PROGRESS REPORT · T+16h
CONVO T-16h FIRST COMMIT T-13h NOW 13:16 CEST · 2026-04-19

Sixteen hours ago, this didn't exist.

T-16h (yesterday, 21:00 CEST): Damon tells Jonathan about a gap in US customs filing. T-13h: First commit. T-0h (now): 336 tests green. Live API. Three jurisdictions mapped. 202 source documents catalogued. One honest path to $5M ARR from a single customer network.

10h
Focused Build-Time
8
Production Modules
336
Tests, All Green
3
Jurisdictions Scoped
T-16h · 2026-04-18 · 21:00 CEST
Input: one conversation. One URL (catair.atwenture.com). One customer with a specific, expensive problem.
T+0h · 2026-04-19 · 13:16 CEST
Output: this page.
The Revenue Math

This is not a TAM.
This is one customer, known volume.

Damon's networks already route ~350,000 customs filings per month per operating company. Every filing goes through the compliance software layer we're building. The pricing benchmark is 12¢ per transaction. The math is trivial. The work is to make it real.

$0.12 per filing
× 350,000 filings / month
$42,000 per month · per company
× 12 months
$504K
ARR · per single operating company
× 10+ companies in Damon's network
$5M+
ARR · single customer network · steady-state

These are not projections. Damon has the filer code. Damon has the ACE sandbox. Damon has the volume. The question is not "will customers buy this?" — the question is "can we ship byte-exact compliance fast enough?"

The Sprint · T-16h to T+0h

What gets built when the spec is already in the PDFs.

A customs agent reading official CBP chapters, generating compliant code, deploying to production — then doing it again for three more jurisdictions. Every step below happened between yesterday 21:00 and today 13:16.

T-16h · 2026-04-18 · 21:00 CEST
The conversation.
Damon describes the gap: CBP-filer code acquired, ACE sandbox open, LOI signed — missing is the byte-exact software layer turning modern JSON into CATAIR-compliant filings.
T-13h · 2026-04-19 · 00:30 CEST
First commit. Three hours of night-shift.
100 official CBP PDFs OCR'd via GPT-4o Vision. Chunked, embedded (e5-large 1024-dim). Structured extractor extracts typed ontology entities.
T-11h · 02:30 CEST
Ontology lives.
115 modules · 240 transactions · 1,897 records · 4,604 data elements · 836 code tables · 862 error codes · 4,121 business rules · 242 workflows. 18,691 typed edges.
T-9h · 04:30 CEST
Live on the open internet.
catair.atwenture.com — landing, pitch deck, pre-meeting checklist, A/B report. Docker on Hetzner, nginx vhost, Let's Encrypt cert. Production-grade from the start.
T-7h · 06:30 CEST
Code generation layer.
Python package catair generated 1:1 from the ontology. Pydantic v2 Records with ClassVar[LAYOUT], byte-exact serialize/parse, ABI envelope, audit log (SQLite+JSONL), transport abstraction.
T-5h · 08:30 CEST
Jurisdiction sprint: UK, Mexico, EU.
Three parallel research terminals produce first-pass customs ontologies. Verify-passes fix 6 hallucinated facts across UK/MX/EU. Cross-jurisdiction concept map with 87 typed concepts emerges.
T-4h · 09:30 CEST
Damon answers the pre-meeting checklist.
14/14 questions in 90 seconds. Production MQ config delivered: QGC1, cbpgw-trade-cert.cbp.dhs.gov:31531, CBITAMERICA, inbound/outbound queues, keystore+CCDT. This is not a pitch anymore. It's a kickoff.
T-3h · 10:30 CEST
Module sprint 2 — "pattern-is-repeatable" confirmed.
Three additional modules shipped in 2h25m: census_warning, courtesy_notice, ad_cvd. 100 new tests, all green first-run. Avg. ~45 min per module via parallel terminals.
T-2h · 11:30 CEST
202 official source documents catalogued.
UK 60 · Mexico 78 · EU 64. 165 directly verified URLs with version + date + license. MX VUCEM WSDL/XSD files confirmed publicly downloadable — no e.firma needed. Ingestion-ready inventory.
T-1h · 12:30 CEST
eBond Input Records byte-exact.
Record10-Input, 20, 30, 35, 36, 45, 46 — seven records, 80 bytes each. 39 tests green. cb_create_update transaction assembler. End-to-end CB Create filing serialization works.
T+0h · 13:16 CEST · NOW
REST API live · demo ready.
POST /filings/ebond/cb · GET /filings/{event_id} · Pre-flight model-validation · Smoke-demo script prints 80-byte-per-line CATAIR bytes. 336 tests green. Two Filer bugs hunted and pinned with regression tests.
Current Inventory

What's deployable, what's in-progress, what's a gap.

Honest accounting. Every claim here is verifiable at catair.atwenture.com or in the us-customs-platform/ repository.

Knowledge Layer
US Customs Ontology
PDFs ingested100 / 100
Chunks embedded1,430
Records extracted1,897
Typed edges18,691
Obsidian vault✓ live
Code Generation
Python Package catair
Modules shipped8
Records byte-exact80+
Tests green336
Pydantic v2✓
Regression-pinned✓
Runtime
REST API · Audit · Transport
POST /filings/ebond/cblive
Idempotency✓
Audit log 7y-ready✓
MockTransport✓
MQ · AS2stubs ready
Demo Surface
catair.atwenture.com
Landing + live Q&A✓
Pitch deck✓
Pre-meeting checklist✓ answered 14/14
A/B comparison86×
Smoke demo script✓
Multi-Jurisdiction · Research
UK · Mexico · EU Scoped
Research docs written29
Cross-jurisdiction concepts87
Verify-passes run3/3
Hallucinations fixed6
Source docs catalogued202
Customer-Integration · eBond CB
Production Path — 80% there
CB Input records7/7
BS parsing✓
pymqi integrationstub → real · 4-6h
Keystore loadingawaits cert
Filer code placeholderpending
MVP · Business-Model Validation

Stripped to the bone: what must exist to validate the business?

Everything else is scope-creep. The following list is what lets Damon send one real CB submission to CBP Sandbox, get an accepted response, and trigger a billable line on our side.

  • ✓
    Byte-exact CB Create serialization · 7 input records. Done. 39 tests green. Verbatim CBP spec citations in every field.
  • ✓
    BS response parsing · structured error codes. Done. Appendix-S code mapping built. 14 tests.
  • ✓
    REST API endpoint accepting JSON, returning CATAIR bytes + event_id. Done. Idempotency + 5 API tests green.
  • ✓
    Append-only audit log · 7-year retention compliant. Done. SQLite + JSONL. Every event signed + timestamped.
  • ·
    Real IBM MQ transport · pymqi wired to Damon's queue config. Needed: 4-6h dev. Config already delivered by Damon (QGC1, queue names, channel). Keystore handoff unblocks.
  • ·
    mTLS channel with CBP-issued x509 keystore. Needed: 1-2h dev once keystore arrives via SFTP/PGP.
  • ·
    Filer code wired to env — placeholder mode for sandbox start. Needed: 30 min. Allows parallel start while C3 confirms.
  • ·
    Per-call billing meter · 12¢ unit counter + invoice-line export. Needed: 3-4h. Straight increment on submitted events. CSV export + webhook-to-billing later.
  • ·
    One real round-trip against CBP sandbox queue. Needed: 2-4h testing after cert + filer code arrive. Everything above was prep for this one moment.
SUMMARY
4 of 9 MVP items are done. 5 remaining require ~12-16 hours of development + cert-handoff dependencies. First real sandbox submission is 5-7 working days away if CBP filer code and keystore arrive inside 48h.
The Three Real Killers

What would actually kill this?

Not scope-creep. Not pricing debates. The three things below are the only outcomes that stop the train. Each has a concrete mitigation path and a timeline impact.

★ Killer #1
CBP filer-code must be new-applied.

If Damon's existing code can't be reactivated and CBP requires a new 60-90-day application, our "7 days to first sandbox" claim becomes "30+ days". Not fatal — but the momentum cost is real.

Probability
~30%
Impact
+3 weeks
Mitigation
Placeholder mode · parallel build against sandbox queue name with dummy code · switch on confirmation
★ Killer #2
Azure-hosted requirement, Hetzner not permitted.

If Damon's security policy forbids outbound MQ-over-the-Atlantic to an EU data center, the "Option B — we host" architecture forces Azure migration. Infrastructure + compliance work: +40-60 hours.

Probability
~25%
Impact
+2 weeks
Mitigation
Stack is already containerized · Azure replication is config-port, not rewrite · lift-and-shift feasible
★ Killer #3
CBP rejects byte-exact output on a technical nuance.

Our compliance guarantee rests on verbatim citation from the official spec PDFs. If an undocumented nuance (hidden filler byte, channel-specific padding) breaks first submission, we need an iteration loop with CBP. Real risk — mitigated by historical-payload diff in meeting #2.

Probability
~15%
Impact
+1 week per cycle
Mitigation
Ask Damon for one anonymized historical CB payload · diff byte-for-byte against our output · catch before submission
Pace Calibration

If we hold this pace, how far is "done"?

16 hours delivered 8 modules, 336 tests, 3 jurisdiction scopes, a live API, and a verified customer-facing checklist. Extrapolation is imprecise — but within one order of magnitude, here's the honest curve.

Demonstrated pace

In 10 focused hours of development (not elapsed hours):

  • 1 new module~35-45 min
  • 1 jurisdiction research pass~60-90 min
  • 1 integration layer (API · auth · audit)~45-90 min
  • 1 verify-pass · 5 live spot-checks~45-60 min
  • 1 source-inventory (jurisdiction-wide)~45 min

Parallelized via multiple terminals, each with isolated write-scope. Coordination overhead < 10%.

IF 5 focus-hours / day
AND parallel-terminal dispatch
AND no Azure-migration (or it's 1 killer-offset)

→ Week 1 MVP to sandbox · CB+BS live
→ Week 2 Entry Summary full + PGA-NHTSA production
→ Week 3-4 UK V1 corpus ingested · 27-30 docs
→ Month 2 UK RAG live · EU-common layer started
→ Month 3 CBAM module · EU V1 (DE+NL+FR)
→ Q2 MX V1 · white-label · Damon network-wide

IF all killers trigger:
→ Timeline × 1.6 · still under 6 months to production
Stakeholder Roadmap

From today to $5M ARR · five phases · conditional on signal.

Every phase has a clear deliverable, a revenue unlock, and a gate condition that has to be met before the next phase starts. No vapor. Each transition point is decided, not drifted into.

2026-04-19 → +7 days
0
MVP — First Sandbox Submission
Goal: one accepted CB create in CBP sandbox
  • pymqi transport wired to QGC1 with Damon's keystore
  • Filer-code placeholder → production switch
  • Billing meter: per-call 12¢ counter + CSV export
  • Historical-payload diff from Damon (byte-exact validation)
  • First real round-trip against ACS.BRK queues
Revenue unlock
$0 — but it unlocks everything below. Signal quality: "it works end-to-end with real CBP."
+7d → +30d
1
Production — First Operating Company
Goal: one of Damon's companies filing through us in production
  • Entry Summary full (R20-R90) — replaces/complements CB flow
  • Multi-tenant auth + filer-code per client
  • SLA monitoring, 99.9% target, status page
  • Onboarding playbook for subsequent clients
  • Invoice-line automation (Stripe or wire-transfer)
Revenue unlock · ARR trajectory
$42K / month · one operating company live at 350k filings/mo · $504K ARR run-rate
+30d → +90d
2
Scale — Damon's Network
Goal: 5-10 operating companies live · white-label option
  • PGA modules: NHTSA, FDA, EPA, APHIS prioritized by Damon's volume mix
  • 5106 · AD/CVD · Courtesy-Notice full production-ready (partial already shipped)
  • White-label brand layer — "powered by" model or invisible-OEM
  • Admin console for broker ops (Paperclip UI)
  • Historical volume backfill (if customers bring legacy filings)
Revenue unlock · ARR trajectory
$2-5M ARR · full network at scale · proves the single-customer-single-category business model
+90d → +180d
3
Jurisdiction #2 — UK Green Lane
Goal: UK CDS filings live · V1 scope · first UK customer
  • HTML ingestion pipeline (gov.uk HTML vs. PDF)
  • 27-30 UK Tier-1 docs ingested (from source-inventory)
  • CDS API integration (v3.0, OAuth2 + mTLS)
  • UKIMS Green Lane default · Red Lane deferred
  • WCO DM canonical layer — mapping once, reusable for EU/MX
Revenue unlock · moat expansion
UK market entry · per-customer > $500K ARR · cross-jurisdiction network effect begins
+180d → +365d
4
Multi-Jurisdiction — EU + Mexico
Goal: EU V1 (DE+NL+FR) + MX V1 live · CBAM integrated
  • EU common layer: UCC Annex B, EUCDM 7.0, WCO DM v4.1 aligned
  • DE ATLAS 10.2, NL DMS 1.4, FR DELTA-H7 envelopes
  • CBAM module (real-time Customs + TARIC + Single Window)
  • MX VUCEM integration (SOAP/XML, e.firma, Anexo 22)
  • ICS2 R3 · AES · NCTS P5 full coverage
Revenue unlock · category dominance
4-jurisdiction coverage · one byte-exact agent per customs regime · enterprise broker category emerges
What we need next

Three answers unblock phase 0.

Filer code timeline (reactivation or new app?), cert handoff channel (SFTP / PGP / Box?), and whether Hetzner-hosted is acceptable or Azure-replication is a hard requirement. Everything else is us executing against the plan above.

Pre-Meeting Checklist → Read the Pitch → Try the Live Agent →