diff --git a/field-guides/README.md b/field-guides/README.md
new file mode 100644
index 0000000..31c27c7
--- /dev/null
+++ b/field-guides/README.md
@@ -0,0 +1,11 @@
+# Condition Monitoring — Field Guides
+
+Practical guides for technicians taking readings on the Mechbase marine demo routes.
+
+| Guide | Sensor / Instrument | Measurement |
+|---|---|---|
+| [Vibration](vibration.md) | TE Connectivity WVS wireless tri-axial accelerometer | Velocity (mm/s RMS) |
+| [Thermography](thermography.md) | FLIR Edge series handheld thermal camera | Surface temperature (°C) |
+| [Ultrasound](ultrasound.md) | Handheld ultrasonic detector | Signal level (dB) |
+
+Each guide covers the instrument, how to set it up and take a reading correctly, and how to enter the result in Mechbase.
diff --git a/field-guides/images/te-wvs-sensor.jpg b/field-guides/images/te-wvs-sensor.jpg
new file mode 100644
index 0000000..c4a7ed6
Binary files /dev/null and b/field-guides/images/te-wvs-sensor.jpg differ
diff --git a/field-guides/thermography.md b/field-guides/thermography.md
new file mode 100644
index 0000000..1281db9
--- /dev/null
+++ b/field-guides/thermography.md
@@ -0,0 +1,79 @@
+# Thermography — Field Guide
+
+Camera: **FLIR Edge series** handheld (visual + thermal, MSX enhancement, FLIR ONE app)
+
+---
+
+## How the camera measures temperature
+
+Thermal measurements show the **surface** temperature of the target. Accuracy depends on three factors:
+
+| Factor | What to watch |
+|---|---|
+| **Distance to target** | Closer = better spatial resolution and spot size. Get as close/zoomed as safely possible. |
+| **Ambient temperature** | Note ambient; it is the reference for ΔT calculations. |
+| **Emissivity** | How efficiently the surface radiates heat (see below). |
+
+### Emissivity
+
+- **Matte / painted / oxidised surfaces** — good emitters; the camera's default ~0.95 setting is a fair approximation.
+- **Bare / glossy / shiny metal** — poor emitters; the default setting will under-read. Apply a strip of high-emissivity tape or a dab of matt paint over the spot, **or** manually set the known emissivity in the camera before reading.
+
+---
+
+## Camera setup
+
+1. Power on; let the camera warm up briefly.
+2. Set the **temperature range** to suit the expected target (e.g. a wider range for electrical gear under fault conditions).
+3. Choose a suitable **colour palette** (e.g. Iron or Rainbow for electrical; Greyscale for quick scans).
+4. Enable **MSX** and set the MSX distance to match the target distance — this overlays visual detail on the thermal image for easier interpretation.
+5. Use the **adjustable measurement spots** to track the hottest and coldest points in the frame. The camera periodically auto-calibrates (mechanical shutter click + brief freeze) — this is normal; wait for it to complete before reading.
+
+---
+
+## Inspection technique
+
+### General
+
+- Inspect under **normal / representative operating load** — an unloaded circuit or idle machine hides faults.
+- Keep the **view angle within ~30° of perpendicular** to the target surface to limit reflection error.
+
+### Electrical (switchboards, distribution panels)
+
+1. Open the inspection cover or use an IR window where required (PPE: arc-flash rated).
+2. Scan **busbars, breakers, incoming terminals, and all connections**.
+3. Compare the **same component across all three phases** — a phase running hotter than its siblings under similar load indicates a loose or corroded connection, an overloaded circuit, or a failing device.
+4. Note the hottest spot temperature (°C) and the ΔT above ambient or the reference phase.
+
+### Mechanical (motors, bearings, pumps)
+
+1. Scan **bearing housings** and **couplings**.
+2. Compare each bearing housing against its pair or against its known baseline.
+3. A bearing running hotter than its baseline or its paired housing suggests a lubrication or wear issue.
+
+---
+
+## Judging severity
+
+Judge findings by **ΔT** (temperature rise over a reference point — ambient air, the established baseline, or a similar phase / component), not by absolute temperature alone.
+
+| ΔT (over reference) | Typical severity |
+|---|---|
+| < 10 °C | Monitor — recheck at next route interval |
+| 10–30 °C | Investigate — plan corrective action |
+| > 30 °C | Act promptly — potential imminent failure |
+
+*These are general guidelines; the Mechbase route item shows the configured limit band for this specific point.*
+
+---
+
+## Safety
+
+- Do **not** point the camera at the sun or laser sources.
+- For live electrical panels, maintain safe working distance and use appropriate PPE; use an IR window or a non-contact safe zone wherever possible.
+
+---
+
+## Recording the reading in Mechbase
+
+Open the assigned **Route** → tap the **Measurement Item** for this thermal point → the item shows the limit band (minor alert: ~65 °C; major alert: ~90 °C in this demo). Enter the **maximum spot temperature** (°C) observed at the target and confirm.
diff --git a/field-guides/ultrasound.md b/field-guides/ultrasound.md
new file mode 100644
index 0000000..1bd678d
--- /dev/null
+++ b/field-guides/ultrasound.md
@@ -0,0 +1,73 @@
+# Ultrasound — Field Guide
+
+Instrument: **Handheld ultrasonic detector** (contact/stinger probe + airborne horn; outputs a dB reading for manual entry)
+
+---
+
+## Three main applications
+
+| Application | Probe | What you are detecting |
+|---|---|---|
+| **Bearing condition** | Contact / stinger probe on bearing housing | Structure-borne ultrasound from rolling-element wear or inadequate lubrication |
+| **Leak detection** | Airborne horn | Compressed-air, steam-trap, and valve leaks |
+| **Electrical** | Airborne horn | Partial discharge, arcing, and corona in switchgear |
+
+---
+
+## Bearing condition checks
+
+1. Fit the **contact (stinger) probe** firmly to the bearing housing at the designated spot — same location and orientation every visit.
+2. Adjust the instrument's frequency / sensitivity to the manufacturer's recommended setting for rotating-equipment surveys.
+3. Listen through the headset and read the **dB level** displayed.
+4. Record the value; compare to this point's baseline (see Interpretation below).
+
+**Tip:** ultrasound detects incipient bearing faults earlier than vibration or thermography, especially on slow-speed bearings. It is also the primary tool for identifying lack of lubrication before a bearing overheats.
+
+---
+
+## Leak surveys (compressed air / steam)
+
+1. Fit the **airborne horn** (focusing cone).
+2. Sweep the area methodically from a safe distance.
+3. Home in on the direction of the **loudest dB reading** — leaks produce a strong high-frequency hiss.
+4. Once located, move closer to confirm and note the location, dB level, and estimated leak size.
+
+**On a vessel:** cover compressor discharge lines, air receivers, valve glands, steam traps, and flexible hose connections.
+
+---
+
+## Electrical surveys (switchgear / panels)
+
+1. Fit the **airborne horn**.
+2. Hold the instrument near (not inside) the panel or cable tray.
+3. Partial discharge, arcing, and corona all produce characteristic ultrasonic signatures.
+4. Note the dB level and location of any hot spot.
+
+---
+
+## Interpreting bearing readings — trend-based thresholds
+
+Always compare against the **same point's historical baseline**, not an absolute number.
+
+| Rise over baseline | Indication |
+|---|---|
+| < +8 dB | Normal variation — continue monitoring |
+| ~+8 dB (sustained) | Early / incipient wear or lubrication deficiency — investigate, consider re-lubrication |
+| +12–16 dB | Advanced wear — plan maintenance |
+| > +16 dB | Imminent failure — act promptly |
+
+A single elevated reading can be caused by load spikes or probe placement variation. Confirm with a second reading before escalating.
+
+---
+
+## Good technique habits
+
+- Use the **same probe, same spot, same pressure** every visit — variability in contact pressure changes the dB reading.
+- Take readings when the machine is running under **normal operating load**.
+- Note any unusual audible signatures (e.g. grinding, intermittent clicks) that the dB number alone does not capture.
+
+---
+
+## Recording the reading in Mechbase
+
+Open the assigned **Route** → tap the **Measurement Item** for this ultrasound point → the item shows the limit band (minor alert: ~40 dB; major alert: ~50 dB in this demo). Enter the **dB value** displayed on the instrument and confirm.
diff --git a/field-guides/vibration.md b/field-guides/vibration.md
new file mode 100644
index 0000000..bf7552e
--- /dev/null
+++ b/field-guides/vibration.md
@@ -0,0 +1,81 @@
+# Vibration Monitoring — Field Guide
+
+Sensor: **TE Connectivity WVS** wireless tri-axial accelerometer
+
+![TE Connectivity WVS vibration sensor](images/te-wvs-sensor.jpg)
+
+*Note the X/Y/Z axis marker on the hex base (align **X** with the shaft), the
+magnetic twist-mount indicator, the threaded/stud base, and the Ex ia IIC (ATEX/IS)
+rating for hazardous machinery spaces.*
+
+---
+
+## Mounting method — decision tree (best → temporary)
+
+| Method | Use case | Notes |
+|---|---|---|
+| **Stud mount** | Permanent monitoring points | Best frequency response; solid metal-to-metal contact; conveys the full spectrum |
+| **Epoxy mount** | Permanent; no stud possible | 2-part hard-curing epoxy (e.g. Loctite AA 330). Detach magnetic base, epoxy it to machine, re-thread sensor. |
+| **Two-pole magnet** | Curved housings (e.g. motor end caps) | Two feet grip curved surfaces. Gently roll sensor into contact — do not thump it down. Do **not** use a flat magnet directly on a curved surface. |
+| **Flat magnet on glued/soldered metal target** | Painted or curved surfaces where direct contact is impractical | Epoxy or solder a small flat metal target to the machine; mount flat magnet on the target. |
+| **Flat magnet on clean flat surface** | Temporary spot-checks only | Magnets shift on dirty or irregular surfaces. For trend data, prefer a permanent mount. |
+
+---
+
+## Surface preparation
+
+- Remove **all paint and rust** from the mounting area — multiple paint layers and rust severely dampen or block the signal.
+- Vibration must travel through solid, continuous metal; any gap or joint in the path corrupts the reading.
+- Target a **robust, flat, fully cleaned bearing-housing area**.
+
+---
+
+## Where to mount — placement
+
+- Mount as close as feasible to the **monitored bearing**.
+- Target the **load zone**: the section of the bearing housing that carries the rotating shaft — defects show up earliest there.
+
+**On marine machinery take points at:**
+
+1. Motor non-drive end (NDE) bearing housing
+2. Motor drive end (DE) bearing housing
+3. Driven-end bearing (pump / compressor / fan / purifier)
+
+---
+
+## What to avoid
+
+Do **not** mount on:
+
+- Thin cooling fins
+- Plastic covers or fan shrouds
+- Component enclosures
+
+These surfaces resonate at their own frequencies or attenuate the true signal, producing inaccurate data.
+
+---
+
+## Sensor orientation (axis alignment)
+
+The WVS is tri-axial (X / Y / Z axes).
+
+1. Align the **X axis** (marked on the sensor body) with the motor **drive shaft**.
+2. Keep the **same orientation on every visit** and across comparable machines.
+3. Take readings at the **same physical point** each visit.
+
+Inconsistent orientation makes trend analysis unreliable — today's X reading becomes last month's Y.
+
+---
+
+## Taking the reading
+
+1. Mount the sensor at the designated point using the appropriate method above.
+2. Allow a brief settle time after mounting (especially magnetic mounts).
+3. Confirm the sensor is paired and transmitting in the WVS app / Mechbase.
+4. Note the velocity value (mm/s RMS) displayed.
+
+---
+
+## Recording the reading in Mechbase
+
+Open the assigned **Route** on your device → tap the **Measurement Item** for this point → the item shows the limit band (minor / major alert thresholds). Enter the velocity value (mm/s) and confirm. The system records the timestamp, value, and your user ID automatically.
diff --git a/python/examples/seed_marine.py b/python/examples/seed_marine.py
new file mode 100644
index 0000000..ae70ed5
--- /dev/null
+++ b/python/examples/seed_marine.py
@@ -0,0 +1,168 @@
+"""Standalone SDK example: drive the full client flow against a Mechbase instance.
+
+This demonstrates the end-to-end client path — discover the structure, batch-push
+manual point readings (with a couple of curated faults), and run + complete one
+execution of each route — using only the public ``mechbase`` SDK surface.
+
+NOTE: the Aeolian Fortune demo itself is now seeded entirely by mechbase-web's
+``python manage.py bootstrap_marine`` command (structure + measurements + route
+executions, via the ORM). This script is kept as a reusable SDK example; pointed
+at an installation that bootstrap_marine created, it will push additional readings
+and route runs through the live API.
+
+    MECHBASE_TOKEN=<token> MECHBASE_URL=https://app.mechbase.io \
+        python examples/seed_marine.py [--dry-run]
+"""
+from __future__ import annotations
+
+import argparse
+import os
+import random
+
+
+def healthy_reading(transducer_type: str, rng: random.Random) -> dict:
+    if transducer_type == "VB":
+        # mm/s. Must stay below the LOWEST ISO 10816 minor threshold across the
+        # machine classes in the demo — Class I is 1.8 mm/s — so healthy readings
+        # on small (Class I) machines don't falsely trip Minor. Keep margin: <1.5.
+        vel = round(rng.uniform(0.4, 1.4), 3)
+        return {"vel_10hz": vel, "rms": round(rng.uniform(0.3, 0.9), 3),
+                "peak": round(rng.uniform(0.8, 1.6), 3), "crest_factor": round(rng.uniform(1.4, 2.2), 2)}
+    if transducer_type in ("IR", "TM"):
+        return {"value": round(rng.uniform(35.0, 52.0), 1)}
+    if transducer_type == "US":
+        return {"rms": round(rng.uniform(20.0, 36.0), 1)}
+    return {"value": round(rng.uniform(1.0, 10.0), 1)}
+
+
+def fault_reading(kind: str, rng: random.Random) -> tuple[dict, list[dict] | None]:
+    if kind == "vb_minor":
+        return ({"vel_10hz": round(rng.uniform(3.2, 4.5), 3), "rms": round(rng.uniform(0.8, 1.4), 3),
+                 "peak": round(rng.uniform(3.0, 5.0), 3), "crest_factor": round(rng.uniform(2.5, 3.5), 2)}, None)
+    if kind == "ir_minor":
+        return ({"value": round(rng.uniform(68.0, 85.0), 1)}, None)
+    if kind == "ir_major":
+        return ({"value": round(rng.uniform(92.0, 105.0), 1)}, None)
+    raise ValueError(f"unknown fault kind {kind!r}")
+
+
+from mechbase import Mechbase  # noqa: E402
+
+INSTALLATION_NAME = "Aeolian Fortune"
+# Mirror of mechbase-web measurements.marine_demo.MARINE_FAULTS.
+FAULTS = {"AF-SWP2-PDE": "vb_minor", "AF-MSB-BRK": "ir_major", "AF-FOT-IRM": "ir_minor"}
+N_READINGS = 8           # readings per point (weekly cadence, ~60 days)
+DAY_STEP = 7
+
+
+def _iso_days_ago(n: int) -> str:
+    # Lazy import keeps the pure-function tests free of datetime determinism issues.
+    from datetime import datetime, timedelta, timezone
+    return (datetime.now(timezone.utc) - timedelta(days=n)).isoformat()
+
+
+def seed_measurements(inst, points, rng, *, dry_run: bool) -> int:
+    items = []
+    for p in points:
+        if p.transducer_type == "PR":
+            continue
+        fault = FAULTS.get(p.external_id or "")
+        for i in range(N_READINGS):
+            days_ago = (N_READINGS - 1 - i) * DAY_STEP
+            recent = i >= N_READINGS - 2          # fault ramps on the last 2 readings
+            if fault and recent:
+                data, _ = fault_reading(fault, rng)
+            else:
+                data = healthy_reading(p.transducer_type, rng)
+            item = {"measurement_point_external_id": p.external_id, "data": data,
+                    "timestamp": _iso_days_ago(days_ago),
+                    "external_id": f"{p.external_id}-{i}"}
+            items.append(item)
+    if dry_run:
+        print(f"[dry-run] would push {len(items)} readings")
+        return len(items)
+    # The batch endpoint caps at 1000 items per request; chunk to stay under it.
+    chunk = 500
+    created = duplicates = errors = 0
+    for start in range(0, len(items), chunk):
+        result = inst.measurements.create_batch(items[start : start + chunk])
+        created += result.created
+        duplicates += result.duplicates
+        errors += result.errors
+    print(f"readings: created={created} duplicates={duplicates} errors={errors}")
+    return created
+
+
+def run_routes(inst, points_by_id, rng, *, dry_run: bool) -> int:
+    done = 0
+    for route in inst.routes.list(limit=50):
+        detail = inst.routes.get(route.uuid)        # dict: {..., "items": [...]}
+        if dry_run:
+            print(f"[dry-run] would run '{route.name}' ({len(detail['items'])} items)")
+            done += 1
+            continue
+        execution = inst.routes.start(route.uuid)
+        for item in detail["items"]:
+            data = _response_for(item, points_by_id, rng)
+            execution.respond(route_item_uuid=item["uuid"], data=data)
+        execution.complete()
+        done += 1
+        print(f"completed execution of '{route.name}'")
+    return done
+
+
+def _response_for(item: dict, points_by_id: dict, rng: random.Random) -> dict:
+    it = item["item_type"]
+    if it == "pass_fail":
+        passed = rng.random() > 0.1
+        return {"passed": passed} if passed else {"passed": False, "severity": "minor"}
+    if it == "numerical":
+        unit = (item.get("config") or {}).get("unit", "")
+        rng_map = {"°C": (20, 60), "bar": (2, 8)}
+        lo, hi = rng_map.get(unit, (0.5, 100))
+        return {"value": round(rng.uniform(lo, hi), 1)}
+    if it == "multi_check":
+        return {"results": [{"label": c["label"], "passed": True}
+                            for c in (item.get("config") or {}).get("checks", [])]}
+    if it == "measurement":
+        pt = points_by_id.get(item.get("measurement_point_id"))
+        ttype = pt.transducer_type if pt else "TM"
+        return healthy_reading(ttype, rng)
+    return {}
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--dry-run", action="store_true")
+    args = parser.parse_args()
+
+    token = os.environ["MECHBASE_TOKEN"]
+    base = os.environ.get("MECHBASE_URL", "https://app.mechbase.io")
+    rng = random.Random(2014)
+
+    with Mechbase(token=token, base_url=base) as client:
+        me = client.me()
+        inst = client.for_installation(me.current_installation_id)
+        names = {i.installation_id: i.name for i in me.installations}
+        if names.get(me.current_installation_id) != INSTALLATION_NAME:
+            raise SystemExit(
+                f"Refusing to seed: token resolves to '{names.get(me.current_installation_id)}', "
+                f"not '{INSTALLATION_NAME}'. Use the token printed by bootstrap_marine.")
+
+        points = []
+        offset = 0
+        while True:
+            page = inst.measurement_points.list(limit=100, offset=offset)
+            points.extend(page)
+            if len(page) < 100:
+                break
+            offset += 100
+        points_by_id = {p.point_id: p for p in points}
+
+        n = seed_measurements(inst, points, rng, dry_run=args.dry_run)
+        r = run_routes(inst, points_by_id, rng, dry_run=args.dry_run)
+        print(f"Done: {n} readings, {r} route executions on '{INSTALLATION_NAME}' ({base}).")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/python/tests/test_seed_marine.py b/python/tests/test_seed_marine.py
new file mode 100644
index 0000000..25fcfac
--- /dev/null
+++ b/python/tests/test_seed_marine.py
@@ -0,0 +1,41 @@
+# tests/test_seed_marine.py
+import random
+import sys
+import pathlib
+
+sys.path.insert(0, str(pathlib.Path(__file__).resolve().parent.parent / "examples"))
+import seed_marine as sm  # noqa: E402
+
+
+def test_healthy_vb_below_iso_minor():
+    rng = random.Random(1)
+    for _ in range(50):
+        data = sm.healthy_reading("VB", rng)
+        assert "vel_10hz" in data
+        # Below the LOWEST ISO 10816 minor threshold in the demo (Class I = 1.8 mm/s),
+        # so small Class-I machines read healthy, not a false Minor.
+        assert data["vel_10hz"] < 1.8
+
+
+def test_healthy_ir_tm_in_safe_band():
+    rng = random.Random(2)
+    for ttype, ceiling in (("IR", 65.0), ("TM", 55.0)):
+        for _ in range(50):
+            assert sm.healthy_reading(ttype, rng)["value"] < ceiling
+
+
+def test_vb_minor_fault_breaches_minor_not_major():
+    rng = random.Random(3)
+    data, clusters = sm.fault_reading("vb_minor", rng)
+    assert clusters is None
+    assert 2.8 < data["vel_10hz"] < 7.1  # Minor band
+
+
+def test_ir_faults_breach_value_thresholds():
+    rng = random.Random(4)
+    minor_data, minor_clusters = sm.fault_reading("ir_minor", rng)
+    major_data, major_clusters = sm.fault_reading("ir_major", rng)
+    assert minor_clusters is None and major_clusters is None
+    # IR alarm rule: minor=65, major=90 (°C). Minor band is (65, 90]; Major is > 90.
+    assert 65.0 < minor_data["value"] <= 90.0
+    assert major_data["value"] > 90.0