Architecture of Grace · For review committees

The logic model

How this curriculum is supposed to work — link by link, with the published research behind each link, and a plain statement of what a logic model does and does not prove.

Who this is for. Curriculum directors, grant writers, and anyone who has to put an evidence level on an adoption form. Last reviewed: September 20, 2026.

Why this page exists

Districts are often required to state an evidence level for anything they adopt. Under the Every Student Succeeds Act there are four, and the first three all require a completed study: a randomized trial, a quasi-experimental design, or a correlational study with controls. No efficacy study of this curriculum has been conducted — Evidence & Roadmap says so at length and commits to not implying otherwise.

The fourth level, “demonstrates a rationale,” asks for something different in kind: a well-specified logic model informed by research, plus an effort already underway to examine the intervention’s effects. It is the level that explicitly does not assert a demonstrated outcome. This page is that logic model, written so a reviewer can judge it rather than take it on trust.

The claim being made, exactly. This page sets out a rationale that meets the criteria described for the fourth ESSA evidence level. It does not assert that Architecture of Grace is a Tier IV intervention. That determination belongs to your district or your state education agency, not to the person who wrote the curriculum. If anyone representing this program tells you otherwise, they are wrong, and the no-overclaim commitment is the page to hold them to.

Where this sits among the four levels

Level	What it requires	Status here
Strong	At least one well-designed, well-implemented experimental study	Not met — no study exists
Moderate	At least one well-designed, well-implemented quasi-experimental study	Not met — no study exists
Promising	At least one well-designed correlational study with statistical controls for selection bias	Not met — no study exists
Demonstrates a rationale	A well-specified logic model informed by research or an evaluation, and an effort underway to examine the intervention’s effects	Both conditions addressed below. The model is this page; the effort is Stage 1 of the validation roadmap

Level names and requirements are paraphrased from the ESSA definition of evidence-based activities. Read the statute and your state’s own guidance rather than this table before you rely on it.

The model

Inputs · what a school puts in

What the program requires to run

A classroom teacher, and the curriculum book for that grade band with its companion novel.
A school counselor available for the flagged lessons — roughly 60 committed hours a year in a one-counselor K–8 building, costed openly on Counselor Capacity.
Scheduled instructional time: 20–25 minutes in K–2, rising to 40–50 in the upper bands, with 60–75 for flagged lessons.
Crisis resources posted and current; the counselor’s name and same-day contact known to the teacher.
The school’s own psychologist or counselor reviewing the most sensitive units and confirming clinical readiness in writing before delivery.
No license, account, subscription or purchase. The materials are open.

Activities · what actually happens

What teachers and students do

A lesson read against a chapter. Each lesson sits against a scene in the companion novel, so the concept arrives as something that happened to a character before it is asked of the student.
Four strands, taught in sequence and revisited: identity, self-compassion, forgiveness, grace — mapped to standards lesson by lesson on Standards & Alignment.
Named practice in the moment of failure — mistake-and-repair routines, the distinction between what I did and what I am, and a shared vocabulary the adults use too.
Regulation tools available continuously, not only during the lesson.
Family letters sent home in English and Spanish together, so the same words are available outside the room.
Disclosure protocols in every lesson, with counselor-in-room requirements on the highest-sensitivity ones.

Outputs · what can be counted

What a pilot school can actually observe

Lessons delivered, deferred, or shortened — and under what staffing. Deferral records are data, not embarrassment.
Counselor hours actually spent against the hours the calendar asked for.
Family letters sent; both languages or neither.
Student self-reflection responses, held in the school’s own spreadsheet.
Disclosures surfaced and handled under protocol — the safety architecture working as designed is an output, not a failure.
Teacher and family accounts of what held up in a real room and what did not.

Short-term outcomes · within a year

What the model predicts first

Students can name an emotion, a mistake, and a repair, using words shared with their teacher and with home.
Students distinguish I did something bad from I am bad when asked.
Students describe self-compassion as something other than self-esteem or excuse-making.
Students can say what forgiveness is not — not reconciliation, not condoning, not forgetting — and that it is never required where harm is ongoing.
Adults in the building use the same vocabulary outside the lesson.

Medium-term outcomes · the hypothesis

What the model predicts if the short-term outcomes hold

Students respond to failure with repair attempts rather than concealment more often than before.
Conflicts reach a repair conversation more often, and reach escalation less often.
Students report knowing at least one adult in the building they would tell.
Disclosures reach a counselor earlier.

These are predictions, not findings. Nothing on this list has been measured in a school using this curriculum. They are written down here so that a study can later show them to be wrong.

Why each link, and the research behind it

A logic model is only as good as its arrows. Each row below names the assumption, the published work it rests on, and — the part most logic models leave out — what would have to be true for the arrow to fail.

The link	The research it rests on	How it could fail
Teaching self-compassion explicitly → students treat their own failure with less harshness	Neff’s three-element model — self-kindness, common humanity, mindfulness — and the Mindful Self-Compassion work with Germer. The curriculum uses that structure directly and distinguishes it from self-esteem throughout.Neff (2003a, 2003b, 2011); Neff & Germer.	Neff’s findings are evidence about self-compassion. They are not evidence that this curriculum teaches it, or that a child taught it in a classroom acquires what the studies measured in adults.
Separating shame from guilt → students attempt repair instead of hiding	The distinction between “I am bad” and “I did something bad,” and the finding that shame tends to produce concealment and withdrawal where guilt tends to produce reparative action. This grounds the mistake-and-repair strand from K–2 upward.Brown; Tangney and colleagues.	The research describes a distinction between emotional states. Whether an eight-year-old who can articulate the distinction then acts on it is the untested step — and it is the step the whole strand depends on.
Teaching forgiveness as a skill with boundaries → students hold a workable definition rather than a coerced one	Forgiveness-therapy research defining what forgiveness is not — not reconciliation, not condoning, not forgetting — plus a staged process model. The curriculum’s most repeated safety line, forgiveness is never required in situations involving ongoing harm, comes straight out of these boundaries.Enright & Fitzgibbons; Worthington (REACH).	This literature is largely clinical and largely adult. Its translation into a whole-class lesson for children is the curriculum’s own construction, and it is the place where a badly delivered lesson could do harm rather than nothing.
The novel as spine → the concept is available before it is demanded	This is the curriculum’s own design claim and the one with the least external support behind it. It is stated here as an assumption rather than dressed as a finding.No published study is cited for this link.	It may simply add reading load. A study could find the lessons work as well without the novel, which would be a useful result and would change the product.
Disclosure protocols and counselor-in-room requirements → surfacing is safe rather than hazardous	Built to trauma-informed practice guidelines, with the AFSP Safe Messaging Guidelines applied by name to every highest-risk lesson. The 6–8 teacher brief states that the most sensitive chapter is statistically the most likely point of disclosure — and that this is the chapter doing its job.AFSP Safe Messaging Guidelines; trauma-informed practice guidance.	The protocol is only as good as the staffing behind it. A building that cannot put a counselor in the room and runs the lesson anyway has not implemented this model — it has implemented something more dangerous than doing nothing.
Shared language across classroom and home → the framework survives outside the lesson	The bilingual family letters and the common vocabulary are the mechanism. The design intent is stated on the site as influencing the hours outside the forty-minute lesson.Design assumption; no study cited.	Letters can go home and be unread. Nothing in the model yet measures whether the language actually crosses the threshold.

The effort underway to examine effects

The fourth evidence level asks for a logic model and an ongoing effort to study the intervention. That effort is Stage 1 of the validation roadmap, and the reason the whole curriculum is published open is to make it possible without a cost or a procurement cycle.

Stage 1 — pilot classrooms. Fidelity logs (what was delivered, deferred, and under what staffing), item-level instrument data under the district’s own data governance and in the district’s own spreadsheet, and teacher and family feedback.
Stage 2 — external psychometric review of the self-reflection instrument, by reviewers with no stake in the result. Findings published whichever way they point, and the instrument revised or retired accordingly.
Stage 3 — outcome evaluation, only after the instrument survives Stage 2. A validated measure is a precondition for a study, not a product of one.

No pilot sites, review partners or timelines are named, here or on the roadmap, because naming them before agreements exist would be the same overclaim in a different suit. If your district would consider being one, that is the conversation worth having.

What this page does not do

It is not evidence of effectiveness. A logic model is a statement of how something is supposed to work. It can be entirely coherent and entirely wrong.
It does not assign an evidence tier. That is your district’s determination or your state’s.
The self-reflection instrument is not validated. No independent psychometrician has evaluated it. Every score it produces is a flag for a conversation — never a diagnosis, never a screening result in the clinical sense, never a basis for a placement or eligibility decision.
It has not been externally reviewed. Nobody outside this program has checked the arrows above.
It is falsifiable on purpose. The medium-term outcomes are written as predictions, and the “how it could fail” column is there so a reviewer can find the weak link faster than a sales conversation would let them.

Found something that gets ahead of the evidence? This site commits to not using “evidence-based”, “proven” or “validated” before the evidence exists — not in marketing, not in a grant application, not in a sales conversation. A line on this page that overstates is a defect. Report it and it will be corrected.