AccountingSim
The instructor writes one line. The AI turns it into a full accounting exercise. The student posts the journal entry, hits Validate, and gets a scored, line-by-line correction in seconds — not next week.
An infinite exercise book that corrects itself
AccountingSim is an AI-powered accounting trainer. Instructors author exercises by describing them in one sentence; the AI writes the case. Students post debit and credit lines against the Spanish chart of accounts and receive an instant, structured correction with a 0–100 score. Practice stops being limited by how many exercises a textbook contains, or by how fast one instructor can mark.
One line in, one case out
The instructor writes a prompt — “a sale with 21% VAT on credit” — and the AI drafts the full exercise, constrained to accounts the student can actually select.
A real journal, not a quiz
Students pick accounts, enter amounts in the debit or the credit column, watch running totals, and see their T-accounts build. No multiple choice anywhere.
Graded, not just marked
The AI returns a 0–100 score, a general verdict, a per-line judgement on every account posted, and one concrete hint when something is off.
Attempt, fix, attempt again
Every validation is numbered and stored. The student can revalidate immediately; the instructor sees the whole progression, not just the last answer.
Double entry is learned by repetition — and repetition is what nobody can afford
Every accounting teacher knows the drill: students only internalise debit and credit by posting entry after entry. But the exercise book runs out, and marking eighty journals by hand is a weekend gone. So practice gets rationed exactly where it should be unlimited.
A finite exercise book
A textbook chapter carries a dozen cases. Motivated students burn through them in an evening and then have nothing left to practise on.
Feedback arrives too late
By the time a marked journal comes back, the student has forgotten the reasoning that produced the error — so the correction teaches far less than it could.
Marking does not scale
Manual correction is slow and expensive, and it is the single reason class sizes cap how much deliberate practice a course can offer.
Criteria drift
Two markers — or one marker at 9am and at midnight — apply subtly different standards. Students notice, and it corrodes trust in the grade.
What AccountingSim actually removes
Not the teaching — the bottleneck. The instructor keeps authoring the cases and running the room; what disappears is the finite supply of exercises and the queue of unmarked journals.
Four moves, two of them run themselves
The whole product is one short loop. The instructor touches it once at the start; after that it runs between the student and the AI for as long as the student wants to keep practising.
Writes a prompt
One sentence describing the case.
Drafts the exercise
In the instructor's language, using only bookable accounts.
Posts the entry
Accounts, debit and credit, until it balances.
Grades it
Score, verdict per line, and one hint.
Try again
Numbered attempt, stored for the instructor.
The instructor authors the exercises
In the authoring table, each row is one exercise. The instructor types only the instruction — “purchase of merchandise on credit with 21% VAT” — and saves. The description column shows a placeholder until the AI fills it in.
The AI writes the case
A background worker picks up the request, generates a mid-level exercise in the instructor's own language, and is explicitly forbidden from using any account outside the platform's chart — so the case is always solvable with the dropdown the student sees.
The student posts the journal entry
Pick an account, type an amount into the debit or the credit column, add the line. Totals update live and the T-accounts redraw underneath. Any line can be edited or deleted before validating.
The correction comes back scored
The platform checks that the entry balances before any AI is called, then queues the grading and waits for the verdict. The student gets a score, a sentence of overall judgement, a mark against each line they posted, and one concrete hint.
The journal, the totals and the T-accounts — on one screen
The workspace mirrors how accounting is actually taught: post to the journal, watch whether it balances, then see the same movements land in the ledger. The student never has to imagine the second step.
| Account | Debit | Credit |
|---|---|---|
| (430) Trade receivables | — | |
| (705) Services rendered | — | |
| (477) Output VAT payable | — |
Change any amount to watch the totals and the ledger react, exactly as they do on the student's screen.
Illustrative reconstruction of the student screen. The amounts follow a 1,200 EUR service invoice with 21% Spanish VAT.
What the student can do
Add a line by choosing an account from the dropdown and typing an amount into the debit or the credit column. Edit or delete any line already posted. Watch both totals recalculate on every change, with an explicit balanced or not balanced state.
Below the journal, the ledger rebuilds itself: one T-account per account touched, with its debit side, its credit side and its resulting balance. It is the same data, shown the way the exam will ask for it.
The balance is checked before the AI is called
An empty or unbalanced journal is rejected instantly with a plain-language message. The student learns the discipline of balancing first, and no AI call is ever spent on an entry that cannot be right.
Not “wrong” — but which line, and why
The AI is asked to return a structured verdict, not prose. That structure is what makes the feedback teachable: a score to gauge progress, a sentence of judgement, a mark against every single line the student posted, and exactly one hint so the next attempt is theirs to make.
- ✓ (430) Trade receivables— correct: the receivable carries the full invoice including VAT.
- ! (700) Sales of goods— wrong account: this is a service, so it belongs in 705.
- ✓ (477) Output VAT payable— correct: 21% on the taxable base.
Illustrative reconstruction of the feedback panel, following the verdict structure the platform requests.
The four verdicts
Every line the student posts comes back tagged with one of four judgements — which is what lets an instructor see error patterns across a whole class rather than reading eighty paragraphs of prose.
| Correct | Right account, right side, right amount. |
| Wrong account | The operation was understood but booked against the wrong PGC account. |
| Wrong amount | Right account, arithmetic or VAT treatment off. |
| Unnecessary account | A line that should not be in this entry at all. |
The score is clamped to 0–100 and the attempt counter increments per student and exercise, so progression across attempts is measurable rather than anecdotal.
46 accounts, curated so every generated case is solvable
The platform ships a working subset of the Spanish Plan General Contable, spanning all seven classes. It is deliberately not the whole PGC: the AI that writes exercises is constrained to exactly this list, which guarantees the student can always find every account the case requires in the dropdown in front of them.
Basic financing
Share capital, reserves, prior-year results, long-term bank debt.
6 accountsNon-current assets
Land, buildings, machinery, furniture, IT equipment and accumulated depreciation.
7 accountsInventories
Merchandise and raw materials.
2 accountsReceivables and payables
Customers, suppliers, bills, payroll payable, and both sides of VAT.
11 accountsFinancial accounts
Cash, current accounts, short-term bank debt.
4 accountsPurchases and expenses
Purchases, rent, professional services, utilities, wages, social security, interest, depreciation.
12 accountsSales and income
Sales of goods, services rendered, own work capitalised, grants.
4 accountsBookable set
Enough to cover VAT sales and purchases, payroll, fixed-asset depreciation and financing — the backbone of an introductory course.
46 accountsSet up in one form, run from one board
Creating a session is a single dialog; from that point on the instructor has a live board of teams, a place to author exercises, and reporting — without leaving the platform.
1 · Create the session
Name, institution, number of parallel worlds, teams per world, participants, session and due dates, instructor, and naming convention. Sensible defaults are pre-filled — a single class is one world.
One dialog2 · Author the exercises
A two-column table: your instruction, and the description the AI writes from it. New, Edit and Remove — nothing else to learn. Exercises are scoped to the session, so each class gets its own set.
Session-scoped3 · Run the room
The games board lists every team with its online state and round, a global play and pause switch, per-team access links, and the ability to enter a team's view to see exactly what a student sees.
Live controlWhat the setup form actually asks
| Field | Why it matters to a teacher |
|---|---|
| Games / Teams / Participants | Sizes the class. One game world is a single class; teams are the decision units inside it. |
| Session date / Due date | Opens the window students work in, and when the set closes. |
| Version | Which build of the simulation this cohort runs. |
| AI integrity monitoring | Whether AI-assistance signals are collected for this session. |
| Naming convention | How teams are labelled — 1, 2, 3 or alternatives. |
From an empty session to students posting entries
One form, one table and a handful of one-line prompts. There is no content to upload, no answer key to write and no marking rubric to configure — the exercise and its correction are generated from the same sentence.
Every one of these is a navigation item the instructor already has.
Nothing reaches the model until it is worth grading
Institutions increasingly ask what an AI feature costs them and what stops it running away. The answer here is built into the product: unbalanced work never reaches the model, duplicate requests collapse into one, and usage per student is bounded by design.
Balance pre-check
Before anything is queued, the server sums the student's own lines for that exercise. An empty journal, or debits that do not equal credits, is rejected locally with a teaching message — no AI call at all.
In-flight de-duplication
If a validation for that student and exercise is already queued or running, the platform hands back the existing job instead of stacking a second one. Impatient double-clicking costs nothing.
Rolling rate limit
Validations per student and exercise are capped within a rolling ten-minute window, so neither a script nor a frustrated student can drive unbounded usage.
Authoring is gated by role
The exercise endpoint refuses students outright, so exercises can only be created by the instructor who runs the session.
Cross-session write guard
Edits and deletions are verified against the session that owns the row, so an instructor in one class cannot touch another class's exercises by submitting their identifiers.
CSRF on the grading endpoint
Every state-changing request is rejected unless it carries a valid token minted in the student's own dashboard.
A question worth asking of any AI tool
How many model calls does one student generate in an hour, and what stops that number growing? Here the ceiling is part of the design rather than something to discover afterwards.
The question every teacher asks second
The first question is whether the tool saves time. The second is always how you know the student did the work. AccountingSim answers it structurally rather than with a promise.
Integrity monitoring
A per-session setting that governs whether AI-assistance signals are collected for that cohort.
Every attempt is kept
Attempts are numbered per student and exercise, so the trajectory from first try to final answer is visible.
Audit log
Creations, edits and deletions on exercises are recorded with their previous values.
LTI access log
Entries arriving through the institution's LMS are logged, tying activity back to an identity the school already trusts.
What the platform does, and what it does not claim
It records attempts, logs changes and can collect integrity signals per session. It does not claim to detect whether a student used an outside AI to produce an answer, and we would rather say so than imply otherwise.
What a student actually walks away with
These map to what an introductory financial accounting course is examined on, which is what makes the tool defensible in a syllabus conversation rather than a nice extra.
The accounting equation
Assets = Liabilities + Equity, and the discipline of keeping it true through every single transaction.
The nature of accounts
Which accounts increase on the debit side and which on the credit side — the distinction most beginners get wrong.
Double entry itself
Every operation touches at least two accounts, and total debits equal total credits. Enforced by the interface, not just stated.
Choosing the right account
Mapping a described transaction onto the PGC: goods versus services, customers versus debtors, purchase versus expense.
VAT and arithmetic
Applying VAT correctly on both sides, and separating the taxable base from the total invoiced.
Analytical breakdown
Splitting a complex operation into its components and checking that the solution accounts for all of them.
Where it fits, and where it does not
AccountingSim is a drilling tool for the foundations. Being clear about that makes the first conversation shorter and the fit better.
Undergraduate accounting
Introductory financial accounting, where volume of practice is the single biggest predictor of who passes.
Core fitVocational training
Administration and finance programmes that need repeatable, gradeable practice against the Spanish PGC.
Core fitMBA refreshers
Pre-programme levelling for students arriving without an accounting background, before the finance core begins.
Strong fitSelf-study and resits
Students who need unlimited extra repetitions between classes, or who are retaking the subject.
Strong fitCorporate onboarding
Finance-adjacent staff who need to read a journal entry without becoming accountants.
SecondaryWhere it does not fit
Consolidation, IFRS-specific treatments and advanced audit are out of scope, as is any market whose chart of accounts is not the Spanish PGC.
Out of scopeAccountingSim at a glance
The essentials, in one place.
Answers to the most frequently asked questions
No. You write one sentence describing the case — a purchase of merchandise on credit with 21% VAT, say — and the AI drafts the full exercise from it. There is no answer key to write and no marking rubric to configure: the exercise and its correction come from the same instruction.
A working subset of the Spanish Plan General Contable: 46 accounts spanning all seven classes, from share capital and fixed assets to VAT, payroll and financial expenses. The exercise generator is constrained to exactly that list, so every case is solvable with the accounts the student sees in the dropdown.
A 0–100 score, one sentence of overall judgement, a verdict against every line they posted — correct, wrong account, wrong amount or unnecessary account — and one concrete hint. It arrives in seconds, in the student's own language.
The journal has to balance before anything is graded, every attempt is numbered and stored, and validations are bounded per student and exercise within a rolling window. The instructor sees the whole progression from first attempt to final answer, not just the last one.
Yes. AccountingSim connects over LTI and logs the accesses that arrive that way, tying activity back to an identity your institution already manages. The interface runs in Spanish and in English.
Let us talk about your course
Tell us how you teach double entry today and we will show you where AccountingSim fits.
General inquiries
info@eurekasimulations.comPhone
(+34) 877 245 676Technical support
tech@eurekasimulations.comReady to get started?
Give your students an exercise book that never runs out
Request a demo and we will walk you through authoring a case, posting an entry and reading the correction.