## AI Planning Review Sheet (practice, fictional data)

### 1. Condition classification
- **Required**: R1 include A (G1) / R2 per-role caps: dev 12, design 4, QA 2, no substitution across roles (G2) / R3 budget 3,600,000 KRW = 18 person-days x 200,000 (G3) / R4 every feature must complete QA; no skipping or substituting (S4) / R5 B starts only after A is done (F-B)
- **Optional requests**: O1 B (multi-item edit), O2 C (first-use guide), O3 D (enterprise SSO)
- **Unverified evidence**: U1 S1's 32 inquiries still include repeat inquiries from the same users, so the user count is unknown / U2 S2's 20 of 50 are sessions, not people, and the cause of drop-off is unknown / U3 S3 is one prospect asking about SSO; no contract, amount or deadline demand / U4 D's external vendor schedule is unknown / U5 C is a pre-validation estimate

### 2. Calculation (person-days)
| Plan | Dev | Design | QA | Total | Cost (KRW) | Pass? |
|---|---|---|---|---|---|---|
| A+B | 9 | 2 | 2 | 13 | 2,600,000 | Pass |
| A+C | 7 | 3 | 2 | 12 | 2,400,000 | Pass |
| A+B+C | 12 | 4 | 3 | 19 | 3,800,000 | Fail: QA over by 1; total over by 1 day and budget over by 200,000 |
| A+D | 12 | 2 | 3 | 17 | 3,400,000 | Fail: QA over by 1 (budget OK); vendor schedule unknown |

A+B and A+C use all 2 QA days, leaving no QA slack. A+B+C and A+D would only fit by cutting QA, which violates R4.

### 3. Two feasible plans compared
- **A+B**: leftover dev 3, design 2, budget 1,000,000. S1 is a demand signal tied to B, but because of U1 it is not converted into a reduction figure. B must wait for A, so the schedule is sequential and any A delay passes straight to B.
- **A+C**: leftover dev 5, design 1, budget 1,200,000. C is independent of A and can run in parallel. But C is a U5 estimate, and due to U2 nothing can be said about improving setup completion.

### 4. Recommended draft: A+B
Finish A first, then do B. It keeps total and role caps, adds a feature with a clear dependency, and is backed by a real support-team signal. The 32 inquiries are a reference signal, not an effect figure, and the 20 of 50 incomplete setups are not used as evidence for C. Revisit C once its estimate is validated. Exclude D for now until S3 and the vendor schedule are clarified.

### 5. Predecessors and successors (not fixed dates)
A dev/design -> A QA -> (after A completes) B dev/design -> B QA -> release decision. C could run in parallel with A but is excluded from this recommendation. Any day-by-day plan is an assumption pending schedule confirmation; the two-week target is a requirement, not a commitment.

### 6. Questions to confirm
1. Converted to actual headcount, leave and calendar days, can the person-day caps be met in two weeks?
2. Can the 2 QA days be available when A finishes and again when B finishes?
3. After removing repeats, how many distinct users are behind the 32 inquiries?
4. What causes the S2 drop-off, and what is the per-person figure?
5. Can C's dev and design estimates be validated?
6. Does the S3 prospect require a deadline or contract terms, and when will D's vendor schedule be known?