The Sheet You Were Sent: Spreadsheet Reasoning Under ConstraintThis does not test whether you know where a menu item lives.
It tests what you do with a sheet you did not build — and it publishes the cost matrix behind your score. Thirty-two written exercises, thirty minutes, no software to open, and a report that shows whether your skill is even or spiky rather than averaging it into one number.
A skill test, not a software quiz — and the scoring is published
Almost every spreadsheet test on the market checks whether you can find a function, name a shortcut or recall a menu path. That measures how long you have used one particular product. This measures the thing the job actually consists of: a workbook somebody else built lands in your inbox, somebody wants a number out of it by four o'clock, and nobody is going to check your working. Twenty reading and diagnosis exercises ask what a total actually covers, which row is wrong, what a formula will do when it is copied, and what a filter took out of view before the file was sent to you.
Twelve of the thirty-two are build exercises, and they are the part no quiz has. Each one states a small sheet in full, then a hard constraint set — it must still be right when a row is added, the data you were sent must go back unchanged, somebody who has never seen it has to be able to check it — and asks which steps you would take. Every list contains a step that actively breaks one of those constraints and a step that merely wastes an afternoon. Both are errors, and they are not the same error, which is why they are not scored the same.
The scoring is a decision-cost matrix rather than a mark out of thirty-two. Every option carries an authored cost: breaking a hard constraint costs five to eight units, a slower route to the same right answer costs one to three. Your score is your cost expressed against what a respondent choosing at the declared rates would have cost, so zero means typical rather than nothing right. The whole weight table is printed on your report as a value judgement you can argue with, alongside the reliability assumption behind every band and a plain statement of what the assessment does not measure. No commercial product, vendor or brand is named anywhere in it, because knowing your way around one particular application is not the skill being measured.
What you walk away with
One cell per exercise, four strips, so you can see the texture: even reading and spiky building look identical in a percentage and are completely different colleagues. Each strip carries a spikiness reading in words, four greyscale steps, and a letter in every cell so nothing depends on colour.
Your cost is split into the part that came from breaking a hard constraint and the part that came from inefficiency, reported separately from correctness. Two people can lose the same number of points and be entirely different hires: one of them is slow, and one of them changes the source data.
The headline is corrected against authored per-option base rates rather than a uniform guess, so zero is the typical respondent rather than an empty page. The raw figure, the chance figure and the expected cost under chance are all printed, so the correction hides nothing.
Eight weights, what each one is the cost of, and the sentence that justifies the ordering: a wrong total that goes to somebody else costs more than a slow method, and a change to data you were sent costs more than a redundant column. If you disagree, you can see exactly what you disagree with.
Every headline carries its standard error at the sixty-eight and ninety-five per cent spans, written out as sentences. The four parts carry a three-way standing and no number at all, and the reliability table that justifies withholding those numbers is printed rather than the refusal being made in silence.
Inside your report
Illustrative sample — your report is generated from your own responses.
An even strip and a spiky strip are different findings. No summary number shows consistency, so the strip shows it and the report says the order the cells are in.
| Weight | What it is | Example |
|---|---|---|
| 5 to 8 | broke a hard constraint | changed the source data, or sent on a total that was wrong |
| 1 to 3 | took the long way round | a redundant column, a manual step that will not survive a new row |
| 0 | the keyed build | does the job and still works when the sheet grows |
A cost matrix is a value judgement, not a measurement. It belongs on the page where you can disagree with it.
Built for
- Employers screening for analyst, operations, finance, admin and coordinator roles, where a workbook is the main tool of the job
- Recruiters who want a spreadsheet check that is not a quiz about one vendor's menus and cannot be passed by product familiarity alone
- Hiring managers comparing shortlisted candidates on how they handle data somebody else prepared
- Candidates who work in spreadsheets daily and want an honest, private read on where their habits cost somebody downstream
- Team leads deciding who can be handed a shared workbook that other people will keep using after them
The sheet lands in somebody's inbox either way.
Thirty minutes, one sitting, the full report immediately — including the cost matrix behind the score.
₹999 (incl. GST) · assessment and full report, nothing further to pay
Frequently asked questions
What a person does with a spreadsheet they did not build. Four things, eight exercises each: what a stated figure actually covers, where the sheet is wrong, what to build when the answer has to survive a new row without disturbing the data, and what to leave behind so somebody else can check it. It does not measure typing speed, memory for where a menu item lives, knowledge of any particular software product, or how fast a person works.
No, and that is deliberate. Every sheet is written out in full inside the exercise — the columns, the rows, the formulas, what is filtered, what is hidden — so nothing has to be opened and no product has to be installed. No commercial product, vendor or brand is named anywhere in the assessment. Familiarity with one vendor's menus is a construct-irrelevant knowledge dependency, not part of the skill, and scoring it would be unfair to anyone who learned on something else.
It is scored by a decision-cost matrix and yes, the matrix is printed on your report. Every option carries an authored cost: breaking a hard constraint carries five to eight units, a slower route to the same right answer carries one to three, and the keyed step carries none. On a build exercise cost accrues for each keyed step you leave out and each harmful step you put in. Your total is expressed against the cost a respondent choosing at the declared per-option rates would have run up, so zero on the reported scale means typical rather than nothing right, and the expected cost under chance is stated as a figure on the page.
₹999 in India or $9.99 internationally, inclusive of GST, for one sitting and the full report. Spreadsheet proficiency tests sit in every incumbent pre-employment library, but they are generally only reachable inside a platform subscription of roughly $200 a month, and the standalone per-test products sit near $20 to $30 with no scoring model published. This is one sitting, a complete report and the cost matrix behind the score, with nothing further to unlock.
It was built against that and then tested for it. Every fixed habit a bored respondent can actually execute was pushed through the real scorer: selecting every step, selecting no step, taking only the first option, always taking the shortest option and always taking the longest option each land between nine and fifty-five points below the typical respondent. Option lengths are balanced within each exercise, the keyed option is the longest in a quarter of them, key positions are spread evenly across all four, and the build exercises never key every option.
Each one takes a single capability, puts you inside the situations where it is actually tested, and scores your choices against published evidence — with a report designed for that capability alone, not a template. They span hiring, compliance, education, operations and personal skill.
Browse the catalogue →Methodology: Measures applied spreadsheet reasoning through original reading, diagnosis and build items scored by an authored decision-cost matrix. Construct statement: It measures what a person does with a spreadsheet they did not build: what a stated figure actually covers, where the sheet is wrong, what to build when the answer has to survive a new row without disturbing the data, and what to leave behind so that somebody else can check it. It does not measure typing speed, memory for where a menu item lives, knowledge of any particular software product, mathematics beyond arithmetic, how fast a person works, or general intelligence. Declared response instruction: knowledge and skill throughout. Every item asks what is true of the sheet that is written out in front of the respondent, or which steps they would take under the stated constraints. No item asks what the respondent is most likely to do, no self-report scale is used, and no situational-judgment item appears, so knowledge framing and behavioural-tendency framing are never mixed in this instrument. Item format: a constraint puzzle build, authored as a fixed and fully stated position. A static item bank cannot hand anybody a manipulable grid, so each of the twelve build exercises writes the sheet out in a few short lines, states a hard constraint set and one soft objective, and scores the build decision rather than the keystrokes. The position is identical for every respondent and does not respond to anything chosen earlier. Every build exercise offers at least one step that actively breaks a hard constraint and at least one that is merely wasteful, so a commission error and an omission error are both reachable, and the keyed set is never every option. Twenty reading and diagnosis items state a sheet and ask one question about it: what a figure covers, which row is wrong, what a formula will do when it is copied, or what a filter took out of view. All content is product neutral. Cells, columns, rows, formulas, filters, totals and lookups are described in plain language and no commercial product, vendor or brand is named anywhere, because product familiarity is a construct-irrelevant knowledge dependency rather than part of the skill. Scoring design: a decision-cost matrix. Every option on every item carries an authored cost, which is the cost of getting the decision wrong in the direction that option represents, and an authored base rate, which is the declared marginal share of respondents expected to choose it. The base rates are deliberately uneven and the plausible-but-wrong option carries more mass than a uniform null would give it. On a build item cost accrues for each keyed step left out and for each harmful step put in; a step that breaks a hard constraint costs between five and eight units while a merely wasteful step costs between one and three, and that asymmetry is the entire point of a cost matrix. On a reading item the cost is the cost of the option chosen. The score is the obtained cost expressed against the cost expected under the declared marginals, normalised so that zero is the typical respondent and one hundred is the keyed build throughout. The expected cost under chance is computed from the authored base rates rather than from a uniform-random null, is stated on the report as a figure, and will be replaced by observed marginals once live sittings exist. The cost matrix is a value judgement rather than a measurement and is printed in full on the report. Only answered exercises enter both sides of the ratio; an unanswered exercise leaves the denominator rather than scoring zero, and coverage below sixty per cent suppresses the headline figure. A second reading, reported separately from correctness, splits the respondent's own cost into the part that came from breaking a hard constraint and the part that came from inefficiency, reporting the signed and the absolute figure and letting the absolute figure decide the verdict. The headline carries its standard error of measurement at both the sixty-eight and the ninety-five per cent spans, in words as well as drawn, and the reliability behind those spans is a stated assumption rather than a measurement until enough sittings exist. Each of the four parts of the skill carries a three-way classification and a per-item heat strip, and never a number or a percentile; the reliability table that justifies that refusal is printed on the report. No percentile appears anywhere. Construct areas and source families drawn on: research on spreadsheet error rates and error taxonomies in operational workbooks, including the distinction between qualitative and quantitative errors and the finding that most field-audited workbooks contain at least one; aggregation and double counting where a subtotal sits inside the range of a grand total; the unweighted average of averages and the difference between a mean of rates and a rate of totals; denominator neglect and the instability of a ratio built on a very small base; missing-data conventions, in which a blank cell, a zero and a text placeholder are three different things with three different effects; displayed precision against stored precision, and the order of rounding and totalling; relative and absolute reference behaviour under copy and fill; range integrity, covering off-by-one ranges, hidden rows, filtered views and what a plain total covers; type coercion, in which numbers and dates held as text are silently stepped over; positional lookup fragility under column insertion, against key-based joining; date arithmetic conventions and inclusive against exclusive day counts; like-for-like comparison and the partial-period trap; the separation of inputs, calculation and output as a maintainability principle in end-user computing; named and labelled parameter cells against constants hard-coded inside formulas; provenance and reproducibility at handover, covering the stated as-of date, the stated source and the stated exclusions; the reviewability literature on what makes an artefact checkable by somebody who did not build it; cost-sensitive decision analysis and asymmetric error cost; and set-overlap partial credit, which penalises over-selection and under-selection at the same rate. All items are original works. No trademarked instrument, branded methodology, commercial software product or vendor is named or reproduced anywhere in this assessment, and no affiliation with any source is claimed.