A calculator bot can collect every field and send an attractive result while still doing the wrong calculation. When a spreadsheet and a Telegram bot disagree, a successful conversation is not enough to accept the delivery.
The useful starting point is an approved calculation specification: valid inputs, units, evaluation order, output meaning and conditions that must stop calculation. Give that specification to an AI builder or developer before judging the generated interface.
A real request: reproduce the calculation, not guess the formula
The historical Freelancer project Custom Telegram Calculator Bot requested seven inputs, calculations matching a supplied sheet, a result message and a subsequent PDF, email and booking workflow. On November 12, 2026, its page was marked Completed and showed a relative posting age of ten months; the exact publication date was not confirmed. This is historical demand evidence, not an invitation to bid.
The buyer placed the actual calculation rules in attachments that we have not verified. We cannot reconstruct their metric or formula. The practical question we can answer is narrower: how do you prove that a bot follows a calculation already approved by the business? One project does not establish the size of a market.
The time-saving example below is an original hypothetical specification. It does not describe that customer's formula, implementation or delivered results.
Write the calculation specification before the conversation
Ask the person responsible for the business rule to provide a sanitized specification rather than only a spreadsheet screenshot:
- Fields and units: Is duration measured in minutes or hours? Is volume daily or monthly? Does a percentage arrive as 20 or 0.2?
- The complete formula: Include parentheses, intermediate steps, constants and their meanings. Translate cell references into named business fields.
- Valid conditions: State which inputs may be zero, which must be positive and which combinations have no meaning. Decide whether blank means unknown.
- Output rules: Define the period, units, decimal places, rounding method and whether rounding happens during calculation or only for display.
- Approved examples: Have the business owner supply inputs and independently confirmed expected outputs. Do not let one generation step invent both the implementation and its supposed correct answers.
Microsoft's Excel calculation operators and precedence documentation explains how evaluation order and parentheses affect results, including the order for operators with equal precedence. Matching field names does not establish that the migrated expression is equivalent.
List hidden-sheet references, external data and manually maintained coefficients separately. Without those dependencies and approved examples, you can build an interface prototype, but cannot claim to have reproduced the workbook accurately.
A worked example with an explicit scope
Suppose a team wants to estimate monthly hours saved by reducing the time spent on a repetitive task. This hypothetical model assumes constant daily volume and handling time. It covers task-processing time only, excludes training, deployment and maintenance costs, and does not promise actual savings.
The following limits are example business rules, not platform limits:
- N: tasks per day, an integer from 0 to 10000.
- B: previous minutes per task, a number from 0 to 120.
- A: new minutes per task, satisfying 0 ≤ A ≤ B.
- D: working days in the month, an integer from 1 to 31.
- H: estimated hours saved, calculated as
H = N * (B - A) * D / 60.
Subtract the handling times first, multiply by daily volume and working days, then convert minutes to hours. With N=20, B=6, A=3 and D=20, the answer is 20 hours. Accidentally writing N * B - A * D / 60 produces 119 for those same inputs. That is a different rule, not a formatting discrepancy.
For this example, retain the unformatted calculation value and round half up to two decimal places only for the displayed result. Show its unit and rule version. More demanding precision requirements need a separately approved numeric representation and error tolerance; a neat decimal display does not prove accurate internal arithmetic.
Accept expected results, not just one successful interaction
Put input sets and expected outputs or rejection reasons in the requirements. Compare each with the bot's actual response:
- Normal case: N=20, B=6, A=3, D=20 must display 20.00 hours, all four inputs and the rule version.
- Zero volume: N=0 with the other normal values must display 0.00 hours. A valid zero must not be treated as missing.
- No time difference: A=B=6, N=20, D=20 must display 0.00 hours.
- Outside the model: A=7 and B=6 must be rejected under this specification with an explanation. Do not silently clamp the result to zero. A model that reports increased time needs a separately approved negative-result policy.
- Incomplete or invalid input: A blank D, 0, 32 or 2.5 must request a corrected working-day count, not an invented default. Non-numeric text for N must receive a clear correction message.
- Correction and recalculation: Change D in the normal case to 10. The result must become 10.00 hours, and the displayed input summary must change too.
For other formulas with a variable denominator, add a division-by-zero rejection case. Missing data must not silently become zero, and a failed calculation must not be presented as a successful result.
These are acceptance criteria for the hypothetical design, not test results for an existing bot. Real projects also need approved workbook comparisons, boundary values and combinations that have caused errors before.
A fixed business formula is a smaller scope than arbitrary expressions
A business calculator often needs only a few numeric fields and a formula reviewed by its owner. Users provide data; they do not change the calculation rule through an input field.
Do not ask a generator to execute incoming text with eval(). The MDN eval() reference explains that it parses and executes a string as JavaScript code. Explicit numeric conversion and validation followed by a fixed, approved calculation are easier to review for this use case.
If users really need to type arbitrary expressions, separately define allowed symbols, parentheses, length limits, parsing and error handling. Verify that the runtime supports the proposed implementation. A business form does not automatically imply support for arbitrary code, spreadsheet functions or third-party calculation libraries.
A requirements prompt you can adapt
This is a prompt for the hypothetical example. Replace it with your own approved formula and expected outputs before using it with a builder or developer:
Create a Telegram private-chat hours-saved calculator. Collect N, B, A and D, explaining each field, unit and allowed range. Use the fixed formula H=N*(B-A)*D/60 and rule version hours-v1. Accept validated numbers only; never execute user-provided code or expressions. Stop on missing, non-numeric, out-of-range values or A>B, identifying what needs correction. A valid result must include the input summary, calculation rule, hours saved and rule version. Round half up to two decimal places only for the final display. Recalculate using corrected inputs. Pass all six groups of acceptance examples in this article. The first version displays results in Telegram only, with no PDF, email, booking, external spreadsheet synchronization or runtime AI decisions.
Saving previous results adds decisions about which inputs, formula versions and outputs to retain, and how retention and deletion work. For cancellation, timeouts and resuming a form, use the multi-step conversation state guide alongside the calculation specification.
Where BotFatherV2 fits, and what still needs verification
As checked on November 12, 2026, BotFatherV2 publicly supports creating and modifying Telegram bots through AI conversation, and its homepage includes a calculator example. That establishes a relevant starting point for AI-assisted development. It does not establish support or successful acceptance for every business formula, Excel workbook import or the buyer's entire workflow.
Its public AI development documentation says the first version does not add bot runtime capabilities that depend on paid AI. The fixed SDK validation scope does not support arbitrary npm modules or Node/Bun runtime APIs. Start by generating and testing a fixed-formula flow against approved examples. AI-assisted bot development does not require a model to guess numerical answers at runtime.
Treat PDF generation, email delivery and booking as separate scopes requiring verified implementation options, external services, costs and data handling. Importing an existing Telegram Serverless bot also does not mean importing an Excel workbook. BotFatherV2 is an independent product with no affiliation to Telegram's official BotFather.
The next step is to prepare a one-page formula specification and approved results, then describe the smallest calculation flow in BotFatherV2. Rerun the examples before release and assign a new rule version when the formula changes. If historical records are stored, consult the code rollback and data compatibility guide; restoring old code does not establish that old calculation records have been restored.
Source dates
The Microsoft and MDN documentation linked above was checked on November 12, 2026; a fixed publication date was not confirmed from the readable pages. The buyer's project evidences a requested task, the technical references support the proposed calculation approach, and the product pages establish capability boundaries. These serve different purposes.