Assessment platform
Summa is one place to write items, assemble papers, sit them under a clock, mark what a machine cannot, and see whether the paper actually measured anything. Your programmes, your modules, your cohorts, your grade scale - configured, not coded.
A list of 1 000 000 sorted integers is searched for a value that is not present. Which growth rate describes the number of comparisons a binary search performs?
A university runs degrees across years. A technical institute runs diplomas across levels. A training centre runs one short course, once. All three are the same five nested things, named however you name them.
The course of study a candidate is enrolled on.
BSc Computer Science · Diplôme TDI
A stage within it. One level is a perfectly good programme.
Year 2 · Niveau 1 · Foundation
A subject that gets taught and assessed. Owns its question bank.
Algorithmics · Databases
The group moving through a level together this year.
TDI-201 · Class 2A
What they sit, assembled from the bank against a blueprint.
Midterm · Final · Resit
They happen in the same order in every institution, and each one hands something specific to the next. Summa is the whole line rather than one station on it, which is why nothing has to be exported at the end of a step and re-imported at the start of the next.
Nine item types, each with the answer key its marking actually needs: a tolerance on a number, accepted spellings on a phrase, a weight per blank on a cloze. Items are versioned, so editing one never rewrites a paper somebody has already sat.
→ a module's question bank
A blueprint says what the paper is made of - three fixed questions, then five drawn at random from the Loops pool at medium difficulty. Every candidate's copy is resolved from their own seed, so two people get different papers of the same shape.
→ one paper per candidate
Assigned to a cohort or to named candidates, opened in a window, timed by the server. Questions and options shuffle, answers save as they are given, and flagged questions come back at the end.
→ an attempt, with its seed
Everything with a key is marked the moment it is submitted, partial credit included. What is left goes to a queue: essays, with the rubric beside them, marked one question at a time across candidates rather than one script at a time.
→ a mark, and the reason for it
Results are held until you release them, then read against the grade scale the exam points at. What a candidate may see afterwards - score only, their answers, the correct answers, the explanations - is yours to set.
→ a grade the institution recognises
Facility and discrimination per item, distractor analysis per option, reliability for the paper. Adjust an item for everyone who sat it, and the marks are recomputed with the change recorded.
→ a better bank next term
Partial credit where partial credit is meaningful. Tolerance on numbers. Accepted spellings on text. Per-blank weighting on a cloze. Essays go to a person, with a rubric if you have one.
Facility and discrimination per item, distractor analysis per option, and a reliability figure for the whole paper. The question your strongest candidates got wrong is usually the one that is miskeyed, and Summa says so.
| Item | Facility | Discr. | Verdict |
|---|---|---|---|
| Q4 | 0.91 | +0.12 | Everyone got it |
| Q7 | 0.62 | +0.48 | Working well |
| Q11 | 0.44 | -0.21 | Check the key |
| Q18 | 0.31 | +0.35 | Hard but fair |
Reliability (α) 0.83 · SEM 4.2 marks · n 126
A machine in an exam hall can be an hour slow, deliberately or not. So the deadline is computed and held by the server when the attempt begins, and the countdown on screen is only ever a display of it. Answers are saved as they are given: a laptop that dies mid-paper loses the walk back to the invigilator, not the work.
Fullscreen, copy-paste blocking and focus-loss limits are offered, and described honestly in the settings that switch them on: they are deterrence, not proof. Anything that matters is decided by the server and written to the audit log.
Clocks are the server’s, not the candidate’s.
| 16–20 | Très bien | ≥ 80% | Pass |
| 14–16 | Bien | ≥ 70% | Pass |
| 12–14 | Assez bien | ≥ 60% | Pass |
| 10–12 | Passable | ≥ 50% | Pass |
| 0–10 | Insuffisant | ≥ 0% | Fail |
68.5% on this scale reads 13.75/20 · Assez bien
A French institute reports 14.5/20 with the mention Bien. A university reports B+ and 3.3 grade points. A training centre reports Pass. All three are the same boundary table with different rows, so a grade scale is data an administrator edits - not code somebody has to change.
Percentage, letter and out-of-20 scales are set up with every new institution, and ECTS and pass/fail are a preset away. Add your own bands, set what rounds to a quarter point, mark which of them pass, and point each exam at the scale it should be read against.
Permissions are capabilities, not job titles with a switch beside them. An instructor may author and mark - for their own modules, and the check that they own the module is made again on the server, every time.
Create your institution, invite your staff, and the structure is yours to shape. Nobody has to edit code to add a programme.
Get started