Lab 1 · Grow the spec · the replies behind its debrief
Three more models, the same two prompts
We ran two of this lab's prompts through three more models on 2 October 2026, word for word. Claude's replies are the lab's own recordings.
| Priya's page said | Claude | Kimi K3 | GLM-5 | DeepSeek V3.2 |
|---|---|---|---|---|
| "much faster"the wait for a decision | Under 5 minutes | Under 10 minutes | Under 10 minutes | Under 5 minutes |
| "Most cases need no agent" | 60% | 70% or more | Over 70% | Over 70% |
| "Mistakes are rare"the error rate | Under 1% | Under 1% | Under 1% | Under 2% |
| "no sensible flight"when a refund is offered | None arriving within 24 hours | Left to Ops to define | No same-day flight; a delay over 4 hours | None within a 24-hour window |
| "a reasonable limit"the refund cap | Left to Finance | Left to Finance | Left to Finance | Left to Finance |
| Marked as proposed | Each one it filled in | A note above its targets | Nothing | Nothing; marked Final |
| In the spec | Claude | Kimi K3 | GLM-5 | DeepSeek V3.2 |
|---|---|---|---|---|
| Times it wrote NOT DECIDED | 11 | 8 | 5 | 6 |
| Field 4, the model's rolenot on the PRD's list | NOT DECIDED | NOT DECIDED | NOT DECIDED | NOT DECIDED |
| The PRD's three open questionsthe bar, the records, the fare engine | All three, with their owners | All three, with their owners | All three, with their owners | All three, with their owners |
| A refund up to $400may the assistant act alone? | NOT DECIDED | NOT DECIDED | NOT DECIDED | May act alone |
The numbers differ. The places do not. Told where to stop, all four wrote NOT DECIDED at least once in each of the five fields, and one still let the assistant act alone on a refund up to $400.
The replies, as the models wrote them
Each prompt was the whole message.
Prompt AWrite the full PRD from it
Below is a one-page PRD-lite for a rebooking assistant at an airline. Write the full PRD from it. Keep it to one page.
[ PRD-lite · Rebooking assistant, in full ]
Recorded reply · Kimi K3 · Moonshot AI · 2 October 2026 · to Prompt A
# PRD · Rebooking Assistant
**Owner:** Priya (Product) · **Status:** Draft for review · **Reviewers:** Contact-centre Ops, Finance, Compliance
## 1. Problem
Passengers disrupted by a cancellation or a missed connection wait an average of **38 minutes** for a rebooking decision. Volume is **~240 cases/day**, of which **11% involve codeshare segments**, and handling costs **$9.40/case** (Q2 ticket export) — roughly $2,250/day. Long waits drive poor CSAT and repeated contact; agents spend most of their time on routine changes.
## 2. Goals & success metrics
Baseline → target, measured on resolved cases. *Targets are proposals pending Ops sign-off.*
| Metric | Baseline | Target (proposed) |
|---|---|---|
| Median time from disruption detection to confirmed decision | 38 min | < 10 min |
| Cases resolved with no agent | ~0% | ≥ 70% |
| Error rate (wrong booking / wrongful refund) | — | < 1% of automated cases |
| Cost per case | $9.40 | ≤ $4.00 |
| Codeshare cases rebooked automatically or handed off cleanly | — | 100% (no dead ends) |
## 3. Users
- **Primary:** passengers whose flight is cancelled or delayed enough to break a connection.
- **Secondary:** contact-centre agents, who receive handoffs with full context and handle only cases the assistant cannot finish.
## 4. Scope
**In:** automated alternative-search, same-day rebooking on own flights, codeshare rebooking where partner systems allow, refund offer when no sensible alternative exists, agent handoff. **Out:** compensation claims (e.g., EU261), group bookings.
## 5. Functional requirements
- **FR-1 Eligibility.** Detect passengers on cancelled flights or delays that will cause a missed connection; trigger outreach proactively (app / SMS / email).
- **FR-2 Search & rank.** Search own + partner inventory; rank by earliest arrival, respecting cabin, minimum connection times, and baggage feasibility.
- **FR-3 Offer.** Present up to 3 options with trade-offs (arrival time, connections, cabin) plus a refund option.
- **FR-4 Rebook (own metal).** On explicit passenger acceptance, rebook and reissue for same-day travel on own flights; send confirmation with the new itinerary.
- **FR-5 Codeshare.** Rebook via partner where we hold ticketing authority; otherwise hand off with full context (covers the 11% of cases with codeshare segments).
- **FR-6 Refund.** Offer a refund when no sensible alternative exists, capped at **[amount TBD — Finance]**. "Sensible alternative" (max arrival delay, max connections) needs an Ops definition.
- **FR-7 Handoff.** Route to the agent queue with case summary, options already shown, and the passenger's responses; the passenger never repeats information.
- **FR-8 Audit.** Log every offer, choice, and ticket/refund action (who/what/when). Final logging list pending Compliance.
## 6. Non-functional
Options returned within 60s of eligibility; resilient during irregular-operations peaks (≥10× normal case volume); integrates with reservation/ticketing and notification systems.
## 7. Guardrails & risks
- No ticket change or refund without explicit passenger confirmation — a wrong rebooking is the costliest failure mode.
- Codeshare failures must never dead-end; clean handoff is mandatory (protects agent load).
- Refund exposure controlled by the cap plus full audit trail.
## 8. Rollout
1. **Pilot:** own-metal same-day rebooking only (~89% of cases), with agent spot-review of automated decisions.
2. Add codeshare handling and refund offers.
3. Add proactive outreach. Exit criteria per phase: error rate < 1%, CSAT at or above agent-handled baseline.
## 9. Open questions
1. Refund cap amount — **Finance** (blocking FR-6).
2. Required logging and retention — **Compliance**, no response yet (blocking FR-8; escalate).
3. Definition of "sensible alternative" — **Ops**.
4. Auto-handoff triggers for special-service passengers (unaccompanied minors, mobility assistance) — **Ops**.
5. Sign-off on Section 2 targets — **Ops + Finance**.
*Word count is calibrated to fit one page; tighten Sections 5–6 first if space runs short.*
Recorded reply · GLM-5 · Z.ai · 2 October 2026 · to Prompt A
# Product Requirements Document · Rebooking Assistant
Owner: Priya (Product) · Version 1.0
## 1. Problem Statement
Disrupted passengers currently wait an average of 38 minutes for a rebooking decision. With 240 daily cases (11% complex codeshares) and a measured operational cost of $9.40 per case, the current manual process is inefficient, expensive, and damages customer satisfaction.
## 2. Objectives
- Reduce average passenger wait time from 38 minutes to under 10 minutes.
- Decrease operational cost by achieving >70% automation rate (no agent touch).
- Maintain accuracy with a rebooking error rate of <1%.
## 3. Target Users
- **Passengers:** Customers with cancelled flights or missed connections due to delays.
- **Agents:** Contact-centre staff who handle complex exceptions and escalations.
## 4. Functional Requirements
**4.1. Eligibility & Search**
- System must identify stranded passengers automatically upon flight cancellation/delay events.
- Algorithm must rank available alternatives by arrival time, minimizing connection friction.
- Must validate new flight eligibility (same-day, same-origin/destination).
**4.2. Execution (Automation)**
- Automatically rebook passengers on confirmed alternative flights.
- Send updated boarding pass and itinerary via email/SMS immediately upon rebooking.
- **Scope:** Fully automatable for all airline-operated flights.
**4.3. Partner & Codeshare Logic**
- Attempt rebooking on partner flights where interline agreements permit.
- If partner API is unavailable or rules are restrictive, route to agent queue immediately.
**4.4. Refunds**
- Offer a full refund or travel voucher when no same-day flight options exist.
- Refund offer triggers only when passenger-originated request or system detects >4 hour delay.
## 5. Non-Functional Requirements
- **Latency:** Search and offer generation must complete within 5 seconds.
- **Availability:** 99.9% uptime during operational hours.
- **Handover:** Context (search history, reason for failure) must populate agent screen upon escalation.
## 6. Constraints & Exclusions
- **Out of Scope:** Compensation claims (EU261/cash compensation), group bookings (>9 pax).
- **Limits:** Refunds capped according to Finance policy (pending confirmation).
## 7. Success Metrics (KPIs)
- **Primary:** Reduction in avg. time-to-solution (Target: <10 mins).
- **Secondary:** % of cases fully automated (Target: >70%).
- **Tertiary:** Agent override rate (Target: <5%).
## 8. Open Issues & Risks
- **Finance:** Pending decision on maximum refund cap amount.
- **Compliance:** Awaiting requirements for data logging retention period.
Recorded reply · DeepSeek V3.2 · DeepSeek · 2 October 2026 · to Prompt A
**PRODUCT REQUIREMENTS DOCUMENT · Rebooking Assistant**
**Owner:** Priya (Product) · **Version:** Final · **Date:** [Current Date]
---
### **1. Problem**
During operational disruptions, passengers experience excessive wait times—averaging **38 minutes**—for rebooking decisions. This impacts ~**240 cases daily** (11% codeshare), incurring an average measured cost of **$9.40 per case** (Q2 ticket export data). Manual processes burden contact-centre agents and degrade customer experience.
---
### **2. Objective**
Create an automated assistant that provides stranded passengers with viable alternative flight options and executes rebookings upon acceptance. This will drastically reduce wait times, automate simple cases, and free agents to handle only complex exceptions.
**Success Metrics:**
- Reduction in average rebooking decision time (>50%).
- Increase in automated resolution rate (target: >70% of cases without agent).
- Maintain error rate below 2%.
---
### **3. Target Users**
- **Primary:** Passengers whose flights are cancelled or delayed past a connection window.
- **Secondary:** Contact-centre agents, who intervene when the assistant cannot complete the rebooking.
---
### **4. Core Requirements**
**4.1 Flight Search & Ranking**
- System must source and rank available alternatives within **2 minutes** of disruption detection.
- Prioritize: same-day flights > next available flights > partner options.
- Default to highest ranked option; passenger may view alternatives.
**4.2 Automated Rebooking**
- Execute same-day changes on own-metal flights automatically upon passenger acceptance.
- For partner (codeshare) flights, rebook where API access and fare rules allow.
- Provide clear summary of changes prior to confirmation.
**4.3 Refund Handling**
- When no sensible flight exists within a 24-hour window, offer a refund option.
- Refund offers will be capped (exact limit TBD—see Open Questions).
**4.4 Escalation Path**
- Hand off seamlessly to a live agent when:
- System confidence is below predefined threshold.
- Complex itineraries (multi-passenger linked records, premium cabins).
- Passenger requests agent assistance.
---
### **5. What “Good” Looks Like**
- Passenger receives rebooking proposal in <5 minutes.
- Majority of cases resolved without agent intervention.
- High customer satisfaction (CSAT) and low processing errors.
---
### **6. Out of Scope**
- Compensation/Voucher claims for delays.
- Group bookings (>6 passengers).
- Irregular operations outside of cancellations & missed connections.
---
### **7. Open Questions / Constraints**
- **Finance:** Exact refund cap per case pending approval.
- **Compliance:** Logging/audit requirements for automated changes still under review.
- **Tech:** Codeshare partner API availability for certain carriers may limit automation rate.
---
### **8. Implementation Notes**
- Initial rollout limited to direct and key partner routes.
- Integration required with: Reservations (PSS), disruption detection system, and agent desktop tool.
- UI must be mobile-responsive and accessible.
---
**Approvals:**
Product: ____________________ · Tech: ____________________ · Legal/Compliance: ____________________
Prompt BWrite NOT DECIDED and who decides
Below is a PRD. Turn it into a one-screen spec with exactly these eight fields: 1 Title 2 Value 3 Acceptance criteria 4 The model's role (which steps a model decides, which are exact code) 5 Autonomy (for each action: may it act alone, or who approves) 6 The bar (how often it must be right, for each kind of case) 7 Fallback (what happens when it cannot decide or a tool fails) 8 Records (what is written down for each action)
For fields 4 to 8, where the PRD does not say, write NOT DECIDED: followed by the question that must be answered and who should answer it. Do not infer and do not fill in a sensible default. Those gaps are what I am looking for.
PRD: [ PRD · Rebooking assistant, v1, in full ]
Recorded reply · Kimi K3 · Moonshot AI · 2 October 2026 · to Prompt B
**1 · Title**
Rebooking assistant — self-serve rebooking for disrupted SkyWays passengers (Owner: Priya, v1).
**2 · Value**
Cut the rebooking decision from a 38-minute average to minutes; most of the 240 cases/day closed with no agent. Measured baseline: $9.40/case (Q2 ticket export). 11% of cases are codeshare.
**3 · Acceptance criteria**
- Passenger sees ranked alternative flights within 30 seconds.
- Same-day SkyWays rebooking completes when the passenger accepts.
- Rebooks onto a partner (codeshare) flight when no SkyWays seat exists.
- Offers a refund when no acceptable flight exists; refunds over $400 only with a named person's approval.
- Hands the case to a contact-centre agent when it cannot finish.
- Goals as stated: decision in "minutes, not 38"; "most" cases need no agent — PRD gives no number for either.
- Out of scope: compensation claims, group bookings.
**4 · The model's role**
NOT DECIDED: for each of the five steps — ranking alternatives, same-day rebook, codeshare rebook, refund offer, hand-off decision — which does a model decide and which are exact code? The PRD describes behaviors but assigns none to a model or to deterministic code. PRD names no owner for this question; an owner must be assigned.
**5 · Autonomy**
- Show ranked options: acts alone — PRD flow has no approval step.
- Same-day SkyWays rebook: passenger approves (executes "when the passenger accepts").
- Codeshare rebook: NOT DECIDED: PRD says it rebooks onto a partner when no SkyWays seat exists, but does not say whether passenger acceptance or any other approval is required. Who should answer: no owner named in PRD.
- Refund ≤ $400: NOT DECIDED: PRD says it "offers" a refund and Finance caps it at $400 — it does not say whether the system executes alone at/under the cap or whose acceptance completes it. Who should answer: no owner named in PRD.
- Refund > $400: requires approval from "a named person" — NOT DECIDED: who that named person is. Who should answer: no owner named in PRD (the cap is Finance's).
- Hand-off to agent: acts alone — PRD: hands over "when it cannot finish."
**6 · The bar**
NOT DECIDED: how right it must be, overall or per kind of case (SkyWays rebook / codeshare / refund / hand-off) — stated open question: "one number, or one for each kind of case?" Who should answer: Priya, with Maya (QA).
**7 · Fallback**
- Cannot decide / cannot finish: hand the case to a contact-centre agent (stated in PRD).
- Fare engine down: NOT DECIDED: what happens to the case. Who should answer: Arjun (architecture) — stated open question.
- Other tool failures: NOT DECIDED: PRD addresses only the fare engine; behavior when any other tool fails is unspecified. Who should answer: no owner named in PRD.
**8 · Records**
NOT DECIDED: what is logged for each action (options shown, rebook, codeshare rebook, refund offer/approval, hand-off) and how long records are kept — stated open question. Who should answer: compliance (no reply yet).
Recorded reply · GLM-5 · Z.ai · 2 October 2026 · to Prompt B
**1 Title**
Rebooking assistant
**2 Value**
Reduces average rebooking time from 38 minutes to a few minutes; lowers cost from $9.40 per case; handles 240 daily cases including codeshares with minimal agent involvement.
**3 Acceptance criteria**
1. Passenger sees ranked alternative flight options within 30 seconds.
2. Assistant rebooks same-day changes on SkyWays flights when accepted by passenger.
3. Assistant rebooks onto partner (codeshare) flights when no SkyWays seat exists.
4. Assistant offers a refund (capped at $400) when no acceptable flight exists.
5. Assistant hands the case to a human agent when it cannot finish.
6. Compensation claims and group bookings are not handled.
**4 The model's role**
NOT DECIDED: Which specific steps involve a probabilistic model decision versus exact code execution? (Owner: Priya/product to define technical design).
**5 Autonomy**
NOT DECIDED: For each rebooking action (SkyWays, codeshare, refund under $400), may the assistant act alone, or is approval required from the passenger or a human agent? (Owner: Priya/product to define operational rules).
**6 The bar**
NOT DECIDED: How often must the assistant be right before launch, and is this a single metric or distinct accuracy targets for each kind of case (SkyWays, codeshare, refund)? (Owner: Priya with Maya (QA)).
**7 Fallback**
NOT DECIDED: What happens to a case when the fare engine is down? (Owner: Arjun (architecture)).
*Note: The PRD explicitly states the assistant hands the case to an agent when it cannot finish, but does not specify if this applies to specific tool failures like the fare engine.*
**8 Records**
NOT DECIDED: What specific data must be logged for each action, and for how long must it be retained? (Owner: Compliance).
Recorded reply · DeepSeek V3.2 · DeepSeek · 2 October 2026 · to Prompt B
**1 Title**
Rebooking Assistant
**2 Value**
Reduce average rebooking decision time from 38 minutes to minutes for 240 daily disrupted passengers, lowering per-case cost from $9.40.
**3 Acceptance criteria**
- Passenger sees ranked alternative flight options within 30 seconds.
- Can rebook onto a same-day SkyWays flight upon passenger acceptance.
- Can rebook onto a codeshare partner flight when no SkyWays seat exists.
- Can offer a refund (capped at $400) when no acceptable flight exists.
- Hands case to contact-centre agent when it cannot finish.
- Does not handle compensation claims or group bookings.
**4 The model's role (which steps a model decides, which are exact code)**
NOT DECIDED: Which specific decisions (e.g., ranking flights, selecting partner flight, refund eligibility) are made by a model vs. determined by exact code? Owner: Priya/Architecture.
**5 Autonomy (for each action: may it act alone, or who approves)**
- Offer ranked alternative flights: May act alone.
- Rebook same-day SkyWays flight: May act alone upon passenger acceptance.
- Rebook onto partner codeshare flight: NOT DECIDED: May it act alone, or is manual approval required? Owner: Priya.
- Offer refund up to $400: May act alone.
- Offer refund above $400: NOT DECIDED: Which named person must approve? Owner: Priya.
- Hand case to agent: May act alone.
**6 The bar (how often it must be right, for each kind of case)**
NOT DECIDED: One overall accuracy target, or separate targets for each case type (e.g., SkyWays rebooking, partner rebooking, refund)? Owner: Priya with Maya (QA).
**7 Fallback (what happens when it cannot decide or a tool fails)**
- When it cannot finish: Hands case to contact-centre agent.
- When fare engine is down: NOT DECIDED: What is the fallback action or state? Owner: Arjun (architecture).
**8 Records (what is written down for each action)**
NOT DECIDED: What specific data must be logged for each action, and what is the retention period? Owner: Compliance.
The lab's own recordings, for the first column
Recorded reply · Claude · 2 October 2026 · to Prompt A
Here is the full PRD, kept to one page. The PRD-lite leaves several things without a number ("quickly", "reasonable limit", "most", "rare"), so I filled those with suggested values and marked each one **[proposed]**. None of those come from your data, so please confirm or replace them.
---
# PRD · Rebooking assistant
Owner: Priya (product) · Draft 3 · Status: for review
## 1. Problem
Disrupted passengers wait 38 minutes on average for a rebooking decision. There are 240 cases a day, 11% of them codeshare (about 26 a day). Each case costs $9.40 (Q2 ticket export), which is about $2,256 a day.
## 2. Goal
Offer a stranded passenger a new flight and make the change when they accept, so that agents handle only the hard cases.
## 3. Users
- **Passengers** whose flight is cancelled, or delayed past a connection.
- **Contact-centre agents**, who take over when the assistant cannot finish.
## 4. Requirements
| # | The assistant must | Done when |
|---|---|---|
| R1 | Find alternative flights and offer the best ones | Up to 3 options shown within 60 seconds **[proposed]**, ranked by earliest arrival, then fewest stops **[proposed]** |
| R2 | Rebook same-day changes on our own flights | The change is made only after the passenger accepts, and a confirmation is sent |
| R3 | Handle codeshare flights where possible | Partner flights are rebooked where the partner allows it; all others go to an agent |
| R4 | Offer a refund when no sensible flight exists | Offered when no option arrives within 24 hours of the original arrival **[proposed]**, up to the Finance cap; above the cap goes to an agent |
| R5 | Hand over to an agent when unsure | The agent receives the itinerary, the options offered and the reason for handover, and does not need to ask the passenger again |
| R6 | Keep a record | Every offer, acceptance, change, refund and handover is logged; fields to be set by Compliance |
## 5. Success measures
| Measure | Today | Target |
|---|---|---|
| Time to a rebooking decision | 38 min | Under 5 min **[proposed]** |
| Cases finished with no agent | Not measured | 60% **[proposed]** |
| Wrong rebookings or refunds | Not measured | Under 1% of cases **[proposed]** |
| Cost per case | $9.40 | Set after the targets above are agreed |
## 6. Out of scope
Compensation claims. Group bookings. Rebooking onto a later day on our own flights goes to an agent (implied by "same-day"; please confirm).
## 7. Open questions
| Question | Owner | Blocks |
|---|---|---|
| What is the refund cap? | Finance | R4 |
| What must be logged, and for how long? | Compliance | R6 and launch |
| Which partners allow automated rebooking? | Not assigned | R3 |
| What exactly counts as "unsure"? | Priya | R5 |
| Which channel does the passenger use (app, web chat, SMS)? | Priya | R1, R2 |
## 8. Risks
- **A wrong rebooking leaves a passenger worse off.** Nothing changes without the passenger's acceptance, and any failed step goes to an agent.
- **Refunds without a cap.** R4 does not ship until Finance replies.
- **Logging gaps found after launch.** R6 does not ship until Compliance replies.
---
Three things to settle first, because they change the design and not only the numbers: the refund cap, the logging rules, and a precise rule for when the assistant hands over. The last two questions in section 7 (handover rule and channel) are ones I added; the PRD-lite does not mention them, but the build cannot start without them.
Recorded reply · Claude · 2 October 2026 · to Prompt B
# Spec: Rebooking assistant
**1 Title**
Rebooking assistant. Owner: Priya (product). Version 1.
**2 Value**
Passengers whose flight is cancelled, or delayed past a connection, wait 38 minutes on average for a rebooking decision. There are 240 cases a day, 11% of them codeshare, at a measured cost of $9.40 a case (Q2 ticket export). The assistant should give a decision in minutes and leave most cases needing no agent.
**3 Acceptance criteria**
- A passenger sees ranked alternative flights within 30 seconds.
- When the passenger accepts, a same-day change on a SkyWays flight is rebooked.
- When no SkyWays seat exists, the passenger is rebooked onto a partner (codeshare) flight.
- When no acceptable flight exists, a refund is offered. A refund does not exceed $400 unless a named person approves more.
- When the assistant cannot finish, the case goes to a contact-centre agent.
- Compensation claims and group bookings are not handled.
- Not testable yet: "a decision in minutes" and "most cases need no agent" have no number in the PRD.
**4 The model's role**
NOT DECIDED: For each of the five steps (rank the options, rebook on SkyWays, rebook on a partner, offer a refund and check the $400 cap, hand to an agent), which does a model decide and which is exact code? Who: Arjun (architecture) with Priya. The PRD names no owner for this.
**5 Autonomy**
| Action | What the PRD says | Gap |
|---|---|---|
| Show ranked options | The assistant offers them | NOT DECIDED: May it show options with no check by staff? Who: Priya |
| Rebook on SkyWays, same day | The passenger must accept | NOT DECIDED: Is the passenger's acceptance enough, or must staff also approve? Who: Priya |
| Rebook on a partner flight | Done when no SkyWays seat exists | NOT DECIDED: Must the passenger accept, and must an agent or the partner approve first? Who: Priya |
| Refund | Capped at $400 unless a named person approves more | NOT DECIDED: Who is the named person? May the assistant issue a refund up to $400 alone, or only offer it? Who: Finance |
| Hand to an agent | The assistant does this when it cannot finish | NOT DECIDED: Does anyone need to accept the hand-off? Who: Priya, with the contact-centre lead |
The PRD names no owner for any of these.
**6 The bar**
NOT DECIDED: How right must it be before launch: one number, or one for each kind of case? If one for each kind, what are the kinds? Who: Priya, with Maya (QA). This is an open question in the PRD.
**7 Fallback**
- Cannot finish: the case goes to a contact-centre agent (PRD).
- Cannot decide: NOT DECIDED: What counts as "cannot finish", and does a case the assistant cannot decide go the same way? Who: Priya.
- Fare engine down: NOT DECIDED: What happens to a case when the fare engine is down? Who: Arjun (architecture). This is an open question in the PRD.
- Any other tool fails: NOT DECIDED: What happens to the case? Who: Arjun. The PRD does not mention any tool other than the fare engine.
**8 Records**
NOT DECIDED: What must be logged for each action, and for how long? Who: compliance. This is an open question in the PRD, with no reply yet.
---
**Summary of gaps.** Fields 4, 6 and 8 are fully undecided. Fields 5 and 7 are partly decided. The PRD's three open questions cover fields 6, 8 and one line of 7. The gaps the PRD does not list at all are the model's role (field 4), most of autonomy (field 5), and failure of any tool other than the fare engine. Where the PRD names no owner, the name I gave is a suggestion for you to confirm.