Landed cost — what we are still closing

This is the roadmap for turning the Landed Cost Simulator (Agent #8) from a customs-grade estimate into a true door-to-door landed cost. It sits alongside How landed cost is calculated: methodology documents what we do model and how; this doc documents what we don't and how we plan to.

Audience: PMs scoping the next 18 months, brokers fielding "is freight in there yet?" from importers, and prospects evaluating the roadmap.

Everything past the "What's in vs out today" section is roadmap — not yet live. The "what we'd build" write-ups describe planned work (the data sources, effort, and designs we intend to use), not shipped features. What runs in the product today is only the customs-grade landed cost described in the next section.

---

What's in vs out today (v1, June 2026)

The current simulator returns a customs-grade landed cost:

Customs landed cost = product (FOB)
                    + base HTSUS duty
                    + trade-program duty (Section 301 / 232 / IEEPA / Fentanyl / …)
                    + MPF (19 CFR 24.23)
                    + HMF (19 CFR 24.24, ocean only)
                    + detention risk addition (modelled, ranged)

Every number above is sourced from a regulatory document (Federal Register, HTSUS schedule, CBP CSMS) or a transparent model documented in How landed cost is calculated. That's deliberately the part that can put goods at the border — the part where being wrong is catastrophic.

What's not in the headline number:

• Ocean freight
• Marine insurance
• Customs broker fee
• Customs bond cost
• Port charges (THC, demurrage risk)
• Inland US trucking
• Anti-dumping / countervailing (AD/CVD)
• PGA filing fees (FDA / FCC / CPSC / USDA / EPA)

Order-of-magnitude these add ~$4K–$13K to a typical $100K ocean FCL PO from China, plus a potentially much larger AD/CVD figure on specific commodities (solar, steel, apparel from China).

That's a real gap. The decision to omit was honest: modelling them poorly is worse than omitting them. This doc is the plan to close each gap with a defensible source.

---

The 8 missing categories

Each section: what it is, magnitude on a typical $100K PO, the data sources we'd integrate (free first, paid second), what we'd build, and the effort + value ranking.

1. Ocean freight

Data sources (free first):

• Freightos Baltic Index (FBX) — fbx.freightos.com — daily spot rate index for 12 major global lanes including CNSHA→USMIA. Free public index. Paid API ~$500/mo for granular lane data.
• Drewry World Container Index (WCI) — weekly, 8 routes, free index. Paid for detail.
• Searates API — searates.com/services/rates-services-api — calculator-style estimates. Free tier (rate-limited).
• Xeneta — xeneta.com — real contracted rates + benchmark percentiles. The gold standard for serious procurement. Enterprise pricing ($$$).
• FreightWaves SONAR — freightwaves.com/sonar — real-time spot + 90-day forecast. $1.5K–3K/mo.
• Flexport API / Project44 — if importer uses them, their actual contracted rates. Free if customer.
• Carrier direct APIs — Maersk Quick Quote, Hapag-Lloyd Quick Quote, MSC's eContact. Per-carrier, varying access models.

What we'd build:

1. Ingest FBX free index daily into tariff_intel.freight_rates keyed by lane + container size + date.
2. Add a paid tier (Searates or Xeneta) for granular lane data — this gets us 25+ lanes vs 12.
3. Surcharge layering: BAF (bunker), PSS (peak season), GRI (general rate increase), port congestion — each as its own rate row so the broker can see where the number comes from.
4. Org-level broker discount override (most brokers have a contracted discount off published rate).
5. Output a range with confidence interval, not a point estimate. Same shape as the detention risk model — defends the methodology.

Effort / value: 2 eng-weeks · Very high. Closes the biggest gap.

---

2. Marine insurance

Data sources:

• Roanoke Insurance API — cargo specialty + customs bonds together. Enterprise.
• Allianz Trade, Chubb Cargo — standard ICC rate cards (Type A: 0.35–0.55%, Type B: 0.20–0.40%, Type C: 0.10–0.20% of CIF).
• TT Club — transport-specialty quotes. Enterprise.
• InsureCargo — aggregator, mid-market.
• Lloyd's casualty data — public loss-ratio data by route + commodity. Used to surface route risk premium (e.g. Red Sea war addendum).

What we'd build:

1. Static rate card per commodity tier (the standard ICC ranges are public). Applied as % of CIF (= product + freight + insurance, iterated once).
2. Route risk adjustment from Lloyd's casualty data (e.g. Red Sea = +0.15% war addendum).
3. Importer policy config — when they bring their own policy, we use it.
4. Surface as an optional line item, not headline (because it's importer-policy specific, not a market rate).

Effort / value: 2 eng-weeks · Medium. No API needed for v1 — just rate-card data.

---

3. Customs broker fee

Data sources:

• The broker themselves — Commers IS a broker; for our own filer code we know our rates. For brokers we onboard, read their fee schedule from org config.
• NCBFAA pricing surveys — public annual benchmarks (NCBFAA = National Customs Brokers & Forwarders Association of America).
• CBP-published broker license registry — license number → brokerage → optional quote URL.

What we'd build:

1. Org-settings table for broker fee tiers (per-entry, per-line, volume discount).
2. Read the broker's filer code → fee schedule → apply.
3. Default to NCBFAA median if not configured.

Effort / value: 1 eng-week · Medium. Internal config, not external integration.

---

4. Customs bond cost

Data sources:

• CBP bond minimum formula — public. 10% of estimated annual duties, taxes, fees with minimum $50K continuous bond.
• Surety carrier rates — Roanoke Insurance, Atlas Insurance, Avalon Risk Management, International Bond & Marine, Westchester Surety. Typical rates 1.0–2.5% of bond amount annually.
• Importer credit history — internal proxy; we can rank by entry volume + flagged history.

What we'd build:

1. Importer's trailing-12 duty exposure (we already compute this) → required continuous bond amount.
2. Apply standard 1.5% surety rate → annual bond cost.
3. Amortize across estimated annual entry count → per-entry bond fee.
4. Surface as a tiny line item (not material on most POs but adds completeness).

Effort / value: 1 eng-week · Low. Small magnitude but trivial to add.

---

5. Port charges (THC, demurrage risk)

Data sources:

• Port authority published tariffs — every US port publishes its tariff. PortMiami's is at portmiami.biz. Public PDFs, scrapeable annually.
• Carrier THC schedules — MSC, Maersk, CMA-CGM, Hapag-Lloyd all publish. Scrapeable.
• JOC / IHS Markit Port Productivity Database — avg dwell time by port (drives demurrage probability). $$$.
• Container Atlas — port-level congestion statistics. Free tier.
• FMC (Federal Maritime Commission) detention/demurrage filings — public. Charge schedules + dispute data.
• PortMonitor / PortBase — European ports, equivalent.

Partnership angle: a port authority is a natural data partner here — ingesting its published tariff directly is more accurate and lower-maintenance than an annual PDF scrape, and keeps the port as the authoritative source of its own rates.

What we'd build:

1. Annual scrape of port tariffs into tariff_intel.port_charges.
2. Carrier THC schedule scrape, keyed by carrier + size + service type.
3. Dwell-time distribution per port → demurrage probability × daily rate.
4. Output: THC (deterministic) + demurrage expected value (modelled range, same approach as detention risk).

Effort / value: 2 eng-weeks · High. Partnership angle adds strategic value beyond the dollar amount.

---

6. Inland US trucking

Data sources:

• DAT iQ — dat.com — the spot-rate benchmark for US trucking. Used by every 3PL. ~$500/mo enterprise.
• Truckstop.com — similar, alternate. Load board + rate data.
• FreightWaves SONAR — forecast + spot + lane density.
• Uber Freight API, Convoy (RIP, replaced by RXO) — instant quote APIs.
• C.H. Robinson Navisphere, XPO Connect, Schneider FreightPower — 3PL quote APIs.
• FMCSA (Federal Motor Carrier Safety Admin) — public data on carriers, mileage, safety scores. Use to filter carrier eligibility.

What we'd build:

1. Port (origin) + destination ZIP → great-circle mileage with road factor.
2. DAT lane rate API call (port-region → destination-region) → spot rate $/mile × distance.
3. Equipment-type adjustment (dry van vs reefer vs flatbed) → final rate range.
4. Fuel surcharge layer (DOE diesel price index, weekly).

Effort / value: 3 eng-weeks · High. Closes the door-to-door story. Needs paid API.

---

7. AD/CVD (Anti-Dumping / Countervailing Duties)

Data sources (all free + public):

• CBP ADCVD database — aceservices.cbp.dhs.gov/adcvdweb — every active case, HTS coverage, cash deposit rates. Free, searchable.
• USITC publications — usitc.gov — investigation findings, sunset reviews, scope rulings.
• Federal Register Trade Remedies notices — rate adjustments. Public.
• ITA ACCESS database — access.trade.gov — manufacturer/exporter- specific cash deposit rates. Public.
• CBP CSMS messages — implementation guidance.

Real example: Hoshine Silicon Industry — currently subject to ADD case A-570-XXX on Crystalline Silicon Photovoltaic cells from China. Cash deposit rate at the manufacturer level can exceed 100% depending on the rate type and the specific exporter ID. On a $94K solar PO this can be a $100K+ surprise at liquidation.

We currently flag ADD/CVD exposure with a warning but don't compute the number. That's the gap.

What we'd build:

1. Daily ingest of CBP ADCVD database → tariff_intel.adcvd_cases, keyed by HTS + origin + manufacturer/exporter ID.
2. On every simulation: HTS + origin lookup → matching active cases.
3. If manufacturer ID known (from extracted documents), look up specific cash deposit rate. Otherwise, use the "all-others" rate.
4. Surface as a dedicated tariff component, separate from the trade- program stack. Bigger number than most components → it should be visually distinct.

Effort / value: 1 eng-week · Very high. Free data, biggest accuracy lift on the commodities most exposed to AD/CVD (solar, steel, aluminum, apparel from China).

---

8. PGA filing fees

Data sources:

• FDA Prior Notice — free filings, no fee. Data filing only.
• FCC equipment authorization fees — published schedule by device class.
• CPSC Section 12 certificate — no fee, data filing.
• USDA APHIS permit fees — public schedule.
• EPA TSCA Section 5/6 — per-substance fees, public.
• DOT/FAA/PHMSA hazmat permits — published schedules.

What we'd build:

1. Static fee schedule per PGA × HTS → fee.
2. We already identify triggered PGAs at classification time → just wire the fee lookup.
3. Surface as a small "PGA fees" line item.

Effort / value: 1 eng-week · Low. Small magnitude, trivial.

---

Beyond costs — the savings angle

The product gets significantly stronger if Commers doesn't just model costs but also opportunities to reduce them. Same data, same agents — different output framing.

Each of these turns Commers from "duty calculator" into "duty optimizer." That's a meaningful product-positioning shift. Same inputs, same data — but the output is "you'd save $X if you did Y."

---

Recommended priority order

For the next 6 months, ranked by value × ease:

If we had to ship one thing first: #1 (ADCVD). Free public data, and the biggest accuracy lift on the specific commodities most exposed to AD/CVD (solar, steel, apparel from China) — the category where an unmodeled duty can turn into a six-figure surprise at liquidation.

---

Partnership patterns

Several categories above are better sourced through direct data partnerships than through scraping or third-party aggregators. We plan data partnerships with port authorities, ocean carriers, sureties, and 3PLs to source their published rates directly.

---

Eval strategy — how we validate predictions

A landed cost roadmap is only credible if we can prove our numbers are right. The eval discipline:

1. Customer billing data — for paying customers, compare predicted vs actual invoice line items. Surface per-category accuracy (freight ±%, broker fee ±%, etc.) like we do classifier accuracy today.
2. Per-lane backtesting — run the same simulation against the past 60 days of FBX data → measure prediction error per lane.
3. ADCVD ground truth — every entry where we predicted ADD/CVD applicability vs CBP's actual finding. Same shape as classifier eval.
4. Honest miss rates — publish per-category prediction intervals (e.g. "freight: ±18% on 90% of lanes"). Importer can decide whether to trust the model.
5. A/B against industry standard — same PO simulated by Commers and by an alternative model (Flexport instant quote, broker manual). Measure rank-order accuracy.

---

Why surface all this rather than pretend

Three reasons:

1. Trust scales worse than features. A sharp buyer (port authority, enterprise importer, regulator) will probe. If we get caught pretending, the damage compounds. If we surface the gap proactively, the same buyer reads it as rigor.

1. A published roadmap is a commitment, not a claim. Each missing category here is scoped with its data source and effort, so a buyer sees the plan rather than a promise. What ships moves into the methodology doc; what's listed here is explicitly not yet live.

1. Honest model bands compound across products. The detention risk model already labels itself as a range, not a prediction. Doing the same for freight, port charges, and AD/CVD compounds the brand of "we tell you what we don't know" — which is what regulated brokers buying customs software actually want.

The landed cost screen's "Out of scope" panel is where this shows up in the product; this doc is the depth behind it.

---

Where it lives