What your students see.
If a student mentions Solas Marker, or you've found it, this page is written for you — not for a sales pitch. Here is exactly what it does, why it's formative, and where we draw the line.
The one-sentence version
Solas Marker gives a student a formative second opinion on a draft before they submit — an indicative mark range (never a single number, never a grade), examiner-style feedback tied to the student's own words, and a referencing check. It is the same category of help as a writing-centre tutor. It never rewrites the essay, never produces submittable text, and never claims to detect AI-written work.
What a student actually sees
A range across a labelled band scale (with an "indicative only" disclaimer fixed to it, non-dismissible), a short examiner-style summary, criterion-by-criterion feedback keyed to each of the rubric's criteria with the student's own words quoted back, the single clearest gap to the next band, and a referencing report. Every quoted phrase is sliced by the server from the student's own submission, so the feedback cannot misquote them. There is no black box "62–68" — the reasoning is shown: the criterion being judged, the questions a marker would ask, and references into the student's own text.
Our complaints stance — decided in advance, and published
The official mark is correct by definition. Solas does not arbitrate between its indicative range and a mark awarded by your institution.
If a student's Marker range differs from their real mark, that is expected — two experienced markers routinely land several marks apart, and Marker has less context than you do. We will not write letters, produce "evidence", or support appeals against university grading on the basis of a Marker output. Support enquiries of this kind receive a kind, canned version of exactly this.
We built the product to make this easy to hold: no single number, no midpoint, no outcome phrasing ("this would get you a First"), and criteria-level feedback rather than word-level number-moving — so the reward signal for gaming a grade simply isn't there.
The two marking passes — described honestly
Marker reads the essay, judges it against the rubric, writes the tough questions a demanding marker would ask, then marks it a second time, blind to the first reading and — where our deployment is configured with a distinct secondary model — on a different model. If it falls back to the same model, the student's report discloses it and tells them to treat the two readings' agreement with extra caution. A third step reconciles the two and settles the range. We are careful to call this what it is: a second reading designed to resist anchoring — not an independent second marker, and never "how real moderation works." Both readings share model lineage and may share blind spots, which is why the fairness safeguards, not the second reading, are the real check on correlated bias. The full pipeline is described on the transparency page.
Fairness and integrity
Presentation may only affect a presentation criterion, and only if the student's rubric has one — grammar, idiom and second-language phrasing are never treated as evidence about argument or knowledge. Marker refuses to rewrite text, generate answers, tell a student the exact words to change for a grade, or assess whether writing is AI-produced. Iterating a draft against criteria-level feedback is learning, not cheating.
If a student asks, Marker can also suggest a short reading list — capped at four papers, drawn only from the public research registries, chosen against the weaknesses its own feedback identified. The suggestions are optional, carry no outcome promises (the same formative language rules apply), name what to check before relying on each paper, and never replace or rank above the module's assigned reading. A student following them is doing exactly what we hope: reading more, critically.
Data and governance
Marker is stateless: a student's essay is sent once, marked in memory, and never stored on our servers; results live only in the student's browser. Marking prompts go to a single AI provider (currently Anthropic) and no other; reference metadata (titles, authors, years, DOIs — never essay prose) goes to CrossRef, OpenAlex, Google Books and the DOI system (doi.org) to check sources exist; if a student asks Marker to suggest readings, short search phrases describing the topics their question, module details and feedback identified — a few keywords, never sentences from the essay — also go to OpenAlex to find candidate papers, and the structured feedback (including its short quoted essay sentences) together with the candidate papers' titles and abstracts goes to the same single AI provider to choose and explain the suggestions. The full detail, including Anthropic's data-usage terms, is in the privacy policy. A data-processing addendum covering Marker is available on request.
Help us calibrate
The most valuable thing an institution can offer is real marked essays (with written consent, essays processed and only the scores retained) or a small lecturer panel willing to blind-review Marker reports against essays they marked. That is the only way to answer "would a real marker endorse this feedback?" honestly. If you're open to it, we'd be grateful.
Talk to us
Questions, concerns, or a Marker output that looks wrong: support@solastool.com. For academic-integrity, governance, or procurement escalation — including a request for the Marker data-processing addendum — email institutions@solastool.com with the subject "Marker — staff escalation". More detail for students and staff alike is on the how-it-marks page.