Lear · Body 2 · 31XX-2 · from one-off study to standing measure

Work Content Monitor

The discovery study, running continuously — measuring where the work actually sits, on the cameras you already have.

The study was a snapshot: two hours, twelve stations, one product mix. Running it continuously turns a finding into a control. Same cameras, same measurement, four clock speeds — from a supervisor interrupted in ninety seconds to a quarterly re-baseline of the whole line.

What it watches

four signals · everything else is derived from these

Signal 01

Is the line moving?

Conveyor displacement, measured directly from the image. No model, no training, no per-station setup — it works the moment a camera is connected.

✓ measured at R² 0.999 over 10 minutes · one camera covers the whole loop

Signal 02

Where is each board?

Boards entering and leaving each station, and how long they stay. Where boards carry a printed tag it also reads the identity, so a unit can be followed around the loop.

✓ station-to-station transit measured at 120 s, constant across 30 minutes

Signal 03

Who is working, not just present

An operator counts as working when their hands are down in the board — judged from their own posture, so it does not depend on where the standard work sheet places them.

✓ 0.28 operators mean error, within one on 97% of frames, against 36 hand-scored frames

Signal 04

Which way operators leave

An operator who follows a board out of their station could not finish inside the pitch. Counting that gives the chase index — the line's own verdict on whether the work fits.

✓ direction confirmed as behaviour, not camera angle (r = −0.01 against position)

Note what is not on this list: the standard work sheet. Nothing here is measured against a plan. That is deliberate — it is what makes the output a reality check rather than a compliance score, and it means the measurement stays valid when the plan changes and nobody updates the document.

Insights Schedule

every output, its cadence, and who it goes to

The same four signals feed every horizon. What changes is how far back you stand — and the further back you stand, the less it looks like an alert and the more it looks like an engineering decision. Only four things interrupt a person; everything else waits for a report.

Cadence	Type	Output	Goes to	Trigger / contents
Liveseconds	Alert	Nobody working the board Board present, no hands on it. Needs no crew-plan data.	Line supervisor	sustained 45 s
	Alert	Board holding Fires while the board is still in front of the operator.	Line supervisor	> 1.5 × 120 s
	Alert	Line down, unattended Not the stop — the stop nobody has answered.	Line supervisor	4 min, no response
	Alert	Downtime budget three-quarters spent The one number the floor cannot know by looking.	Shift lead	75% of ~82 min
Shiftevery 8 h	Report	Shift sheet One page. Boards vs takt requirement, downtime vs budget, and per station: engaged, idle, time at plan, chase index — with exceptions flagged.	Shift lead Production mgr	see prototype A
Shiftevery 8 h	Report	Same-shift-last-week comparison Drift against your own history, not against a target.	Production mgr	auto-attached
WeeklyMondays	Report	Balance chart Work content per board per station against person-minutes available. The sheet a rebalancing decision is made from.	Industrial engineer	see prototype B
WeeklyMondays	Insight	Chase index trend A station drifting out of balance before anyone reports a problem.	Industrial engineer	4-week rolling
On changeas needed	Report	Before / after Any work-element move, speed change or layout change gets the same chart on both sides of it.	Industrial engineer OpEx	see prototype C
On changeas needed	Insight	Work content by product variant A variant that overloads one station gets its own line rather than being averaged away.	Industrial engineer	per mix run
Quarterlyre-baseline	Insight	Line trend Is the spread between stations closing? Did the changes hold?	Plant manager OpEx	rolling quarter
Quarterlyre-baseline	Alert	Target at risk By exception only — when the shift trend puts the weekly number at risk.	Plant manager	exception only

What deliberately does not alert. Instantaneous headcount — not precise enough on a single frame, and presence is a poor proxy for work anyway. Idle share — a real finding, but a conversation about work content, not something a supervisor fixes in two minutes. Line-wide totals — no owner on the floor. And nothing at all goes to the industrial engineer in real time: paging an engineer about a two-minute event trains them to ignore the system.

What it does and does not do

stated plainly, so nobody is surprised in month three

It does

Measure work content per station per board, per product mix, without standard times
Tell you which stations cannot finish inside the pitch, and which have room
Count every stoppage, including the short ones nobody logs
Run on the cameras already installed, on CPU, with no line time to set up

It does not

Judge quality, or whether the work was done correctly
Identify individuals, or track a named person between stations
Replace a time study — it says where to look, not what the element times should be
Work where a camera cannot see the operator's upper body at the board

Setup is the honest cost. Signals 1 and 2 work the day a camera is connected. Signals 3 and 4 need each station's work area marked and the measurement checked against what a person sees on that camera — that step is where accuracy is won or lost, and in the study it took more effort than the analysis did. Budget for it as commissioning, not configuration.

What the first quarter looks like

from connection to a decision you can defend

Weeks 1–2 · connect and commission. Cameras connected, work areas marked, engagement measurement checked station by station against what an engineer sees on the same footage. Line motion and board tracking are live from day one.
Weeks 3–6 · baseline across mixes. Work content per station for each product variant actually run. One evening of one mix — which is what the study had — cannot tell you whether a station's load is a property of the station or of the harness.
Weeks 7–8 · first move. Take the widest neighbour gap, move a work element, and measure both sides of it. A small, reversible change whose effect the system can prove.
Weeks 9–12 · alerts on. Only once the supervisor has agreed the thresholds. The number that decides how often these fire is how long a station may run short-handed before it is worth interrupting someone — and that is their call, not ours.

Addendum · report prototypes

built from the 25 August dataset · real numbers, not mock-ups

Each report below is generated from the two hours actually measured. Where a report needs history the study does not have — week-on-week, quarter-on-quarter — the format is shown against the two hours we do have, and labelled as such.

Prototype A · Shift sheet

one page, issued at end of shift · 6 PM hour shown as a shift block

LineBody 2 · 31XX-2

Block25 Aug · 18:00–19:00

Boards built20 vs 17.1 required

Downtime0 min of ~82 budget

At planned crew28%

Station	Engaged p-min	Idle p-min	Idle %	At plan	Chase	Exception
B2C5	98	22	18%	32%	14.2	work not fitting
B2C6	74	9	11%	28%	12.3
B2C7	28	4	11%	21%	17.3	work not fitting
B2C8	102	14	12%	35%	2.2
B2C10	93	10	10%	35%	14.8	work not fitting
B2C11	58	20	25%	22%	10.9	camera view — see note
B2C12	39	10	21%	13%	3.9	spare capacity
B2C13	71	46	39%	26%	12.1	high idle — check
B2C14	99	34	26%	35%	6.0
B2C15	86	22	21%	23%	3.9
B2C17	82	57	41%	27%	9.2	high idle — check
B2C18	92	19	17%	35%	12.1
TOTAL	922	267	22%	28%	—	4 exceptions

Exceptions only — a shift lead reads the four flagged rows, not the twelve. Chase index above 14 or idle above 35% raises a flag; both thresholds are set with the supervisor, not by us.

Prototype B · Weekly balance chart

work content per board against what is available · the sheet a rebalancing decision is made from

B2C14

5.53

B2C17

4.25

B2C15

4.23

B2C10

4.17

B2C18

4.15

B2C8

4.12

B2C13

3.76

B2C6

3.61

B2C7

3.23

B2C5

3.20

B2C11

2.78

B2C12

2.10

5.97 person-minutes available per board at the current 179 s pitch (85% of the axis); 7.03 if run at the 211 s takt. Widest neighbour gap: B2C12 ↔ B2C13, 1.66 p-min — the recommended first move. Orange = chasing boards downstream, green = works ahead.

Prototype C · Before / after

format shown against the two hours measured · not a real intervention

Station	Period A engaged	Period B engaged	Change	A idle	B idle	Movement
B2C7	28	66	+39	4	8	loaded up sharply
B2C13	71	81	+10	46	52
B2C17	82	90	+8	57	56
B2C12	39	46	+6	10	12
B2C15	86	86	0	22	26
B2C6	74	71	−3	9	9
B2C11	58	54	−4	20	30
B2C18	92	76	−16	19	21
B2C10	93	76	−17	10	11
B2C8	102	64	−39	14	8	unloaded sharply
B2C5	98	31	−67	22	19	unloaded sharply
LINE	922	740	−182	267	251	−20% labour

Periods A and B here are simply the 6 PM and 7 PM hours — no change was made between them. It is shown to demonstrate the format, and because it surfaces something real: the line ran 20% leaner in the second hour, with three stations swinging by more than 35 person-minutes. That is exactly the signature this report is designed to catch, and nobody has yet explained it.

Two of these prototypes carry a caveat from the study. B2C11's camera views a walkway rather than its work area, so its numbers are flagged rather than trusted. And every figure here comes from one evening of one product mix — the weekly chart only becomes a balance decision once several mixes have been measured.