Qianlin Brain · Final Demo Recording Script (Screen-by-Screen Timed Version · Aligned with Environment)

Organized based on the verbatim transcript of the "Qianlin Brain · Final Demo Recording Script (Full Version)" you provided, and aligned with the actual test results from the local 3006 environment.

Total duration target: Approximately 2 minutes (0:00–2:00). All three AI dialogue scenarios use L1 deterministic routing (auto_execute=true), not relying on LLM inference, and have been tested for stable reproducibility.

0. Recording Method & Environment (Confirmed)

• Recording: You will manually record locally (OBS / Win+G), we provide the timed script.
• Environment: Local https://ylxt.chat. Confirmed that the service on this port is the same codebase as BossAgents / ylxt.chat. Local recording will not touch production data.
• Browser: Dark theme; address bar shows https://ylxt.chat (corresponding to the "record locally" decision). If needed to present the ylxt.chat brand externally, you can overlay the address bar with a graphic in post-editing, or use local hosts file redirection.
• I have now started and am maintaining a clean local service instance for you on port 3006 (MTCLAW pre-routing is disabled to ensure L1 deterministic paths).

1. Pre-Recording Checklist (Mandatory)

Scenarios 2 & 3 L1 Deterministic Routing (Tested and matched ✅)

| Scenario | AI Conversation Input | L1 Rule | Tool | Expected Output | Test Result |

|---|---|---|---|---|---|

| 1 BOM Impact Analysis | Query the impact scope of P-1001 | bi_change_impact (p9) | analyze_change_impact | 3 BOMs (ASSY-A001/002/003) + 4 parts + Medium | ✅ Matched L1, score 10 |

| 2. Data Quality Inspection | Check Part data quality | bi_data_inspect (p8) | inspect_quality | Five-dimensional scoring + issue statistics | ✅ Matches L1, score 13 |

| 3. Governance Record Traceability | View recent governance records | bi_gov_history (p8) | get_governance_history | 20 audit logs | ✅ Matches L1, score 9 |

Test method: POST /api/intent/recognize. Under UTF-8 encoding, three statements respectively stably matched the above L1 rules, level=L1, not proceeding to L3 LLM.

⚠️ Note: Passing Chinese directly via curl in Git Bash may cause garbled characters, leading to misjudgment as unknown; the browser UI using UTF-8 is unaffected, record normally.

3. Supplementary UI Page Numbers (Script Annotations · Requires Data Alignment ⚠️)

• Quality KPI Dashboard: 10 KPI cards + 185 alerts (System 80 + Employee 105, high-level "SCSAI request failed")
• Data Asset Management: 328 assets / Part 200 / Vendor 80 / Doc 48
• Dashboard Home: Part 1223 / 28 ECRs Pending / 62 Employees Online

⚠️ Data Alignment Note (Important): The current local 3006 test returns a different set of data (e.g., total_parts≈12847, audit total≈9999, /api/dashboard/asset-overview 404), which is inconsistent with the numbers in the "script annotations" above. This is because the local environment hasn't loaded the demo seed data (the data/phosphorus-demo/ directory is empty, and seed-quality-demo.js does not exist).

Two approaches, choose one before recording:

• (A) Load the demo seed data (Recommended, to make on-screen numbers match the script exactly): Our team will generate/import phosphorus-industry-themed seed data into the local database. After restarting, the dashboard/asset/governance numbers will align with the script.
• (B) Record with the actual numbers: Record directly using the numbers actually displayed on screen, and replace the numbers in the "script annotations" above with the values seen, adjusting the voiceover accordingly.

Whichever is chosen, the three AI dialogue scenarios in Chapter 2 do not rely on this batch of seed numbers and have been tested to be independently and stably demonstrable.

IV. Timed Script (0:00–2:00)

Part 1 · Opening & Topic Introduction (0:00–0:15)

Part 2 · BOM Impact Analysis (0:15–0:55)

Part 3 · Panoramic Governance of Data Assets (0:55–1:35)

Part 4 · Governance Closed Loop + Closing (1:35–2:00)

V. Editing & Visual Design

• Unified dark background + cyan/blue highlights; risk level Medium uses amber, High uses red.
• Add an "L1 Deterministic Routing" corner mark to each of the three scenarios, emphasizing no reliance on large models.
• Use zoom + stroke animation when key numbers appear to reinforce memory points.

VI. Digital Annotation Standards

• Add same-color strokes or arrows to key numbers in screenshots/recordings; voiceover must strictly match on-screen numbers.
• Annotate the rule IDs for the three scenarios (bi_change_impact / bi_data_inspect / bi_gov_history) to reflect "determinism."

7. Risk Mitigation Plan

• If AI returns empty for a scenario → Check if the 3006 service is alive (/api/change/stats), then resend once.
• If the on-screen numbers do not match the script → Follow the "3. Data Alignment" procedure (Option A: Load seed data / Option B: Switch to real numbers).
• If the address bar needs to display ylxt.chat → Use editing overlay to cover it.
• If MTCLAW/large model occasionally blocks → Confirm MTCLAW_ENABLED=false in .env (already set for you), forcing L1 routing.

8. Final Recording Checklist

• [ ] 3006 service is alive (/api/change/stats returns 200)
• [ ] The three AI statements each hit the corresponding L1 rule (test first via POST /api/intent/recognize)
• [ ] On-screen numbers match the script (or decision to use real numbers is made)
• [ ] Dark theme + ylxt.chat visible in address bar throughout
• [ ] All three scenarios display the "L1 Deterministic Routing" corner mark
• [ ] ECR-2026-0005 status is new (confirmed)