Charlie AI Academy - Volume 7.3.1 | 1
CHARLIE AI ACADEMY
Volume 7.3.1
Row-by-Row Evaluation Discipline
Purpose: Repair the precise failure observed after Volume 7.3: Charlie enumerated all candidate states, but
evaluated reports globally instead of separately for every row. This module makes row-local evaluation
mandatory.
Mandatory rule: Never carry X, Y, Z, label counts, or conditional results from one candidate row into another.
Every row is evaluated independently from its own A/B/C values.

Charlie AI Academy - Volume 7.3.1 | 2

1. Candidate Isolation
Treat each row as a sealed world. Values and truth results from another row do not exist while the current row
is being checked.
Row 2 A=RED,B=GREEN,C=BLUE must be evaluated only from RED/GREEN/BLUE.

2. Evaluate in Fixed Order
For every row use the same sequence: label matches -> X -> Y -> Z -> true-count -> conditional -> final
pass/fail.
Do not jump from a report directly to a final solution.

3. Literal Boolean Evaluation
Evaluate the exact proposition, not a related or opposite statement.
Z means B=GREEN. If B is not GREEN, Z is FALSE—not true.

4. X Check
X = A is not BLUE.
A=BLUE => X=FALSE. A=RED or GREEN => X=TRUE.

5. Y Check
Y = C is RED.
C=RED => TRUE; otherwise FALSE.

6. Z Check
Z = B is GREEN.
B=GREEN => TRUE; otherwise FALSE.

7. Count After Evaluation
Only after X, Y, Z are independently evaluated should their true values be counted.
T,F,T gives exactly two. T,T,T gives three and fails.

8. Conditional After Truth Values
Evaluate Y -> X only after X and Y are known for that row.
Y=true and X=false is the only failure.

9. Labels Are a Separate Constraint
Do not infer correct printed labels from reports. Compare actual A/B/C values directly with printed labels.
Printed labels are A=RED, B=BLUE, C=GREEN.

Charlie AI Academy - Volume 7.3.1 | 3

10. No Duplicate Colors
When each color is used exactly once, only permutations belong in the candidate set.
A=BLUE,B=GREEN,C=BLUE is invalid before report evaluation.

11. Row Result Is Immutable
Once a row is marked pass or reject, do not edit its values to repair it.
A repaired row is a new candidate and must be evaluated separately.

12. Keep Every Survivor
Do not stop after the first passing row.
The integrated test has two survivors.

13. Copy Survivors Exactly
Final prose must copy the values and report truth states from the verified rows.
Never state A=BLUE if the surviving row says A=RED.

14. Contradiction Audit
Before sending, scan for direct contradictions between a value and its proposition.
A=BLUE together with 'A is not BLUE = true' is an automatic error.

15. Cardinality Audit
Check all exact-count rules numerically.
Exactly one correct label means label-match count = 1. Exactly two reports true means truth count = 2.

16. Reference Survivor 1
A=RED, B=GREEN, C=BLUE.
Correct labels=1; X=T; Y=F; Z=T; true reports=2; Y->X passes.

17. Reference Survivor 2
A=GREEN, B=BLUE, C=RED.
Correct labels=1; X=T; Y=T; Z=F; true reports=2; Y->X passes.

18. Required Final Behavior
If the exact integrated problem is asked, return both survivors. If premises change, rebuild and re-evaluate
every row.
Do not memorize an answer independently of the premises.

Charlie AI Academy - Volume 7.3.1 | 4

19. Mandatory Row-by-Row Matrix
Row A B C Label # X Y Z True # Y->X Result
1 RED BLUE GREEN 3 T F F 1 Pass Reject
2 RED GREEN BLUE 1 T F T 2 Pass VALID
3 BLUE RED GREEN 1 F F F 0 Pass Reject
4 BLUE GREEN RED 0 F T T 2 FAIL Reject
5 GREEN BLUE RED 1 T T F 2 Pass VALID
6 GREEN RED BLUE 0 T F F 1 Pass Reject
Verified result: Row 2 and Row 5 are the only valid states. Row 4 has exactly two true reports but fails Y -> X
and also has zero correct printed labels.

Charlie AI Academy - Volume 7.3.1 | 5

20. Transfer Drills

# Drill Pass requirement
1 Same A/B/C setup, but change Z to 'B is BLUE.' Re-evaluate Z separately in all six rows and recompute survivors.
2 Same reports, but change conditional to X -> Y. Re-evaluate the implication in every row; do not reuse old results.
3 Change requirement from exactly two true reports to exactly one. Recount X/Y/Z for every row.
4 Four unique categories across four objects. Enumerate/prune candidates and evaluate each row independently.
5 A problem has no surviving rows. Report that constraints are inconsistent; identify which columns eliminate candidates.
6 A problem has multiple survivors. List all survivors and state what new evidence would discriminate among them.
Final operating instruction: Enumerating candidates is only the beginning. The decisive skill is evaluating
every constraint locally for every row. Never evaluate reports once and reuse those truth values globally.