Skip to content

demo-seed-content.md

Latest commit

 

History

History
323 lines (236 loc) · 24.9 KB

demo-seed-content.md

File metadata and controls

323 lines (236 loc) · 24.9 KB
Raw

Demo Seed Content

Note

Companion to demo-plant-design.md (story) and demo-build-spec.md (how). This doc is the literal content the user's seed SQL / Python needs to populate for the demo to read as lived-in rather than canned.

All timestamps are relative (NOW() - INTERVAL '...') so the seed stays fresh across rehearsal days. All IDs are logical references; the user's seed resolves them.

Important

The user owns the actual seed code (per build spec §2.1). This doc is the content reference — 20 specific log entries, 12 specific WOs, 5 specific failures, per-machine signal envelopes. Copy the rows, translate them to whatever seed format you use (raw SQL, asyncpg.executemany, SQLAlchemy, etc.). No runtime dependency; nothing imports this.

Table of contents

  1. Cells + equipment KB (content)
  2. Signal envelopes per machine
  3. 7-day signal history — shape per signal
  4. Work orders — the 12-row backlog
  5. Failure history — 5 recognisable rows
  6. Logbook entries — 20 scripted entries
  7. Shift assignments — 7-day rota
  8. Machine-status transitions
  9. Production events

1. Cells + equipment KB (content)

Five cells. Same parent enterprise / site / area / line (pick any — all five machines sit on one production line).

# Cell name equipment.equipment_type equipment.manufacturer equipment.model installation_date onboarding_complete
1 Source Pump Centrifugal Pump Grundfos CR 64-3-2 NOW() - INTERVAL '22 months' TRUE
2 UV Sterilizer UV Reactor Trojan UV3000Plus NOW() - INTERVAL '14 months' TRUE
3 Bottle Filler Centrifugal Pump Grundfos CR 32-2 NOW() - INTERVAL '18 months' TRUE
4 Bottle Capper Servo Drive SEW-Eurodrive DRN90L4BE5 NOW() - INTERVAL '16 months' TRUE
5 Bottle Labeler Labeler Krones Contiroll NOW() - INTERVAL '2 months' FALSE

Notes per cell (equipment_kb.notes):

  • Source Pump — "Pumps raw water from the well to the pre-treatment tank. 24/7 service."
  • UV Sterilizer — "Kills bacteria and viruses in the incoming water with UV light."
  • Bottle Filler — "Fills empty bottles with clean water. Replaces both bearings on a 12-month PM cycle."
  • Bottle Capper — "Screws caps onto filled bottles. Drive motor replaced 16 months ago."
  • Bottle Labeler — "New machine — not yet onboarded. Upload the IOM PDF to calibrate."

2. Signal envelopes per machine

The thresholds land in equipment_kb.structured_data.thresholds.<kb_threshold_key>. Shapes follow the existing P-02 KB precedent: {nominal, alert, trip, unit, source, confidence} for single-sided bounds, {nominal, low_alert, high_alert, unit, source, confidence} for double-sided.

2.1 Source Pump

Signal kb_threshold_key Display name Unit Nominal Alert Trip
Motor current motor_current_a Motor current A 12.0 18.0 22.0
Discharge pressure pressure_bar Water pressure bar 4.8 low_alert 4.0 / high_alert 5.8 —
Flow flow_l_min Water flow L/min 820 low_alert 720 / high_alert 900 —
Vibration vibration_mm_s Motor shake mm/s 2.5 4.5 7.1

2.2 UV Sterilizer

Signal kb_threshold_key Display name Unit Nominal Alert Trip
UV intensity uv_intensity_mw_cm2 UV lamp brightness mW/cm² 28.0 low_alert 22.0 low_trip 18.0
Lamp runtime uv_runtime_h UV lamp hours h 4500 7500 9000
Flow flow_l_min Water flow L/min 820 low_alert 720 / high_alert 900 —
Motor current (drive) motor_current_a Motor current A 3.2 5.0 6.5

2.3 Bottle Filler (demo star)

Matches the current P-02 KB. Keep these values stable — the Investigator's numerical outputs cite them verbatim.

Signal kb_threshold_key Display name Unit Nominal Alert Trip
Vibration vibration_mm_s Motor shake mm/s 2.2 4.5 7.1
Bearing temp bearing_temp_c Bearing temp °C 48 75 90
Discharge pressure pressure_bar Water pressure bar 5.5 low_alert 4.5 / high_alert 6.5 —
Flow flow_l_min Water flow L/min 533 low_alert 480 / high_alert 580 —
Bottles per minute bottles_per_minute Bottles per minute /min 180 low_alert 150 —

Bearing geometry (keep in failure_patterns[0].signal_signature per §5): bearing_reference=6206, n_balls=9, pitch_diameter_mm=46.0, ball_diameter_mm=9.5, contact_angle_deg=0, shaft_rpm_nominal=2900.

2.4 Bottle Capper (memory-scene target)

Signal kb_threshold_key Display name Unit Nominal Alert Trip
Drive vibration vibration_mm_s Motor shake mm/s 1.8 4.2 6.5
Cap torque cap_torque_nm Cap tightness N·m 3.5 low_alert 2.8 / high_alert 4.2 —
Motor current motor_current_a Motor current A 4.1 6.0 7.5
Jam counter (per hour) jam_events_per_h Jams per hour /h 0 3 8

Bearing geometry (Capper drive motor — typical 6203 bearing): bearing_reference=6203, n_balls=8, pitch_diameter_mm=29.0, ball_diameter_mm=6.75, shaft_rpm_nominal=1450.

2.5 Bottle Labeler

No signal definitions until the operator onboards it. That is the whole point of the onboarding-wizard scene. Leave this cell's process_signal_definition empty; leave equipment_kb with onboarding_complete=FALSE and minimal structured_data ({equipment: {...}, kb_meta: {onboarding_complete: false, completeness_score: 0.1}}).

3. 7-day signal history — shape per signal

Volume target: ~350 000 rows (7 d × 30 s sample period × ~18 monitored signals across 4 cells = ~350 k). Skip the Labeler (no signals).

3.1 Generator recipe

For each (cell_id, signal_def_id) pair:

value(t) = clamp(
    nominal
    + seasonal(t)          # day/night sinusoidal bias
    + ar1_noise(t)         # mean-reverting noise around zero
    + shift_change_pulse(t), # short bump at each shift boundary
    low_bound, high_bound
)

Parameters (per signal class):

Parameter Motor current Vibration Pressure Flow UV intensity Bearing temp Bottles/min Cap torque
seasonal_amplitude (% of nominal) 4% 3% 2% 6% 2% 5% 8% 2%
seasonal_period_h 24 24 24 24 24 24 24 24
ar1_phi (mean-reversion) 0.90 0.88 0.95 0.92 0.97 0.93 0.85 0.90
ar1_sigma (% of nominal) 1.5% 2.5% 1.0% 2.0% 0.5% 1.5% 3.0% 1.5%
shift_change_pulse (% of nominal, triangular ~5 min) 2% 3% 1% 3% 1% 1% 4% 2%

Determinism: seed the RNG from hashlib.blake2b(f"{cell_id}-{signal_def_id}".encode()).digest()[:16]. Every make db.reset → make <seed> pair produces an identical history — critical for reproducible dry-runs.

3.2 Mandatory clamp on the last 6 hours

For every monitored signal, the last 6 h of history must read as abs(slope) / reference < 0.005/h — forecast-watch's drift floor. Enforce via a post-hoc pass:

for each (cell, signal):
    last_6h = history[-6h:]
    slope, _ = np.polyfit(hours, last_6h, 1)
    reference = abs(mean(last_6h))
    if abs(slope) / max(1e-6, reference) > 0.005:
        # replace last-6h with pure mean-reverting noise around last_6h.mean()

Without this clamp, forecast-watch fires on seeded drift instead of the drift seed-forecast injects — the predictive banner fires immediately on stack boot, which is wrong and also defeats the demo endpoint.

Unit test to assert: SELECT signal_def_id FROM process_signal_data WHERE time > NOW() - INTERVAL '6 hours' GROUP BY signal_def_id HAVING regr_slope(raw_value, extract(epoch from time)/3600) / nullif(abs(avg(raw_value)), 0) > 0.005 must return zero rows.

4. Work orders — the 12-row backlog

Twelve rows. Mix: closed/cancelled/open, agent/manual, across all four monitored cells. All times are NOW() - INTERVAL '<X> days' offsets so the seed stays fresh.

# Cell Created Status Priority Title Generated by agent rca_summary (short)
1 Bottle Filler 6 d 04 h ago completed high Bearing replacement — Filler pump TRUE "Sandbox: slope_per_hour=0.019, r_squared=0.94, eta_to_trip_hours=3.1. Bearing wear near end-of-life replaced under scheduled PM."
2 Source Pump 5 d 10 h ago completed medium Replace flow sensor calibration drift FALSE "Operator-initiated recalibration after logbook note from Priya. Field team recalibrated in 40 minutes."
3 Bottle Capper 4 d 18 h ago completed medium Cap torque alarm — reseated jammed cap TRUE "Sandbox: rho_vibration_cap_torque=0.71, n_samples=14400. Jammed cap cleared; torque returned to nominal within 8 minutes."
4 UV Sterilizer 3 d 22 h ago completed low Lamp hours exceeded 7500 — replace scheduled TRUE "UV lamp approaching end of rated life; not an immediate failure. Replacement deferred to next planned maintenance window."
5 Bottle Filler 3 d 05 h ago cancelled medium False alarm — anti-surge valve cycled TRUE "Transient pressure spike traced to a downstream valve cycle, not the Filler. Cancelled without action."
6 Source Pump 2 d 14 h ago completed high Seal replacement after flow drop FALSE "Manual flag from operator — mechanical seal showed wet trace. Replaced seal kit and verified normal flow."
7 Bottle Capper 2 d 02 h ago completed low Routine — change-over between cap sizes FALSE "Line change-over from 500 ml to 1.5 L caps. 30 min. No incident."
8 Bottle Filler 1 d 09 h ago completed medium Vibration nudge — recentred impeller TRUE "Sandbox: slope_per_hour=0.008, r_squared=0.62, eta_to_trip_hours=18.4. Mild vibration uptrend; preventive impeller re-centring."
9 UV Sterilizer 1 d 05 h ago open medium Replace UV lamp 2 of 4 FALSE "Scheduled replacement — lamp #2 at 7800 hours. Field team arrives 09:00 tomorrow."
10 Source Pump 1 d 02 h ago open low Visual inspection — scheduled quarterly FALSE "Walk-around + bolt torque check. Due Friday."
11 Bottle Filler 20 h ago cancelled high Spurious alarm — clock skew on simulator? TRUE "Threshold breach flagged during a sampling gap. Investigator found no physical evidence. Cancelled pending review."
12 Bottle Capper 14 h ago in_progress medium Torque drift — bench-test spindle FALSE "Night-shift observed intermittent torque variance. Day-shift technician running bench test."

Populated fields per WO (beyond the table above):

  • triggered_by_signal_def_id — for agent-generated rows (#1, 3, 4, 5, 8, 11), set to the cell's vibration or pressure signal_def.id; for manual rows (#2, 6, 7, 9, 10, 12) leave NULL.
  • completed_at — set for completed/cancelled rows only, created_at + INTERVAL '<duration>' with realistic durations (30 min-5 h).
  • recommended_actions — a ~3-item JSON array of short strings (e.g. ["Replace upper bearing", "Verify alignment", "Re-torque coupling bolts"]). Keep it plausibly-pump-ish for rows 1/3/5/6/8/11; routine for the rest.
  • created_by — operator username for manual rows (rotate through sarah.miller, marco.ferrari, priya.patel); leave NULL for agent-generated rows.

Constraint check: no WO must have status IN ('detected', 'analyzed', 'open', 'in_progress') on the Bottle Filler's vibration signal at seed time, or Sentinel's 30-min per-(cell, signal) debounce will swallow the first demo anomaly. WO #12 is on the Capper's torque drive, not vibration, so it's safe.

5. Failure history — 5 recognisable rows

Five rows spread across six months. Each row carries a signal_patterns jsonb the Investigator's memory-scene matches against.

# Cell failure_time resolved_time failure_mode root_cause signal_patterns
1 Bottle Filler NOW() - INTERVAL '3 months' + INTERVAL '4 hours' bearing_wear "Discharge bearing wear near end-of-life — replaced under PM-2026-01-18." {"vibration_mm_s": {"peak": 5.4, "duration_min": 14, "slope_per_hour": 0.024}, "bearing_temp_c": {"peak": 78, "slope_per_hour": 0.4}}
2 Bottle Filler NOW() - INTERVAL '6 months' + INTERVAL '6 hours' mechanical_seal_leak "Seal face wear; replaced mechanical seal kit and flushed seal chamber." {"flow_l_min": {"drop_from": 533, "drop_to": 488}, "pressure_bar": {"oscillation_amplitude": 0.6}}
3 Bottle Capper NOW() - INTERVAL '3 months' + INTERVAL '4 hours' bearing_wear "Drive bearing replaced. Same pattern as Filler's Jan incident — both pump-side rotating kit." {"vibration_mm_s": {"peak": 4.8, "duration_min": 22, "slope_per_hour": 0.018}, "cap_torque_nm": {"rho_with_vibration": 0.71}}
4 Source Pump NOW() - INTERVAL '4 months' + INTERVAL '2 hours' impeller_imbalance "Cavitation damage on impeller vanes; inspected and re-balanced, no replacement." {"vibration_mm_s": {"spike_peak": 5.2, "return_to_nominal_min": 3}}
5 UV Sterilizer NOW() - INTERVAL '5 months' + INTERVAL '3 hours' lamp_replacement "UV lamp #3 reached end-of-life. Replaced with Trojan P/N UV3K-LAMP-440W; all 4 banks verified." {"uv_intensity_mw_cm2": {"drop_from": 28, "drop_to": 19}, "uv_runtime_h": {"at_failure": 7640}}

Why row 3 specifically: The memory-scene beat narrates "ARIA saw this exact pattern on the Capper in January." Row 3's failure_time = NOW() - 3 months anchored to today means it reads as "roughly January" whenever the demo runs — adjust the interval if your recording date is already far from April (e.g. on a July demo, use NOW() - INTERVAL '5 months'). The signal_patterns dict must reference both vibration AND a coupled signal (torque) so the Investigator's coupling-class rule fires.

6. Logbook entries — 20 scripted entries

All times relative to NOW(). Rotate authors through sarah.miller (day), marco.ferrari (evening), priya.patel (night), tom.anderson (supervisor, occasional).

# entry_time Cell author_username category severity Content
1 NOW() - 6 d 22 h Bottle Filler sarah.miller observation info "Quiet morning. Filler humming along. Bottles per minute holding at 180."
2 NOW() - 6 d 14 h UV Sterilizer marco.ferrari maintenance info "Lamp #2 runtime counter reset after PM. Back to zero. All four lamps confirmed lit."
3 NOW() - 5 d 20 h Source Pump priya.patel observation warning "Flow dropped briefly (~720 L/min) around 02:30, recovered in 90 seconds. Logged just in case."
4 NOW() - 5 d 10 h Source Pump sarah.miller incident warning "Reached 02:30 note — field tech recalibrated upstream flow sensor (drift ~4%). Back to spec."
5 NOW() - 4 d 21 h Bottle Capper priya.patel incident warning "Cap jam at 04:10, resolved 04:18. Usual suspect — cap rim deformity. Two bottles rejected."
6 NOW() - 4 d 14 h UV Sterilizer marco.ferrari observation info "UV intensity across all four banks inside green band. No action."
7 NOW() - 3 d 23 h Bottle Filler priya.patel observation info "Filler a bit noisier than usual around 03:00. Not above the limit, just noticing. Probably nothing."
8 NOW() - 3 d 15 h UV Sterilizer tom.anderson maintenance info "Planned — lamp #1 runtime 7500h alarm. Scheduled replacement for next Wednesday."
9 NOW() - 3 d 05 h Bottle Filler sarah.miller incident warning "Transient pressure spike, 6.8 bar for 4 seconds. ARIA flagged. Cancelled after no correlation with downstream."
10 NOW() - 2 d 16 h Source Pump sarah.miller incident critical "Seal leak — water trace on the mechanical seal housing. Called field. Seal replaced in 2 hours."
11 NOW() - 2 d 11 h Bottle Capper marco.ferrari changeover info "Line change-over 500 ml → 1.5 L caps. Capper torque spec updated from 3.5 to 4.1 N·m. 30 min downtime."
12 NOW() - 1 d 22 h Bottle Filler priya.patel observation warning "Filler ran rough for about ten minutes after shift change. Sounded like a pulley. Settled on its own. Will watch next shift."
13 NOW() - 1 d 20 h Bottle Filler priya.patel incident warning "Vibration uptrend continues. ARIA opened a WO. Scheduling impeller re-centre for next maintenance window."
14 NOW() - 1 d 13 h UV Sterilizer sarah.miller maintenance info "Verified lamp #2 runtime still under 7900. Holding for Wednesday."
15 NOW() - 1 d 06 h Bottle Capper marco.ferrari observation info "Line running clean at 180 bpm. No jams past two shifts."
16 NOW() - 18 h Source Pump priya.patel observation info "Nothing to report from the pump — steady flow."
17 NOW() - 14 h Bottle Capper priya.patel incident warning "Torque readout bouncing between 3.2 and 3.9 N·m. No audible jam. Will run a bench test on the spindle day-shift."
18 NOW() - 10 h Bottle Capper sarah.miller maintenance info "Bench test on capper spindle started. Running baseline for 2h then loading."
19 NOW() - 5 h Bottle Filler sarah.miller observation info "Filler motor shake back to ~3.1 mm/s after the impeller nudge. Feels right."
20 NOW() - 2 h UV Sterilizer sarah.miller observation info "Pre-replacement check on lamp #2 — intensity 27.8 mW/cm². Holding steady."

Plot-thread entries (12 → 13 → 19) lay down the Filler story that row 1's RCA + row 12/13 logbook + row 19 post-action check all triangulate on. The Investigator's memory-scene narration can cite entry 12 by author and timestamp — "Priya's note at 02:15 flagged the Filler running rough…".

7. Shift assignments — 7-day rota

21 rows (7 days × 3 shifts). Plus one extra for the current shift anchor.

Rotation rule (A-B-C pattern with weekends running the same):

Day offset Day shift (06-14) Evening (14-22) Night (22-06)
-6 d Sarah Miller Marco Ferrari Priya Patel
-5 d Sarah Miller Marco Ferrari Priya Patel
-4 d Sarah Miller Marco Ferrari Priya Patel
-3 d Sarah Miller Marco Ferrari Priya Patel
-2 d Sarah Miller Marco Ferrari Priya Patel
-1 d Sarah Miller Marco Ferrari Priya Patel
today Sarah Miller Marco Ferrari Priya Patel ← current shift at demo time

Note

The rota is deliberately not rotated. Keeping the same operator on the same shift every day makes the memory-scene pronoun ("Priya was on the night shift…") work regardless of what time the demo runs — a rotating rota would force the narration to branch on the calendar date.

Tom Anderson is the supervisor. He is NOT in shift_assignment rows (that table is for on-shift operators). Instead, he appears as created_by on a couple of manual WOs (#8 in §4, for example) and in one logbook entry (#8 in §6). The Shift page's header can show him as "Supervisor on call: Tom Anderson" if desired; that string is a presentation-layer decision, not a table row.

Shift table (if not already seeded): 3 rows — Day/Evening/Night with start_time/end_time 06:00-14:00, 14:00-22:00, 22:00-06:00.

8. Machine-status transitions

One maintenance window per machine in the last 7 d, plus a couple of transient faults to break up the monotony. Shape: rows in machine_status with {cell_id, time, state_code}. Typical states: running, maintenance, fault, changeover.

Cell When State Duration Why
Source Pump ~5 d 10 h ago fault → running 40 min Flow sensor recalibration (from WO #2 + logbook #4)
Source Pump ~2 d 14 h ago fault → maintenance → running 2 h 15 min Seal replacement (WO #6)
UV Sterilizer ~3 d 15 h ago running — Tom's note about scheduled lamp swap — no downtime yet
Bottle Filler ~6 d 04 h ago maintenance → running 4 h Bearing PM (WO #1)
Bottle Filler ~1 d 09 h ago running — Impeller nudge (WO #8); no state change, runtime work
Bottle Capper ~4 d 18 h ago fault → running 8 min Cap jam (WO #3 + logbook #5)
Bottle Capper ~2 d 11 h ago changeover → running 30 min 500 ml → 1.5 L change-over (WO #7 + logbook #11)
Bottle Capper ~10 h ago running — Bench test, still on spindle; line kept running on spare spindle

All other time: running. Generating these is a matter of emitting one machine_status row at each transition boundary.

9. Production events

Target: ~2 000 rows in production_event across the 7-day window, distributed realistically so OEE computation reads healthy.

Distribution per monitored cell:

Cell Events/shift (avg) Typical event types
Source Pump 10 run_hour (1/h) + occasional low_flow_transient
UV Sterilizer 6 run_hour + lamp_hour_tick (each hour)
Bottle Filler 25 bottle_batch (300 bottles) every ~8 min, + changeover_complete
Bottle Capper 25 bottle_batch (300 bottles) every ~8 min (mirrors Filler), + occasional jam_clear

Yield/quality target aggregating across the 7 days:

  • ~55 000 bottles produced total (realistic for a 5-machine bottling line running 24/7)
  • 99.3% first-pass yield (380 rejected bottles logged as quality_fail)
  • OEE computation ends up around 78-85% depending on how the user's KPI query weights availability vs performance vs quality

Correlation with WO downtimes: during the Source Pump seal incident (~2 d 14 h ago, 2 h 15 min), the whole line stopped (no bottles produced downstream). Tag those 2 rows in production_event as line_halt_due_to_upstream. Without this correlation the KPI chart will show bottles being produced during a documented outage — sharp-eyed judge could spot it.

10. Final integrity checks

Before declaring the seed "done", run these SQL assertions. Each one catches a specific failure mode I've seen hurt demos:

Check Query (schematic) Expected
No current-tense breach SELECT count(*) FROM process_signal_data psd JOIN process_signal_definition psd_def ON psd_def.id=psd.signal_def_id JOIN equipment_kb k ON k.cell_id=psd_def.cell_id WHERE psd.time > NOW() - INTERVAL '5 min' AND raw_value > (k.structured_data -> 'thresholds' -> psd_def.kb_threshold_key ->> 'alert')::float 0
No debounce-blocking WO on Filler vibration SELECT count(*) FROM work_order w WHERE cell_id=<filler_id> AND triggered_by_signal_def_id=<filler_vib_id> AND status NOT IN ('completed','cancelled') AND created_at > NOW() - INTERVAL '35 min' 0
No drift in last 6h (all signals) see §3.2 for exact query 0 rows
Shift assignments cover current time SELECT count(*) FROM shift_assignment WHERE assigned_date = CURRENT_DATE 3 (one per shift)
Memory-scene target has a 3-month-old failure SELECT failure_mode FROM failure_history WHERE cell_id=<capper_id> AND failure_time BETWEEN NOW() - INTERVAL '4 months' AND NOW() - INTERVAL '2 months' 'bearing_wear'
KPI query returns a non-null number SELECT oee FROM kpi_oee_view WHERE cell_id=<filler_id> (adapt to actual view name) not null, in [0,1]

A single failure here means the demo fails in a specific rehearsal-visible way. Fix at seed time, not at rehearsal time.

11. How to use this doc

  1. Read §1-§9 once.
  2. Translate each table into your seed's native format (raw SQL / asyncpg.executemany / SQLAlchemy / whatever).
  3. Run make db.reset && make <your-seed-target>.
  4. Run the §10 integrity checks. Fix any failures.
  5. Proceed to dry-run per demo-build-spec.md §3.

Nothing in this doc is imported or consumed at runtime — it is a content reference you translate once and forget.

Tip

The two most demo-visible pieces are logbook entry #12 (Priya's "Filler ran rough" note, which the Investigator's memory-scene narration cites by author + timestamp) and failure_history row #3 (the Capper's 3-month-old bearing-wear pattern that the memory-scene PatternMatch card renders against). If you short-change any row, short-change the production_event rows, never these two.