Sibme AI Transparency — AI-Assisted, Human-Confirmed

AI Transparency

AI-Assisted, Human-Confirmed

Human agency is at the center of Sibme's mission, and guides the design decisions around our AI.

Download this guide Security overview

A guide for districts, coaches, teachers, and the people who advocate for them.

Sibme Review · Mrs. Chen · Grade 3

District rubric selected

Done

Evidence submitted

Done

System analyzes artifacts

Done

Evaluator reviewing

Human judgment required.

Findings inform a conversation

∗ System boundaries

What Sibme does. What it doesn't.

Neither list is short. Both matter. Read both before you decide how you feel about this system.

What Sibme does

Produces evidence-cited analysis, every claim points to a specific moment in a recording or section of a document
Generates pattern-based ratings reflecting multiple artifacts, using your district's exact descriptor language
Honestly flags when artifacts don't support a confident rating, with guidance on what evidence would resolve the gap
Works against the framework your district chose, not its own definition of effective teaching
Produces a draft that the evaluator reviews, edits, confirms, or overrides before anything enters the formal record
Logs every Copilot session with the evidence sources used, auditable and traceable

What Sibme doesn't do

✕
Make personnel decisions, no teacher is hired, fired, or placed on improvement plans by AI
✕
Replace direct observation, Sibme analyzes submitted evidence, not a substitute for an evaluator who knows the students and context
✕
Rate practice not in the artifacts, insufficient evidence is named, not guessed
✕
Override evaluator judgment, the evaluator makes the final rating decision, always
✕
Operate without a district-selected rubric, every analysis is grounded in your framework
✕
Train on your district's data, evidence submitted to Sibme is never used to improve AI models

∗ The review process

Five stages. One non-negotiable.

Stage 4. Human review. Cannot be skipped, automated, or bypassed. Every report that enters the formal record has been confirmed by a credentialed evaluator.

Stage 1

District rubric selected

Admin selects the framework, rubric, and indicators. Sibme operates within those boundaries exclusively.

Stage 2

Evaluator collects evidence

Observations, artifacts, lesson plans, recordings, notes. Submitted by evaluator or teacher per the workflow.

Stage 3

System analyzes artifacts

Sibme generates a draft analysis. Evidence cited, indicators rated, gaps flagged. No human has confirmed it yet.

Non-negotiable

Stage 4

Evaluator reviews and confirms.

A credentialed evaluator reads the draft, checks citations, applies professional judgment, and either confirms, edits, or overrides.

Stage 5

Findings inform a conversation

The confirmed report enters the record and becomes the basis for a post-observation conference. Not the end of the process.

∗ Non-negotiables

Six protections that don't change when the model does.

These aren't features. They're constraints. They apply regardless of which AI model is powering the analysis.

District-selected rubric

Sibme uses your framework. Not its own definition of effective teaching.

Citation-backed claims

Every assertion points to a specific, verifiable moment in the evidence.

Pattern requirement

Ratings reflect patterns across multiple artifacts. Never isolated moments.

Direct observation governs

Self-report cannot override behavioral evidence from recordings.

Human review required

A credentialed evaluator confirms every report before it informs any decision.

Honest gaps + forward framing

What can't be rated is named. Every section includes coaching-oriented guidance on what to work on next.

∗ Questions we've heard

Questions from teachers, coaches, and union representatives.

We gathered these from real conversations. If your question isn't here, reach out and we'll answer it directly.

What if the AI gets something wrong?

Every claim is cited to source evidence. If a citation is wrong, the evaluator corrects it before the report informs any decision. The system does not have authority over the rating. It produces analysis the evaluator reviews. An AI error that passes through uncorrected is a human review failure, not a design feature.

What if the teacher disagrees with the report?

Disagreement is normal and expected. A Sibme report is structured analysis, not a verdict. The teacher can submit additional evidence, contest specific citations, or request a re-review. The formal rating reflects the evaluator's professional judgment, after that conversation, not instead of it.

Who owns the data?

The district. Recordings, documents, transcripts, and reports belong to the district that submitted them. Districts can export their data, delete it, or restrict its use. Sibme does not use district data to train AI models, ever.

Can a report be used in a personnel decision without my knowing?

No. Districts using Sibme are bound by their own evaluation policies and applicable employment law. No personnel decision is made based on a Sibme report alone, and teachers have whatever rights of review, response, and appeal their district policy and contracts provide.

What does “AI-Assisted, Human-Confirmed” actually mean?

It means the AI does the analysis work. Transcribing, citing, pattern-finding, drafting, and a credentialed human evaluator confirms whether that analysis accurately reflects what they observed. The report that enters the formal record has been read, checked, and signed off by a human. The AI assists; it doesn't decide.

What happens if I submit evidence I'm worried about?

Sibme analyzes what's in the evidence, honestly and without softening. If a lesson shows areas for growth, the analysis will reflect that. This is by design: the point of evidence-based review is to have a shared, accurate picture of practice. The report becomes the starting point for a coaching conversation, not an endpoint for a judgment call.

AI-Assisted, Human-Confirmed

What Sibme does. What it doesn't.

What Sibme does

What Sibme doesn't do

Five stages. One non-negotiable.

District rubric selected

Evaluator collects evidence

System analyzes artifacts

Evaluator reviews and confirms.

Findings inform a conversation

Six protections that don't change when the model does.

Questions from teachers, coaches, and union representatives.

How Sibme protects instructional data.

Download the complete AI transparency guide.