Teacher evaluation where every rating is tied to a line of evidence, and the evaluator decides and signs
Feedback written in days, not weeks, with every rating tied to the evidence.
See one case, screen by screen ↓Teacher Evaluation Evidence Review is teacher evaluation software for principals and assistant principals: seven agents tag every line of the observation script and the teacher's artifacts to the district rubric, propose a level only where the rubric's own wording fits, and draft the written feedback with every sentence cited. The evaluator decides every tag and rating and signs, and a second person checks the procedure before the teacher sees it.
AI in education: a buyer's guide to agentic AI for colleges, universities and school districts →
Updated
Teacher evaluation in a US public district runs on a cycle the state sets and the collective bargaining agreement details: notice before a formal observation, written feedback within a fixed number of school days, a post-observation conference, a window for the teacher to respond in writing, and, when practice falls short, an improvement plan with its own steps and dates. Principals and assistant principals do this work between lunch duty and discipline, against a teacher evaluation rubric of 22 components.
Today the evaluator re-reads a 45-minute script against that rubric, copies lines into each component's box on the teacher evaluation form, chooses levels and writes the feedback, often at night. Informal visits mostly never become written feedback, ratings drift with the time of day and the evaluator, and when an improvement plan is grieved the file is missing a notice, a date or the evidence behind a rating. Teacher Evaluation Evidence Review puts seven agents in front of the reading, so the evaluator spends the time deciding.
Estimated split of one formal write-up, by hand and with the agents, in hours: about 2.5 hours by hand against under an hour of review. Our arithmetic from principal-reported write-up times, kept modest.
Seven agents tag every line, propose levels only where the rubric's words fit, draft cited feedback and run the contract's clocks; the evaluator decides.
Jonathan Pruitt teaches Grade 5 at Halvery Elementary in Marlow Bay Unified School District and has been on improvement plan IP-2026-014 since September 8, for questioning and student engagement. His assistant principal, Simone Okafor-Grant, scripted his math lesson on Wednesday, October 14. Here is that teacher observation, screen by screen, in the working solution.
Simone evaluates 15 teachers. Her home shows four plain numbers, the work that needs her and a briefing from the agents: Jonathan Pruitt's written feedback is due in 2 school days, and notices for three formal observations on Thursday must go by 5:00 p.m. under Article 12.2. Released inside the clock so far: 22 of 22.
The Intake agent brought in the 15-line script, a 2-page lesson plan and Jonathan's self-reflection, masked 11 student names as S1–S27 labels and skipped 3 coaching notes. The Tagging agent placed 25 tags on 14 lines, each with its confidence, and the rubric map beside the script reads 16 of 22 components with evidence.
Written feedback due Wed, Oct 21, in 2 school days under Article 12.4; if late, the visit cannot count as formal observation 2 of 3.
For 3b Questioning and discussion the agent proposes Effective, wording match 0.88 on two lines: at 9:21, 22 of 27 students talking about S7's area model, and at 9:24, "Say more. How do you know?" It also shows what it weighed against that: the 9:09 recall question. Levels were withheld on 2b, 3a and 1d, and the components with no evidence yet become the Look-for list for the November 18 visit.
"Most questions ask students to explain their thinking, and many students take part in the discussion." Rubric v4 · 3b · level 3
At 9:40 the agent tagged 2b Culture for learning at 0.72 on a line about one student reading a library book after finishing early. One student says little about the culture of the whole class, so Simone removes the tag with the reason "Over-reads the evidence". The reason stays with the record and goes to the agent's review set.
S8 and S21 finished the set at 9:38; no extension task; S21 reads a library book until 9:43.
The draft runs in five parts — the lesson, each plan goal, checking for understanding, and questions for the post-observation conference — written in Simone's voice from her 38 released write-ups. The sentence that rested on the removed tag is struck, so the draft now carries 15 citations, and each one opens its script line, artifact or rubric wording.
dropped: rested on the tag you removed
Simone keeps the proposed levels — 3b Effective, which meets goal 1, and 3c Student engagement Developing — and leaves 1d, 2b and 3a not rated this visit. She signs version 2 with an attestation: 6 components rated, 3 not rated, 1 tag removed with its reason.
"I observed this lesson; the script is my record; I decided each rating."
Renée runs the release check: notice at least two school days before the visit, sent October 8; feedback inside 5 school days, with 3 used; every rating cites a line and the rubric wording; both plan goals addressed; student names masked and no coaching evidence. She confirms and releases it to Jonathan and the evaluation records system. Simone cannot run this check on her own observation.
Jonathan sees each rating with the lines behind it. He adds a photo showing the class has 7 fraction-strip sets for 13 pairs, with his October 2 request for more, and the exit-ticket results: 21 of 27 correct on all three problems. The same agents tag his evidence, and his written response is attached to the record under Article 12.5.
With the October 14 visit released, Jonathan's plan file passes all 14 grievance-ready checks: notice of deficiencies, human resources review by Gloria Teshome, the plan meeting the association representative attended, coaching attendance with content kept confidential, and every formal observation on its date. Goal 1 is met; goal 2 is still Developing, with the third formal observation set for November 18.
Across Simone's 61 observations, Domain 3 averages 2.6 after 1:30 p.m. (19 visits) against 3.1 before (42); her peers' gap is 0.1. The agent calls it a pattern, not a finding, shows its sample sizes and changes nothing. Simone plans a co-observation with Renée on Monday, October 26 at 2:10 p.m.
The same observation, seen by the four people who carry it — what their day looked like, and what it looks like now.
Seven agents tag every line, propose levels only where the rubric's words fit, draft cited feedback and run the contract's clocks; the evaluator decides.
Brings in the evaluator's script — typed, written on a tablet or dictated — with the teacher's artifacts and self-reflection, masks student names, and reads the class roster for size and schedule.
Tags each script line and artifact passage to the rubric component it shows, with confidence, and proposes a level only where two or more lines match one level's wording at 0.85 or higher, quoting it.
Drafts written feedback in the evaluator's voice — strengths, next steps tied to the plan goals, questions for the conference — with every sentence cited.
Runs the bargaining agreement's clocks in school days on the district calendar, warns and escalates, and checks every release and improvement plan file against the agreement.
Compares each evaluator's tags and ratings across time of day, subject and observation type, against peers and the district's anchor videos, and flags patterns with their sample sizes.
Assembles the year's evidence for each component and applies the state's scoring model to the ratings the evaluator decided.
Writes released observations to the evaluation records system and the HR system, attaches teacher responses, and builds grievance packs and the state educator reporting file.
Decide, release, respond, approve. The agents propose; a named person decides.
The evaluator can ask about any tag, level or sentence and see the line it came from. These are the questions the working solution suggests, with its answers.
Is anything in Jonathan's draft not backed by the script?
Every sentence cites a line or an artifact, 16 in all. One rests on a weak tag: 2b Culture for learning at 9:40, confidence 0.72. It supports "expectations are not yet high for every student", but one student reading after finishing says little about the class, and the rubric's 2b wording is about the whole class. If you remove the tag, I drop that sentence.
Which clocks are at risk this week?
Two today. Jonathan Pruitt's written feedback is due Wed, Oct 21, in 2 school days, and 3 notices for Thursday must go by 5:00 p.m. Nothing at Halvery is overdue; district-wide 2 are, both escalated to human resources.
Why did the Drift agent flag my afternoon ratings?
Across your 61 observations, Domain 3 averages 2.6 after 1:30 p.m. (19) against 3.1 before (42); for the 7 teachers you saw at both times the gap is 0.4, and your peers' gap is 0.1. It is a pattern, not a finding: a co-observation with Renée will show whether it is the lessons or the rating.
What does Jonathan's plan still need before the midpoint?
Three things: release the Oct 14 feedback by Wed, Oct 21; the third formal observation on Nov 18 by Renée Castellano; and evidence on goal 2, engagement, still Developing.
13 screens from the working solution, on its sample data. Pick one to see it large.
The rubric, the level gate and the agreement's clocks are settings, versioned and on the audit trail. A change becomes a new version and takes effect only after a second approver confirms it.
| Setting | Default | Choose from |
|---|---|---|
| Rubric | Framework for Effective Teaching v4: 4 domains, 22 components, 4 levels | Locked for the school year |
| Level gate | 2 or more lines matching one level's wording at 0.85 | A change runs the golden set first, then a second approver |
| Tag floor | 0.70 | Tags below it are not shown to the evaluator |
| Levels on informal visits | Never proposed | Locked by policy |
| Written feedback (Article 12.4) | 5 school days after a formal observation | Warn at 2 school days; 1 left escalates to the principal, overdue to HR |
| Teacher response (Article 12.5) | 10 school days, attached to the record | Counted on the district calendar |
| Retention | Evaluation records 6 years after separation | Drafts and removed tags 1 year; legal holds override |
| Override reason codes | Over-reads the evidence, Wrong component, Missing context I saw in the room, Evidence comes from coaching (not evaluative), Other | Overrides feed a monthly review set |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
Sources: Education Commission of the States, citing NCTQ (by the end of 2017, 24 states and DC required every teacher to be evaluated every year) · sample bargaining agreement clauses on observation timelines (written report within 5 school days) · Illinois Public Act 103-0085 (remediation plan within 30 school days of an unsatisfactory rating) · THE Journal, October 31, 2025 (principal-reported write-up time of two to three hours before an AI pilot) · FERPA, 34 CFR Part 99 · "target" = design goal, measured in the live solution · "demo" = seen in the working solution, on its sample data · "typical" = published figures · "estimated" = our arithmetic from published inputs. Marlow Bay Unified School District, Halvery Elementary, its bargaining agreement and the people named on this page are fictional — characters and sample data in the working solution.
Teacher Evaluation Evidence Review is teacher evaluation software for principals and assistant principals: seven agents tag every line of the observation script and the teacher's artifacts to the district rubric, propose a level only where the rubric's own wording fits, and draft the written feedback with every sentence cited. The evaluator decides every tag and rating and signs, and a second person checks the procedure before the teacher sees it.
Every script line and artifact passage is tagged to the rubric component it shows. A level is proposed only when at least two lines match one level's wording at 0.85 or higher, and the rubric's wording is quoted beside those lines. Where the wording does not fit, the component reads "No level proposed"; where nothing was seen, "No evidence yet", which becomes the Look-for list for the next visit.
Principals report two to three hours per formal write-up when they re-read the script against the rubric by hand. Here the evaluator reviews a script already tagged and a cited draft, and the design goal is every formal observation's written feedback released inside the agreement's five-school-day clock, with a warning at two school days.
In the working solution, the classroom observation feedback covers the lesson, each improvement plan goal with the evidence for it, what the teacher did to check for understanding, the next step, and questions for the post-observation conference. Each sentence cites its script line, artifact or rubric wording, and it is written in the evaluator's own voice from their released write-ups.
Every step the agreement requires, on its date and on the record: notice of deficiencies, human resources review, the plan meeting, supports, each formal observation and its feedback inside the clock, and the midpoint and final reviews. The Due-Process agent checks the file against the agreement's plan article, and HR approves every plan before it is issued.
The Drift agent compares each evaluator's ratings by time of day, subject and observation type, against peers and the district's anchor videos, and shows its sample sizes. It never changes a rating: the evaluator acknowledges the pattern, plans a co-observation or moves visits.
Yes. The agents never rate a teacher, release feedback, send a notice or use demographics. The evaluator decides every tag and rating, with a reason code when removing a tag or differing from a proposal; a principal or HR checks the procedure before release; and HR approves every improvement plan.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your rubric and its level wording, your bargaining agreement's evaluation article and a sample of past scripts — connected to your evaluation records, HR and student information systems. It goes live once every quality gate has passed.
Plans drafted in minutes, and no student judged on sessions that never happened.
1-min demoAI Lesson PlanningEvery lesson drafted from the adopted curriculum, every activity cited to a standard.
1-min demoAI Essay FeedbackStudents get feedback in days, not weeks, with every comment signed by their teacher.
1-min demoWe'll run Teacher Evaluation Evidence Review on a sample of your own scripts, rubric and agreement.