SOLVENT: a concept cash-flow cockpit and what-if simulator for a small business

VITON¹³ builds concept products to show how it approaches software for a business, not only websites. SOLVENT is the finance tool in that series: a cash-flow cockpit and scenario simulator for owners of small companies, shown on Kite & Co., an invented design-and-print studio with 14 people, ten named clients, fourteen software subscriptions and a lease. The company, its clients, its suppliers and every figure are made up, and the product says so on every screen, along with the fact that it is not financial advice.

The brief we set ourselves had two parts. Answer the three questions an owner asks before payroll (how much cash is there, how long will it last, what happens if we hire, raise prices or lose the biggest client) on one screen, with the definition of every number on the page. And make decisions cheap to try: change something, watch the next 18 months move at once, and see exactly how much of the difference each change is responsible for.

The cockpit opens on cash today ($148,210, the closing balance of the operating account), runway (14.8 months: the time until the projected end-of-day balance first drops below zero) and this month’s net flow. One chart carries 12 months of actual balance and 18 months of forecast: the daily balance, a month-end curve, an 80 % band taken from the model’s own past error, the cash buffer, the zero line, hatching where cash sits under the buffer, the lowest point and the day the money would run out. Below it are the next 30 days as a strip of money in and out, the largest payments either way, alerts computed from the forecast, the share of revenue each client brings and a short glossary.

The what-if simulator has eight levers (hire people, raise prices, lose the biggest client, win a new retainer, move to a bigger office, customers pay later, cut software spend or cancel subscriptions, take a loan), each with its own settings and start month, and nine presets. Every change recomputes the forecast at once: the four key figures roll to their new values, the curve morphs while the baseline stays as a dashed ghost, and a waterfall shows what each lever adds to or takes from cash at the end of the horizon. Scenarios can be saved, renamed and reloaded, the buffer is editable, and Compare lays any two scenarios over each other with a table of differences and a one-line verdict. On phones the levers live in a bottom sheet.

Around the simulator sits the rest of a finance tool. Invoices shows receivables in ageing buckets, days to get paid and filters; Chase drafts a reminder from templates, the chosen tone and the client’s own payment habits, and Mark paid moves cash, runway and the 30-day low everywhere, with undo. Spending has a treemap, 12-month trends, anomalies such as software spending up 38 % after three new subscriptions, 90 days of transactions sorted by rules with a confidence score, and subscriptions with a “What if we cancel?” switch that feeds the scenario. P&L gives the monthly statement and a year view. Ask Solvent (/ or ⌘K) answers 15 kinds of question in English and Russian from the same engine, with a one-click “Apply this scenario”. Export writes the scenario as a CSV file and a one-page PDF.

Kite & Co.’s books are generated deterministically: each calendar month has its own seed, so a month’s ledger never changes when “today” moves, and the balance is anchored at $148,210 on 28 September 2026. Two years of history hold invoices per client with each client’s usual lateness, payroll on the 10th and 25th, rent, subscriptions, freelancers and print costs that follow the volume of work, quarterly tax, seasonality and growth. The forecast is built from the same drivers without the noise: open invoices arrive on the client’s usual payment date, and each lever edits dated events. Levers are attributed one after another, which is why the waterfall adds up exactly to the scenario minus the baseline. The project’s own test suite checks that sum, the runway definition, the balance roll-forwards, every lever’s identity, determinism and the assistant’s answers: 1,927 checks in all.

The app is plain JavaScript modules bundled with esbuild into a static folder with relative URLs; on the portfolio it lives at /demos/solvent/. The charts are our own SVG, the CSV export is our own (UTF-8 with a byte-order mark, and semicolons with decimal commas in Russian, so a spreadsheet opens it cleanly), and the PDF is drawn on a canvas and wrapped by a minimal PDF 1.4 writer we wrote: there is no chart, UI or PDF library. The design is a financial newspaper on warm paper, with Noto Serif Display for headlines, Inter Tight for text and JetBrains Mono for figures, all self-hosted, and a night edition. The interface is in English and Russian, shows dollars or roubles at a fixed demo rate of 90 to the dollar, opens with an intro of up to three seconds that can be skipped, and calms its motion for visitors who ask for reduced motion.

What exists now is a working concept at tarasovvitalii.com/demos/solvent/ with six sections (cockpit, what if, compare, invoices, spending and P&L) plus Ask Solvent and the exports, built on 28 September 2026. That day its test suite passed all 1,927 checks. In headless Chrome we opened every section in English at 1440 px and in Russian on a 390 px phone, applied a preset, marked an invoice paid (cash in the header moved from $148,210 to $151,500), asked a question and downloaded the CSV and the PDF, under the portfolio’s production content security policy: no console error, no policy violation, no sideways scroll and no request to any other server.

All the JavaScript is 11 files and 220 KB, or 72 KB gzipped; the whole folder with its fonts is 0.65 MB. On the live address, Lighthouse 13.5 (five runs per device, medians, on a MacBook with a load average of 3–9) gave performance 81 on mobile and 98 on desktop, with accessibility and best practices at 100 on both, and the page transfers 0.47 MB on load. Before measuring we fixed three things in the app: a footer that jumped on load, a chart and a button without an accessible name, and toggles under 24 px on phones.

SOLVENT is a concept VITON¹³ built to show range, not a commissioned product. Kite & Co., its clients, suppliers, staff and every figure are fictional and generated in the browser: there is no client, no users, no bank connection, no traffic and no revenue, and nothing in the demo is for sale. It is not financial advice: the forecast is a simple model that assumes flat demand beyond seasonality and no clients lost when prices rise, and its 80 % band describes the model’s own error on invented data.

Ask Solvent and the reminder drafts are demo AI: rules and statistics in the browser, not a language model, and no reminder is ever sent. The 1,927 checks prove that the maths agrees with its own definitions, not that the forecast would be right for a real business. Every figure is a lab measurement in headless Chrome on one machine on one day, and Lighthouse ran on the live address with a simulated phone, not on real devices. A production version would still need bank feeds, an accounting integration, accounts and roles, encrypted storage and a review by finance professionals.

How to check a web studio’s portfolio →