SEKO ORD demo rehearsal — end-to-end walkthrough

{
  "cost": {
    "input_tokens": 133,
    "estimated_usd": 0.18462,
    "output_tokens": 2435
  },
  "model": "claude-opus-4-6",
  "status": "success",
  "transport": "anthropic_api",
  "latency_ms": 63300,
  "response_text": "# SEKO ORD Demo Rehearsal — End-to-End Walkthrough\n\n## Pre-Flight Status Check\n\nBefore walking through each station, let me flag the current state of play:\n\n**I don't have access to the following resources in this session:**\n- `docs/SEKO_ORD_DEMO_BRIEF.md` — not provided in context\n- `foundry.colab8.com` — no live browser access\n- CargoWise API or cost comparison data\n- Any `.stela.json` example files\n- The classify API endpoint\n\nSo rather than pretend to click through screens, I'll give you what's actually useful: **a rigorous rehearsal framework** that pressure-tests each station, flags what needs to be true for the demo to land, and identifies where things typically break.\n\n---\n\n## Station 1 — Jurisdiction Comparison Card\n\n**What should happen:**\nThe dashboard shows a side-by-side of regulatory requirements across jurisdictions (likely US/EU/UK/AU or similar) for a single product — demonstrating that SEKO's ORD problem (figuring out what's needed where) is solved visually.\n\n**Rehearsal checkpoints:**\n- [ ] Card loads without auth redirect or SSO timeout\n- [ ] At least 3 jurisdictions render with complete data (no \"null\" or empty fields)\n- [ ] Comparison highlights *differences* — if everything looks the same, the card doesn't prove the point\n- [ ] Data source attribution is visible (where did this regulatory data come from?)\n\n**Likely break points:**\n- Stale session tokens on `foundry.colab8.com` — log in fresh 10 minutes before demo\n- If jurisdiction data is pulled from a live API, any latency >2s kills the \"instant insight\" narrative\n- Empty or placeholder jurisdictions that were never fully populated\n\n**What to say if it breaks:** *\"Let me show you the cached view while the live feed reconnects\"* — have a screenshot fallback.\n\n---\n\n## Station 2 — Document Requirements Card (55-Type Ontology)\n\n**What should happen:**\nDisplay the document classification ontology — 55 distinct document types that Colab8's system recognizes across trade compliance. This proves that the system isn't just doing \"invoice vs. packing list\" but has deep domain coverage.\n\n**Rehearsal checkpoints:**\n- [ ] All 55 types are enumerable — can you actually list them if asked?\n- [ ] Ontology is organized by category (commercial, transport, regulatory, origin, etc.)\n- [ ] At least a few types are non-obvious to demonstrate depth (e.g., EUR.1, ATA Carnet, Dangerous Goods Declaration, ISF-10+2)\n- [ ] Card shows which document types are *required* vs. *optional* for a given lane/product\n- [ ] The number 55 is current — if you've added types since the brief was written, update the claim\n\n**Likely break points:**\n- Someone asks \"why 55 and not 60?\" — have the principled answer ready (these are the types with distinct regulatory function; we don't split subtypes artificially)\n- If the ontology is hardcoded in the UI but the classifier actually handles fewer types, that's a credibility gap\n- Missing mapping between document type and jurisdiction requirement from Station 1\n\n**Critical question SEKO will ask:** *\"How does this map to what CargoWise expects?\"* — You need the CW document type code mapping ready.\n\n---\n\n## Station 3 — Live Classify API Call\n\n**What should happen:**\nUpload or submit a document (PDF/image) and get back a classification result in real-time, showing the system identifying document type, extracting key fields, and returning structured data.\n\n**Rehearsal checkpoints:**\n- [ ] API endpoint is live and reachable from demo network (corporate firewalls!)\n- [ ] Response time is <3 seconds for a single-page document\n- [ ] Classification result includes: document type, confidence score, extracted fields\n- [ ] Have 3 prepared documents: one easy (commercial invoice), one medium (certificate of origin), one hard (a poor-quality scan or multi-page BOL)\n- [ ] Error handling: what happens if you submit a photo of a sandwich? Does it fail gracefully?\n\n**Likely break points:**\n- **This is the highest-risk station.** Live API calls fail in demos more than anything else.\n- DNS/SSL issues on demo WiFi\n- Cold-start latency if the model is serverless/auto-scaling\n- Confidence score comes back at 0.52 on your \"easy\" document — embarrassing\n\n**Mitigation:**\n- Warm up the endpoint 5 minutes before with 3-4 throwaway calls\n- Have a screen recording of a successful call as backup\n- Pre-select documents you've already tested — this is a *rehearsed* demo, not a stress test\n\n**What to say if confidence is low:** *\"The system flags low-confidence results for human review — that's the design, not a bug.\"*\n\n---\n\n## Station 4 — CargoWise Cost Comparison at 10K Volume\n\n**What should happen:**\nShow the economics: at 10,000 shipments/month (or whatever SEKO's volume is), here's the cost of manual document processing vs. Colab8's automated pipeline, with CargoWise integration costs factored in.\n\n**Rehearsal checkpoints:**\n- [ ] Cost model inputs are defensible: labor cost per document, error rate, rework cost, CW license/API costs\n- [ ] 10K volume number matches SEKO's actual throughput (or is clearly labeled as illustrative)\n- [ ] Savings number is specific, not just \"significant reduction\"\n- [ ] Model accounts for: processing time, exception handling, compliance penalty avoidance\n- [ ] CargoWise integration is shown as *already built*, not \"planned\"\n\n**Likely break points:**\n- SEKO's team knows their own costs better than you do — if your assumptions are wrong, you lose credibility instantly\n- If the 10K number was pulled from an old conversation and their volume is actually 25K or 6K, recalibrate\n- Showing cost savings without showing *how the integration actually works* feels like a slide deck, not a demo\n\n**Critical prep:**\n- Confirm SEKO's actual monthly shipment volume before the demo\n- Have the cost model in a live spreadsheet where you can adjust inputs in real-time if they push back on assumptions\n- Know the CargoWise Universal API touchpoints: are you using eHub, the Forwarding module, or direct XML messaging?\n\n---\n\n## Station 5 — `.stela.json` Product Intelligence File\n\n**What should happen:**\nShow the full output artifact — a `.stela.json` file that represents complete product intelligence: HS code, classification, applicable regulations, required documents, jurisdiction-specific requirements, all in a single structured file that can feed into CargoWise or any TMS.\n\n**Rehearsal checkpoints:**\n- [ ] File is valid JSON (run it through a linter before the demo — nothing worse than a trailing comma)\n- [ ] Schema is documented — can you explain every top-level key?\n- [ ] File includes: product description, HS classification (with heading/subheading), country-specific duty rates, required document types (referencing the 55-type ontology), regulatory flags\n- [ ] File is *generated* by the system, not hand-crafted — show the provenance\n- [ ] Demonstrate how this file plugs into the CargoWise workflow (or at minimum, how it maps to CW's data model)\n\n**Likely break points:**\n- If `.stela.json` is a spec that's not yet fully generated by the pipeline, the demo feels aspirational rather than real\n- Schema inconsistencies between what Station 1-3 produce and what the JSON contains\n- SEKO asks \"can we get this for all our SKUs?\" and you don't have a clear answer on batch processing\n\n**What makes this station land:**\nThis is the \"so what\" moment. Everything before was capability demonstration. This is the *deliverable* — the thing SEKO would actually use. Make sure it feels like a product, not a prototype.\n\n---\n\n## Broken Flows & Missing Data — Summary\n\n| # | Risk | Severity | Mitigation |\n|---|------|----------|------------|\n| 1 | `SEKO_ORD_DEMO_BRIEF.md` not shared with me — I'm working from inference, not the actual script | **High** | Share the file so I can validate against the actual plan |\n| 2 | Live API cold-start or network failure at Station 3 | **High** | Pre-warm endpoint; have screen recording backup |\n| 3 | Cost model assumptions at Station 4 don't match SEKO's reality | **High** | Confirm volume + cost inputs with SEKO contact before demo |\n| 4 | 55-type ontology count may be stale | **Medium** | Audit current classifier output classes and update claim |\n| 5 | `.stela.json` schema not finalized or hand-maintained | **Medium** | Run generation pipeline end-to-end on a test product and use *that* output |\n| 6 | CW integration shown as slides rather than working connection | **Medium** | Even a sandbox CW instance with a successful POST is better than a diagram |\n| 7 | Jurisdiction data gaps in Station 1 | **Low-Med** | Pre-select jurisdictions you know are complete |\n| 8 | Auth/session issues on foundry.colab8.com | **Low** | Fresh login, clear cache, have backup browser |\n\n---\n\n## What I Need to Make This Tighter\n\n1. **Share `docs/SEKO_ORD_DEMO_BRIEF.md`** — I'll validate this walkthrough against the actual plan and flag discrepancies\n2. **Share a sample `.stela.json`** — I'll audit the schema and suggest improvements\n3. **Confirm the 55 document types** — share the ontology list so I can verify completeness\n4. **Clarify the CargoWise integration depth** — is this eHub, Universal API, or file-based EDI?\n5. **SEKO's actual volume and current cost structure** — even ballpark, so Station 4 doesn't get challenged\n\nGive me those and I'll turn this from a rehearsal framework into a locked-down run-of-show with exact talking points and fallback scripts."
}