adds the Word file, the editable drawing and project folders for a team.

Version 1.1 — numbers you can defend, and a first run that proves something. MORE ACCURATE NUMBERS - Parallel branches now deliver at the pace of the slowest branch, not the fastest: 10 a day claimed, 8 a day real. - Shared resource pools are counted once instead of several times. Utilisation per role or system is measured against the pooled capacity — in one test case "Sales 150%" became 30%. - Past the load limit, queues stay finite: waiting time is capped at a 30-day horizon instead of running to infinity. One model went from 422 working days of lead time to 32.8. - The bottleneck ranking splits the load across the steps that share a pool, and names the pool. - Average handling time is the mean, no longer the median shown as one. The length of the working day comes from your model, not from a constant. - The report no longer calls a step at 157% utilisation comfortable. RESULTS THAT SAY WHAT THEY REST ON - Capacity, handling time or spread you never entered are named instead of quietly filled with defaults. - Three states instead of one: blocked (the number would be invented), provisional (it computes, and tells you on what), reliable. - Every figure in the comparison opens into its parts: running cost, throughput value, ongoing cost, investment. Comparison and options read from the same money source. A FIRST RUN THAT PROVES SOMETHING - The walkthrough now applies a real fix: one click cuts handling time at the bottleneck, the saving appears in money, and the bottleneck visibly moves on. - All bundled examples recalibrated: 85-95% load on an average day, 110-140% on a heavy one. - New template: Order Handling. LAYOUT AND PROCESS LANDSCAPE - Separate flows get their own bands side by side instead of one vertical tower — inside expanded sub-processes too. - Process landscape: connect areas by drag & drop, with a share editor at the drop point and one step of undo. Circular, self- and duplicate connections are refused before you let go, with a reason. - The "loss path" switch is explained where you set it. LOOK AND WORDING - One design system: verified contrast (WCAG AA), consistent controls with hover, press and focus states, keyboard operation, respects reduced motion, light and dark checked separately. - Columns of figures use tabular digits, so they no longer jump on refresh. - One term per thing across app, guide and reports, with a glossary in the guide. - Contact as selectable text in the guide and in settings: jonas.hoettler@balane.tech

Version 1.0.3 — sharper numbers, readable diagrams, better client documents. WHAT YOU HAND THE CLIENT - The options comparison is now in the proposal PDF: every variant you tested, ranked, with the ones that are not worth doing marked as such. - Payback is calculated from your own rates instead of a fixed assumption. - Documentation PDFs carry the date you exported them. - The cover page no longer claims a change resolves the bottleneck when the analysis says it does not. - English exports are English throughout — headings, file names and footers. MORE ACCURATE RESULTS - Parallel branches that merge again keep their full volume. Steps after the merge are no longer under-counted and no longer disappear from the bottleneck ranking. - One overloaded side branch no longer drags down the throughput of the whole process, so the per-year figures stay realistic. - A loop with no way out is now reported as a blocking error instead of a quiet 10,000% load. DIAGRAMS YOU CAN READ - Arrowheads on every connection — direction is visible in a screenshot, not only in the animation. - Split percentages are printed on the connections: a 70/30 split looks like 70/30. - Rework loops route around the steps in between instead of cutting straight through them. - Clicks respond immediately, and hovering shows what you are about to hit. FEWER SURPRISES - Deleting a sub-process removes everything inside it — no more ghost areas in the navigator or the landscape view. - Quitting with unsaved work now offers Save, not only discard. - Undo refreshes the inspector, so old values cannot be written back. - Your last project reopens on launch, and recent files are one click away. Thanks for using FlowVisual.

Version 1.0.2 sharpens the numbers and makes saving safer. The calculation engine is now consistent and correct end to end — figures may differ from 1.0.1; the new ones are right: - Utilization matches everywhere: Canvas, handbook and proposal show the same peak-day / P90 value - Levers on sub-processes now take effect - No saving is shown without a documented change behind it - Media breaks at merge points are no longer counted twice - Lead times near the capacity limit no longer jump Plus: crash-safe saving, and named versions are now stored inside the document. Thanks for using FlowVisual.

Version 1.0 — the first release of FlowVisual. Model a business process, stress-test it with a Monte Carlo simulation, find the real bottleneck, and prove the fix in euros with a before/after business case. - Monte Carlo stress test with full distributions — the typical day, the bad day, and the breaking point - Bottleneck ranking, fragility and volatility analysis - Before/after euro business case with a confidence band - Branded proposal and documentation PDFs - Structured data export (JSON and CSV) Thanks for trying FlowVisual.