Course evaluation comments in the same themes for two terms, every quote cited, every removal confirmed by a person
Fixable teaching issues by course in minutes, and no biased comment in a faculty file.
See one case, screen by screen ↓Course Evaluation Comment Analysis gives deans and chairs their course evaluation comments grouped by theme, quoted and cited, with abuse removed. A chair asks a plain question and agents line up two terms in the same themes, with counts per 100 responses and two quotes per theme cited to the section. Comments about protected traits are held for Faculty Affairs to confirm, and comments that may describe harassment go to the Title IX Coordinator.
AI in education: a buyer's guide to agentic AI for colleges, universities and school districts →
Updated
Course evaluation comments arrive by the thousands each term, and a department chair reads them the way the course evaluation report delivers them: section by section. A fourteen-section course like BIO 1010 brings about 1,200 student evaluation comments a term. Whether a summer lab redesign worked becomes a judgment from memory, because nobody can count last fall's complaints against this fall's in the same terms.
The same reports carry the worst comments. Remarks about an instructor's accent, appearance, race or gender sit beside real criticism of teaching and go into the annual-review file, and published research on bias in student evaluations of teaching shows why that matters. An anonymous comment that describes unwanted contact by a teaching assistant is notice to the university, yet in a PDF read in February it is weeks late. Course Evaluation Comment Analysis puts agents in front of the reading, and people in front of every removal.
Estimated split of one department chair's working days on a term's end of course evaluations, by hand and with the agents; our arithmetic, kept modest.
Eight agents strip names, sort comments into the same themes each term, quote and compare them, and hold biased comments for Faculty Affairs to confirm.
Halden State University released its fall 2026 student course evaluations to chairs on Tuesday, January 19, 2027. Biology chair Dr. Celia Marchetti wants to know whether the summer lab redesign for BIO 1010 worked. Here is what happens, screen by screen, in the working solution.
The home screen is a single question box with her scope underneath: Biology, fall 2026, 186 sections, 8,940 comments. Across the university, 1,412 student names were removed before any agent read a comment, and 214 sections under the 5-response floor are not shown at all.
She asks "What do students say about BIO 1010 labs since the redesign?" The agents run step by step in place, then the answer lines up fall 2025 (1,524 responses, 66%) and fall 2026 (1,571 responses, 68%) theme by theme, with a count, a rate per 100 responses and two quotes for each. "Pre-lab videos help" reads "Not mentioned" for fall 2025, because the videos started in fall 2026.
"Lab instructions unclear" fell from 212 mentions to 48: 13.9 to 3.1 per 100 responses.
Clicking the Fall 2026 · Section 004 citation opens the comment: "The new checklist at the top of each lab made it obvious what to do first." Beside it are the course, section, instructor, the course evaluation survey question and the section's 120 responses. The Quote agent says why it picked this one: 30 of the 48 comments in the theme say something close, it is 16 words, and it is not held.
The Quote agent never quotes a held comment or a comment with a name in it.
"Lab instructions unclear" opens section by section for both terms, and section 011 goes from 19 mentions to 8. The Theme agent proposes splitting out "Titration step 6": 11 of the 48 fall 2026 mentions are about when to read the burette, in 6 sections. The chair accepts the split or declines it with a reason; the agent never changes the themes on its own.
Under the answer, one line says 3 comments were removed. Two are about the instructor's accent: the Bias Filter held H-5520 and H-5521 under B-2, national origin or accent, at 0.94 and 0.88. H-5521 also says the lab handouts skip steps, so that point is counted in "Lab instructions unclear" but never quoted. The third went to the Title IX Coordinator instead of the filter.
Text is shown only to Faculty Affairs and the Title IX office.
Omar works a short queue of held comments on a 3-business-day clock. For H-5521 he reads the comment, reason code B-2, the agent's reason and its confidence of 0.88 against the 0.70 threshold, then confirms the hold. He could instead restore it to the report with a K-code. Neither comment enters any faculty file.
The Bias Filter never deletes a comment and never holds criticism of teaching.
The comment from section 006 describes unwanted contact by a lab teaching assistant. The Disclosure agent routed it on January 15 at 11:20 AM with the course, section and meeting time, under code T9-1, while the respondent stays anonymous. Teresa acknowledged it at 2:47 PM, opened case TIX-2027-0031, and can prepare a resources notice to the section, which her deputy checks before it is sent.
Dr. Adeyemi teaches BIO 1010 sections 008–014. He opens the same question and gets the same themes, the same quotes and the same removed line, so his course evaluation results are never read differently behind his back.
"You see the same answer as your chair, Dr. Celia Marchetti: the same themes, the same quotes and the same removed line."
For Dr. Adeyemi's fall 2026 sections (793 responses, 621 comments), the File Rules agent builds the annual-review summary from theme counts only, under Article 17.4 of the faculty agreement, with the 2 held comments excluded. The chair signs it, Dr. Adeyemi has until February 2 to respond, and Faculty Affairs releases it to the faculty information system.
The trail shows Celia's question Q-30418, the Comparison agent's answer from 2,421 comments, 5 themes and 18 quotes, both holds and Omar's confirmation, and the Title IX routing marked not filtered. Gregory builds an export pack for a faculty review, an internal audit of removal controls or an accreditor: 38 files, a SHA-256 manifest, student names removed.
The same student feedback on teaching, seen by the five people who carry it: what their term looked like, and what it looks like now.
Eight agents strip names, sort comments into the same themes each term, quote and compare them, and hold biased comments for Faculty Affairs to confirm.
Pulls comments and responses by term and section from the course evaluation platform, and joins sections, meeting times and instructors of record from the student information system.
Removes student names from comment text and holds back every section with fewer than 5 responses, before any other agent reads a comment.
Assigns each comment to the course's pinned themes at 0.80 confidence or higher, and proposes splits, merges and new themes to the chair with counts.
Picks two representative quotes per theme and term, under 40 words, and cites each to its term and section.
Lines up two terms theme by theme, with counts and rates per 100 responses, and says whether a change could be a response-rate effect.
Holds comments about race, ethnicity, religion, national origin or accent, gender, appearance, age or disability, and slurs, threats or sexual remarks, with a reason code; marks any teaching point to be counted.
Routes any comment that may describe sexual harassment, stalking, dating violence or sex discrimination to the Title IX Coordinator, with the course, section and meeting time.
Builds the annual-review summary under the faculty agreement's article on student comments: theme counts only, verbatims only if the faculty member adds them.
Confirms or restores every hold, with a reason. The agents propose; a named person decides.
Anyone working in the solution can ask the assistant in plain words. Every answer cites its source and can open the screen it comes from.
Why were 3 comments removed from the BIO 1010 answer?
Two fall 2026 comments are about the instructor's accent, so the Bias Filter held them under B-2 (national origin or accent). A hold is final only when Faculty Affairs confirms it. One of them also says the lab handouts skip steps; that point is counted in "Lab instructions unclear" and never quoted. The third describes unwanted contact by a teaching assistant, so it went to the Title IX Coordinator, not to the filter.
Where else in Biology is "lab instructions unclear" rising?
In BIO 2150 Genetics: 31 mentions in fall 2025, 74 in fall 2026 (4.4 to 10.3 per 100 responses). Its lab manual has not been redesigned.
Draft Dr. Adeyemi's annual-review summary under Article 17.4
Drafted from fall 2026, sections 008–014: seven themes with counts only, as Article 17.4(a) requires. No verbatims unless Dr. Adeyemi adds them, and the 2 held comments are excluded. You sign it; Faculty Affairs releases it after his 10-business-day window.
Build the audit pack for BIO 1010, fall 2026
The pack will hold the question and answer with every cited quote, the 2 holds and their decisions, the Title IX routing as a time and a code only, and the agent versions in force. Internal Audit, Faculty Affairs or the Director builds it.
12 screens from the working solution, on its sample data. Pick one to see it large.
The Director of Faculty Affairs sets what is held, what is never held and what may enter a review file. Every change is saved as a new version and logged.
| Setting | Default | Choose from |
|---|---|---|
| Hold threshold | 0.70 | Stepped, with a preview of how many comments a person will read |
| Never held: criticism of teaching | On (K-1) | Harsh comments about teaching, grading or pace stay in the report |
| Never filtered: possible harassment | Routed to the Title IX Coordinator | Under the university's reporting policy |
| Bulk confirmation | A-1 profanity at 0.95 or higher | Or off; bias holds are always decided one by one |
| Response floor | 5 responses | 5, 8 or 10 responses |
| Keep comments | 6 years | 4, 6 or 8 years; legal hold suspends deletion |
| Hold clock | 3 business days | 2, 3 or 5 business days, then escalated to the Director |
| Instructors may read held comments about them | On request | On request, never or always; with Faculty Affairs present |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
Sources: "Where's the harm? Screening student evaluations of teaching for offensive, threatening or distressing comments", Australasian Journal of Educational Technology (6.9% of 62,049 comments flagged for a second look) · FERPA, 34 CFR Part 99 · Title IX, 34 CFR 106.44 · on bias in student evaluations of teaching: MacNell, Driscoll and Hunt, Innovative Higher Education (2015); Subtirelu, Language in Society (2015) · "target" = design goal, measured in the live solution · "demo" = seen in the working solution, on its sample data · "typical" = published figures · "estimated" = our arithmetic, kept modest. Halden State University, its faculty agreement, policies, courses and the people named on this page are fictional: characters and sample data in the working solution.
Course Evaluation Comment Analysis gives deans and chairs their course evaluation comments grouped by theme, quoted and cited, with abuse removed. A chair asks a plain question and agents line up two terms in the same themes, with counts per 100 responses and two quotes per theme cited to the section. Comments about protected traits are held for Faculty Affairs to confirm, and comments that may describe harassment go to the Title IX Coordinator.
Each course has a pinned set of themes, so both terms are sorted into the same themes in the same order. The Comparison agent gives counts and rates per 100 responses and says whether a change could be a response-rate effect. A theme missing in one term reads "Not mentioned", and the themes change only when the chair accepts a proposal, so the student feedback analysis stays comparable from term to term.
They are held, never deleted. The Bias Filter holds comments about race, ethnicity, religion, national origin or accent, gender, appearance, age or disability, and slurs, threats or sexual remarks, with a reason code, and a Faculty Affairs reviewer confirms or restores every hold. Criticism of teaching is never held, and a teaching point inside a held comment is still counted.
It is routed, never filtered. The Disclosure agent sends it to the Title IX Coordinator the day it arrives, with the course, section and meeting time, while the respondent stays anonymous. The coordinator can open a case and prepare a resources notice to the section, which her deputy checks; chairs and instructors never see the text.
Yes. The instructor opens the same question and sees the same themes, the same quotes and the same removed line. Instructors can ask for a review of a comment and add a response to their annual-review summary before it is released.
Student names are removed from comment text before any agent reads it, respondent identities are never imported from the course evaluation platform, and sections with fewer than 5 responses are never shown. Retention and legal hold are settings, and comments are never used to train models.
Yes. The agents read, sort, quote and screen; they never delete a comment, edit a student's words, identify a respondent, release a file or send a notice. Every hold needs a Faculty Affairs reviewer, the chair prepares and Faculty Affairs releases each review summary, and every override carries a reason code.
It reads comments from the course evaluation platform you already run and joins sections and instructors from the student information system. The Agentic Solution Engine builds and deploys it from your requirements and documents, such as your removal rules, faculty agreement and a past term of comments, and it goes live once every quality gate has passed.
Every course file fixed before April 2027, and heard right by a screen-reader test.
1-min demoAcademic Integrity Case ManagementReferrals screened on how the work was written, with both sides in one packet.
1-min demoLearning Outcomes AssessmentAssessment reports built from scored student work, where every number opens its papers.
1-min demoWe'll run Course Evaluation Comment Analysis on a past term of your own course evaluation comments.