Run the project retrospective

Where the evidence comes from

• Instructional improvement cycle: A teacher’s toolkit for collecting and analyzing data on instructional strategies
• Gold Standard: Project Design Elements | PBLWorks

Strong for reflective practice and improvement cycles; AI benefit unvalidated.

1. Reconstruct the dated record

   Privately remove identifying details. Use optional D04 intentions, D05 sequence and C01 progress records, or equivalent teacher documents. Check dates against originals. Retain conflicting accounts without identifying contributors.

   Prompt 1 · Chronology Copy

   CONTEXT
   Safe dated records with source IDs and intended outcome: [records].
   Teacher scope and private source/privacy check: [gate].
   REQUEST
   Organize the supplied record chronologically, separating plans from recorded events and undated material.
   QUALITY BAR
   Use only supplied dates, events and IDs. Never invent missing events, causes or learner details. Preserve conflicting and minority accounts. If the private check or dated records are missing, return HOLD with the missing input only. Mark undated or unsupported claims HOLD; never silently date them.
   FORMAT
   Table: ID | Date | Plan or record | Exact evidence | Limit. Then Missing inputs. Preserve source IDs and exact evidence wording.

2. Compare intentions and experience

   Check each chronology row against the original. Supply the approved version and explicit corrections. Select themes yourself after reading the AI proposals, including disagreement and evidence that challenges your first interpretation.

   Prompt 2 · Compare Copy

   CONTEXT
   Original safe records: [records].
   Exact approved step 1 output: [previous].
   Teacher corrections and approval of this input version: [decision].
   REQUEST
   Propose at most two themes comparing the intended outcome with the record, retaining dissent and unanswered questions.
   QUALITY BAR
   Stop with HOLD if approval, sources or essential evidence are missing. Do not infer learning from completion. Do not assign causes, rank experiences or erase minority accounts. Cite source IDs for every observation; label implications as proposals. Leave unsupported claims HOLD.
   FORMAT
   Table: Theme ID | Proposed theme | Evidence IDs | Dissent or contradiction | Question. Then Limits.

3. Write the bounded retrospective

   Choose themes and write interpretations outside AI. Distinguish design, implementation and contextual possibilities. Supply alternative explanations and uncertainty. Read the draft and edit conclusions yourself before the final comparison.

   Prompt 3 · Retrospective Copy

   CONTEXT
   Original safe records: [records].
   Exact approved step 2 output: [previous].
   Teacher selections, edits, causes with alternatives, surprises, failed assumptions and next questions: [decision].
   REQUEST
   Draft a one-page retrospective of no more than 350 words using only the supplied choices.
   QUALITY BAR
   Missing selections or interpretations mean HOLD; do not supply them. Keep dated events faithful, source-linked and distinct from teacher interpretations. Describe possible causes as unverified, including alternatives. Preserve dissent and uncertainty. Never claim that chronology establishes causation or learning.
   FORMAT
   Use headings: Intended outcome; What happened; Evidence and dissent; Teacher interpretations and alternatives; Surprises and failed assumptions; Next questions. Include source IDs throughout.

4. Audit your selected version

   Check the draft against originals and rewrite conclusions. Paste your selected version and verification notes. Resolve every flagged issue outside AI before retaining it; this comparison cannot authorize use.

   Prompt 4 · Audit Copy

   CONTEXT
   Original safe records: [records].
   Exact step 3 output reviewed by the teacher: [previous].
   Teacher-selected retrospective, edits and private verification notes: [decision].
   REQUEST
   Compare the selected retrospective with the records and supplied decisions. Flag changed dates, invented events, unsupported causal conclusions, missing dissent and unanswered questions removed.
   QUALITY BAR
   If the selected version or verification notes are absent, return HOLD. Do not rewrite, approve or release the result. Distinguish a textual match from verified fact. Retain uncertainty and all unresolved source gaps; the human must resolve them.
   FORMAT
   If the selected version or verification notes are absent, output only HOLD, the missing input and the required human check. Otherwise output, in order: a table with exactly these columns: Check | Selected text | Source or decision | Finding | Human action; a Remaining limits section; one conclusion sentence; and the final line Human release pending. Finding must be Text match, Mismatch or Unresolved. Text match means agreement with supplied text, never verified fact. Always retain the table and limits, including when every finding is Text match. In that case the conclusion must be exactly: The supplied-text comparison found no mismatch. Otherwise it must be exactly: The supplied-text comparison found mismatches or unresolved checks. Put provenance limits, including any synthetic-input status, in Remaining limits. Do not add a preamble or other sections.

Check before you use it

• Have you matched every event and date to an original record?
• Are plans, observations and teacher interpretations clearly separated?
• Have you preserved dissent and minority experiences without identifying anyone?
• Are possible causes qualified and alternative explanations retained?
• Are surprises, failed assumptions and next questions supported or explicitly uncertain?
• Have you resolved every HOLD privately and made the final retention decision yourself?