Context and Token Budgets for School Teacher Training Credits: Find Supporting Evidence

Teacher Credits: review context & tokens for evidence lookup. Check source versions, input mappings and the values received by downstream nodes.

What this category's data looks like

This page addresses the primary and secondary school teacher training credit management scenario, and the materials come from training notices, credit recognition forms, and attendance rosters. The training notice provides the training program name, start and end dates, the conversion basis between class hours and credits, and the registration and sign-in requirements; the credit recognition form records the teacher's name, ID document number, training name, recognized credits, reviewer signature, and recognition date; the attendance roster records the sign-in, sign-out, absence, and make-up training receipt for each session. The three types of materials are updated by training batch: the notice is published first, the roster is supplemented on a rolling basis with each session, and the recognition form is filled in centrally after the training ends and signed level by level. The field characteristics are concentrated in name wording, date format, quantity basis, and signature position: the same training may use its full name in the notice but an abbreviation in the roster; the conversion basis between class hours and credits is inconsistent across different materials; and the recognition date and completion date may span months. When the materials enter this feature, the notice serves as the basis source, the roster serves as the working draft for factual verification, and the recognition form serves as the conclusion carrier; together, the three support basis entries that can be traced back to the original text.

Configuration and implementation checks

This part is executed by the application settings (response limit, etc.), which output the length of a single answer and the range of context that can be accommodated. The "basis entries that can be traced back to the original text" are then completed by the model. The response limit is controlled by a slider for the length of a single answer; if set too low, the answer will be truncated, and if set too high, consumption will increase.

ObjectCheckVerification action
Training noticeThe program name, start and end dates, conversion basis between class hours and credits, and registration and sign-in requirements can each be matched item by item to the original clauses; the name and dates of the same training are consistent across the two materialsIn the application settings, check the response limit switch and slider position, ask a basis question that requires a long answer, see whether the ending is complete, and record maxResponse and maxContext
Credit recognition formThe teacher's name, ID document number, training name, recognized credits, reviewer signature, and recognition date are complete; entries missing a signature or receipt are marked separatelyCompare the name, quantity, and date in the recognition form and the attendance roster item by item, record inconsistent items, record the number of quoted entries separately by array length, and check the tokens of the quoted text with quoteMaxToken
Attendance rosterThe sign-in, sign-out, absence, and make-up training receipt for each session can be traced back to specific rows; when the name, quantity, and date of the same record do not match across the two materials, the signed receipt prevailsSpot-check cross-material basis questions, see whether the answer gives a traceable source, and compare the usage changes between two runs

Three common errors

  • Symptom: the name and date of the same training do not match between the notice and the roster. Cause: the notice uses the full name while the roster uses an abbreviation, and the update batches are not synchronized, causing the basis entries to be traced back to the wrong place.
  • Symptom: the recognition form is missing the reviewer signature or the make-up training receipt. Cause: during centralized filling, the level-by-level signatures were not completed, and no completeness check was performed when the materials entered the feature, so the basis entries cannot be closed.
  • Symptom: the answer breaks off in the middle of a sentence or the length exceeds what is needed. Cause: the response limit slider position does not match the length required by the question, and the truncation behavior and run consumption need to be checked against the actual ending of the answer.

How to verify the setup

Feature acceptance: ask a cross-material basis question that requires a long answer, see whether the ending of the answer is complete, and check whether maxResponse and maxContext match the answer length required by the page.

Business test: ask a cross-material basis question, see whether the answer gives a traceable source, and confirm that the basis comes from the relevant materials, that the version is the current one, and that the entries match the original text.

  • In the application settings, check the response limit switch and slider position, ask a question that requires a long answer and look at the ending, and compare the usage changes between two runs.
  • Compare the name, quantity, date, and signature in the training notice, credit recognition form, and attendance roster item by item, mark entries missing a signature or receipt, and record the number of quoted entries and quoteMaxToken.