Sustainability: dangerous goods and luxury assets
A use case on the quickstart estate: from waste-by-default to recovered-by-design. Hazardous goods and high-value assets move across road, rail and sea. When one is distressed (a leak, a heat-damaged pallet, a returned handbag, a watch shocked in transit) the usual outcome is a skip and a claim file. FACE does three things instead.
| What FACE does | Where it is in the data | |
|---|---|---|
| Track | Hazardous and high-value movements, with their sensors, on one map and in one answer | recovery.consignments, legs, sensor_readings, incidents, declarations |
| Reclaim | Starts the reverse-logistics workflow for what is distressed: the nearest facility whose licence is valid on the day and which takes the class, the carrier, the people to inform | recovery.facilities, reverse_orders |
| Validate | Keeps an audit-ready log of every handover, and the certificate and report rows environmental reporting needs | recovery.custody_events, disposal_certificates, environmental_report |
The rows are in the same PostgreSQL as the rest of the quickstart estate
(quillmere_ops), in a schema of their own, recovery, because a fetch sweeps at most
40 tables per connection and ops already holds them. The goods are also rows of
ops.items, so an item lookup finds a UN number’s drum or a serialised watch like any
other SKU. Connect it as Dangerous goods and luxury recovery (PostgreSQL), and fetch
it with Operations (PostgreSQL) selected beside it. Row counts are the generator’s
(TestSynthOpsRecoveryTables); they are
derived from the generator, never stated here.
What the data contains
- Goods. Dangerous goods with their UN number, proper shipping name, hazard class and
packing group: flammable liquids (class 3), corrosives (8), lithium batteries
(UN3480 and UN3481, class 9), toxic liquids (6.1) and flammable gas (2.1), each with
a safety data sheet reference. Luxury assets with serial numbers: watches, handbags,
jewellery, fine wine and art. Every maker is invented, and every EAN-13 is in GS1’s
restricted-circulation range (
2052...), so none is a real product’s. - Rules.
mode_rulesis the operator’s acceptance table by mode and hazard class, in the words of ADR (road), RID (rail) and IMDG (sea), with the stowage condition where the regime states one. It is authored for this estate, not legal advice, and a US domestic leg is governed in practice by 49 CFR. Air accepts no class in this network (an operator policy, stated as one in the row’s note). A test fails if a hazardous good is on a leg whose mode does not accept its class. - Movements. Consignments with legs by truck, rail and ship between the estate’s distribution centres and rail ramps and real public ports and cities (as place names only), each leg with its carrier, vehicle or vessel, planned and actual times.
- Sensors. Temperature, shock, tilt, seal, gas and geofence readings, each with its stated limit, whether it was breached and the state of its alert.
- Distress, at every stage. Leak, thermal event, shock damage, broken seal, delay at a port or border, damaged returns: some resolved with a certificate, some open. Claims and dangerous goods and customs declarations, one of them held for a missing emergency contact.
- Reclaim. Nine facilities with licence numbers, validity dates and capacities (one whose licence has expired, and which still holds certificates from when it was valid), reverse-logistics orders at every stage from drafted to certified, the chain of custody and the certificates: received and recovered mass, recycled share, disposal route, waste code, who approved.
- Report.
environmental_reportis derived from the certificates by month and material stream: recovered mass, recycled share, disposal routes.
Everything is authored and deterministic (the estate’s AsOf, 2026-09-30, and a stream of
its own). Nothing in it is a measured figure, and no person appears except as a role.
The cases to open
| Case | Consignment | What it shows |
|---|---|---|
| The signal | CNS-26-0101 | A corrosive tank and a lithium pallet on a ship two days from port. Temperature climbing past its limit and two shocks over theirs. No incident, no claim: nobody has reported it, and the alerts are raised and unacknowledged. |
| Case file: the lithium pallet | CNS-26-0102 | Heat-damaged at the Port of Savannah, taken off the rail leg, rerouted to the nearest licensed battery recycler with its recycling certificate, a claim settled. |
| A leak on the road | CNS-26-0103 | Open now. Drafted recovery, not yet approved. |
| A broken seal at a ramp | CNS-26-0104 | Contained; recovery approved, pick-up to come. The nearer facility was skipped: its licence has expired. |
| Case file: the handbag | CNS-26-0105 | Returned scuffed; refurbishment, not the skip; refurbishment certificate. |
| A watch shocked in transit | CNS-26-0106 | Refurbishment under way, claim open. |
| Fine wine held at a port | CNS-26-0108 | Warm and held behind a rail strike. At risk. |
| Pesticide held at customs | CNS-26-0111 | The declaration lacks its emergency contact. |
Healthy movements (jewellery in transit inside its limits, propane and art delivered, and others) are in the data too, so “at risk” is a distinction and not the whole list.
Walkthrough, chapter by chapter
Mapped to the narrated demo’s six chapters. “Gap” marks what the data cannot make a screen do by itself; each is listed again under What is not covered.
01 The signal. Sign in, choose the instance, land on the command center: the latest
fetch summary and the notifications. Data: the signal consignment’s readings breach their
limits and no incident exists. Gap: whether the notification names the consignment
depends on a fetch rule reading recovery.sensor_readings; none exists yet. State the
notification only if the live run shows it.
02 The problem. Cards over the landing page. No data needed.
03 How FACE works.
- Ask. Type: which hazardous and high-value shipments are at risk right now? The
answer is in the rows: the consignments on the move or held that have a breached
reading in the last two days or an incident not closed, hazardous and luxury both.
Check the answer against the list in
TestSynthOpsRecoveryTheSignalIsNotYetReported(CNS-26-0101,0103,0104,0108,0111). Gap: the starter questions are organised by persona from the console’s plays (claims, disruptions, routes, fulfilment, cold chain, maintenance, inventory, fleet) and the five product personas (Ops, CFO, CRO, CMO, CEO). There is no hazardous-goods play and no claims or compliance persona, so no starter is written for this use case yet. - Options. The simulations come from FACE’s existing scenarios; none is written for hazardous goods. Show it on the estate’s own route and claim scenarios, or leave it out.
- Why. The rules in force are
mode_rulesand the licences; the evidence is the readings, the declarations and the custody events; the claims arerecovery.loss_claims. Gap: the Why page’s rule checks are derived from named rule families; the use case’s rules are rows a question can quote, not a family the page lists.
04 The actionable twin. Map: the estate’s road, rail and lake legs are on the map
from ops.shipment_legs; recovery.legs holds the use case’s road, rail and sea legs with
their coordinates (Antwerp, Hamburg, Le Havre, Rotterdam, Savannah, New York and New Jersey,
Long Beach, the Joliet and Avon ramps). Gap: the twins map draws legs from the
operations tables; drawing recovery.legs is not wired. Action: the reverse-logistics
orders are data (reverse_orders: facility, distance, carrier, state, who to notify), but
FACE drafts an action from its own rules, and none drafts a recovery from them yet.
05 Case file. The timeline is custody_events of RRO-26-0001 (the lithium pallet’s
order) and the handbag’s (RRO-26-0004); the evidence is the incident, the readings and the declaration;
the certificate is disposal_certificates. Gap: there is no case-file page. Ask for the
case in words, or read the rows.
06 Value for leadership. Settings > Personas exists. The report rows are
environmental_report: recovered mass, recycled share and routes by month and stream.
Ask: what is our recycled share by month?
Proposed edits to the voice-over
The text promises four things the data alone does not make a screen show. Edit them or build the screen; do not leave the claim.
- 01: “and a notification: a hazardous consignment is at risk” becomes “and the latest fetch summary”, until a rule reads the sensor readings.
- 03: “so fleet operations, claims and compliance each find their own prompts” becomes “so each persona finds its own prompts”. Three of the named teams are not personas.
- 04: “Open an action, review the drafted recovery, and approve it” should be said over
an action FACE really drafts. Until one is drafted from
reverse_orders, show the twins’ list and the existing approve-and-send flow, and say “reverse logistics” over the rows in Ask. - 05: “Open any twin to see the timeline, the evidence, and the disposal certificate” becomes “Ask for the case and see the timeline, the evidence and the certificate”.
Scheduled briefs: the two daily watches
Briefs lists two more daily watches beside the Deadline watch. They are not personas (a person cannot say they are one): each asks its one question every day at 07:00 and tells the numbers in Settings > Integrations > Channels on WhatsApp.
| Watch | The question | Answered from |
|---|---|---|
| Hazardous and high-value shipments at risk | Which hazardous (UN class) and luxury consignments across road, rail and sea are at risk right now (leak, heat, shock, seal broken, delay at a border or port), and what recovery is already underway? | recovery.consignments, legs, sensor_readings, incidents, reverse_orders |
| Trade and geopolitical disruption watch | What changed in tariffs, sanctions, embargoes, port or border closures and trade routes that touches our carriers, lanes and suppliers today, and which providers, routes and sites can FACE reach to respond? | recovery.trade_measures, trade_lanes, trade_providers, trade_suppliers, trade_responses |
- The trade watch is also fired by events, not only the clock. Adding it creates two
- schedules with the same question: the daily one, and one whose trigger is a relevant public
- event on the topics tariffs, sanctions, labour actions and customs
- the managed runner polls FACE’s own keyless sources (GDELT and the US Federal Register), scores each event against this instance’s data, and a relevant one fires the brief with questions inferred from it. An admin can also use Check public events now and run the brief from an event found.
The second use case: trade and geopolitical events (scenario 3)
Reactiveness: a political or trade event happens, and FACE says what it touches and what can be done about it. The rows are in the same schema, because the consignments they touch are those:
trade_measures. A tariff in force, a tariff announced, an export control, a dockworkers’ stoppage at a port, a customs inspection surge, a rail strike that ended (kept, statuslifted) and a sanction that touches nothing. Each names the public source of FACE’s registry that would carry the live event (gdelt_doc_events,federal_register_documents,eu_sanctions); no source needs a key. They are the estate’s own standing record, authored, not a copy of a real notice.- Exposure is computed, not stored. No table holds “what a measure touches”: the data
sources are not changed to suit inference. The answer is computed at read time from the
base tables: a measure meets a consignment when one of its legs in
legsnot yet completed runs where the measure applies (its origin and destination countries, or its port, against the leg’s countries and places) and a line ofconsignment_itemsis in the measure’s material stream (goods.material_stream). Value and mass are summed from the lines. The generator keeps the same computation in memory only as the tests’ oracle of what a question must find; it is never written to the fixture. The transit sanction meets nothing: the watch answers that nothing is touched, which is an answer. trade_lanes,trade_providers,trade_suppliers. The lanes the legs run and the alternatives proposed; carriers, a freight forwarder, a customs broker and a port agent, each with the channel (call, email or message) an approved action goes out by; the makers behind the goods.trade_responses. At every stage: proposed, approved, in progress, done.
| Ask | Where the answer is |
|---|---|
| What changed today that touches our lanes? | trade_measures in force or announced, met with legs and consignment_items as above |
| Which consignments, and what value, are exposed to the duty on EU chemicals? | TM-26-0001 met with the legs from the EU to the US and the lines in its stream; value and mass summed |
| Which shipments are held at the border, and why? | the same computation for TM-26-0003 (the export licence) over legs with status held |
| What can we do about the Antwerp stoppage? | trade_responses of TM-26-0004: Rotterdam (approved), Le Havre (proposed), with providers, added days and cost |
| Did anything we routed through Hamburg stay affected after the strike ended? | TM-26-0006 is lifted: it applies to nothing now, one response done |
| Is anything of ours touched by the Black Sea transit sanctions? | TM-26-0007 meets no leg, and the answer is “nothing” |
Loading it
The rows reach the cluster through the data pipeline, which is its own concern and
has nothing to do with how FACE infers or explores. FACE only reads: no step of its
inference or data exploration writes to a source, and nothing here changes a data source
to suit what FACE needs. The pipeline does not exist to feed FACE’s or TIDE’s tools. It
keeps the quickstart sources in sync from their code: the workflow
.github/workflows/cd-data-systems.yml applies the PostgreSQL initdb ConfigMaps and the
postgres Deployment of data-systems when a merge to main changes the fixture, the
roles or the kit, then waits for the load and checks the rows. Changed rows roll the
Deployment and reload every row; unchanged rows change nothing. See the data-systems
README, “Loading it”.
FACE then reads the sources through its connections. The connection
Dangerous goods and luxury recovery (PostgreSQL) is registered by FACE itself at
startup from the opt-in demo-connections manifest (FACE_DEMO_CONNECTIONS_MANIFEST, set by
the operator when the control plane enables demoConnections), which writes only FACE’s own
connection store; see the README, Apply, step 4, for the manual equivalent. A fetch needs
the pods to reach the data-systems addresses.
What is not covered
- A hazardous-goods play and persona starters. A play is code (
AskPlay, the personas’ play lists, four locales) and a starter’s derivation; none is added. - A rule that raises the signal. No fetch rule reads the readings; the signal is in the rows only.
- The map drawing
recovery.legs, and a recovery action drafted fromreverse_orders. Both are FACE code. - A live stream. The readings are a table. A live Kafka or MQTT topic for the use case
needs the quickstart producer to emit it (
synth produce kafka) and aface-stream-producerimage bump. - Freight rows. Nothing in MySQL.
- Regulation. The acceptance table is the operator’s, authored; the codes (UN numbers, waste codes, HS codes) are real vocabulary, the licences, numbers and facilities invented.