N+1 LabsFactory N+1

Build risk reportSample

Synthetic coin-cell tracker puck: where the next build loses units

S4 (press cover) scraps 151 of the 172 units the line loses out of 5,000. The largest risk, no-engage / loose at S4, falls from 70 to 1 unit when base diameter goes from 29.80 ±0.08 mm to 29.88 ±0.08 mm.

SIMULATEDUNCALIBRATEDSYNTHETIC PARTASSUMPTION Engine output on a synthetic product; nothing is measured. ASSUMPTION marks equipment values and drawn shapes.

1 Line summary SIMULATEDUNCALIBRATEDSYNTHETIC PART

Units run

5,000

seed 42, S1 to S5

First-pass yield

96.30%

185 fail first pass

Yield after retries

96.56%

13 recovered by retry

Bottleneck

462/h, assumed timing

SYNTHETIC PARTASSUMPTION Drawn to scale from the spec's nominal dimensions. Labels mark the feature each fix changes. Hook and rib shapes are ASSUMPTIONS.

ASSUMPTIONIllustrative plan from the bench layout, station spacing to scale, footprints not. 24 pallets at 12 m/min. Losses are after retries.

2 Failing stations, ranked by units lost SIMULATEDUNCALIBRATEDSYNTHETIC PART

Rank	Station	Of units run	Risks: failure class, units lost (range over seeds 42 to 46)
1	S4 press cover	3.02%	R1 No-engage / loose 70 (55 to 70)R2 Jam / misalignment 61 (54 to 63)R3 Stack interference 20 (15 to 20)
2	S5 inspect gap, power test	0.40%	R4 Gap uneven 20 (18 to 39)
3	S3 insert cell	0.02%	R5 Cell not seated 1 (0 to 1)

3 Risks, cause and fix

The cause is the engine's model of the failure. The fix is the tested change that leaves the most good units, for the first build to confirm. Station errors are 3 sigma.

R1S4 No-engage / loose70 units lost (1.40%); 55 to 70 over seedsSIMULATEDUNCALIBRATEDSYNTHETIC PART

Cause. A hook misses its window, engages too little, has too much axial play, or the press stops before it snaps.

Top drivers	Input	First-pass units with this failure → driver held at nominal	Direction
base diameter	29.80 ±0.08 mm SYNTHETIC	70 → 5	larger is safer
cover inner diameter	30.10 ±0.08 mm SYNTHETIC	70 → 6	smaller is safer
base hooks undercut	0.45 ±0.04 mm SYNTHETIC	70 → 26	spread only

Fix. Best of 7 what-if reruns: move the nominal of base diameter, 29.80 ±0.08 mm to 29.88 ±0.08 mm SYNTHETIC +74 good units. Tied: cover inner diameter, 30.10 ±0.08 mm to 30.02 ±0.08 mm, leaves as many.

Rerunthis failure 70 → 1all lost at S4 151 → 75first pass 96.30% → 97.78%after retries 96.56% → 98.04%

Trade-off. The same rerun loses more units to hook overstrain (crack risk) at S4 (+2); gap uneven at S5 (+2). Check these before choosing this fix.

First build. Measure hook-face diameter and window position on first shots; record press curves. Owner: design and tooling supplier ASSUMPTION

R2S4 Jam / misalignment61 units lost (1.22%); 54 to 63 over seedsSIMULATEDUNCALIBRATEDSYNTHETIC PART

Cause. The cover's lead-in does not capture the base, or a hook lands outside its window along the rim.

Top drivers	Input	First-pass units with this failure → driver held at nominal	Direction
S4 gripper part offset	±0.10 mm ASSUMPTION	61 → 3	smaller is safer
S4 gripper rotation offset	±0.30 deg ASSUMPTION	61 → 9	spread only
pallet locating error	±0.05 mm ASSUMPTION	61 → 32	smaller is safer

Fix. Best of 2 what-if reruns: specify S4 gripper part offset, ±0.10 mm to ±0.05 mm ASSUMPTION +48 good units.

Rerunthis failure 61 → 13all lost at S4 151 → 103first pass 96.30% → 97.26%after retries 96.56% → 97.52%

First build. Measure cover pose on the gripper and base pose on the pallet separately. Owner: manufacturing engineering ASSUMPTION

R3S4 Stack interference20 units lost (0.40%); 15 to 20 over seedsSIMULATEDUNCALIBRATEDSYNTHETIC PART

Cause. A cover rib presses on the stack below it (cell or board) by more than its allowed crush, so the cover cannot seat.

Top drivers	Input	First-pass units with this failure → driver held at nominal	Direction
cell height	2.9 to 3.2 mm TYPICAL	20 → 0	smaller is safer
pcba thickness	1.00 ±0.08 mm ASSUMPTION	20 → 1	smaller is safer
cover cell ribs reach	1.60 ±0.04 mm SYNTHETIC	20 → 12	smaller is safer

Fix. Best of 4 what-if reruns: move the nominal of cover cell ribs reach, 1.60 ±0.04 mm to 1.56 ±0.04 mm SYNTHETIC +25 good units.

Rerunthis failure 20 → 0all lost at S4 151 → 131first pass 96.30% → 96.80%after retries 96.56% → 97.06%

First build. Measure the seated board and cell stack and the rib reach on first-shot covers. Owner: mechanical design ASSUMPTION

R4S5 Gap uneven20 units lost (0.40%); 18 to 39 over seedsSIMULATEDUNCALIBRATEDSYNTHETIC PART

Cause. The camera's cover-to-base gap varies around the part by more than the spread limit.

Top drivers	Input	First-pass units with this failure → driver held at nominal	Direction
base hooks catch depth	2.20 ±0.06 mm SYNTHETIC	29 → 6	spread only
cover windows latch depth	2.30 ±0.06 mm SYNTHETIC	29 → 6	spread only
S4 gripper pick loss	1 in 1,000 units HYPOTHESIS	29 → 23	smaller is safer

29 units show this failure on the first pass, some alongside another; 20 are lost to it after retries, each lost unit counted once under its main failure.

Fix. Best of 3 what-if reruns: tighten the tolerance of base hooks catch depth, 2.20 ±0.06 mm to 2.20 ±0.03 mm SYNTHETIC +21 good units.

Rerunthis failure 20 → 0all lost at S5 20 → 0first pass 96.30% → 96.72%after retries 96.56% → 96.98%

First build. Measure the gap at each hook on samples, compare with the camera, and gauge it. Owner: quality and design ASSUMPTION

R5 S3 cell not seated, 1 unit lost (0 to 1 over seeds): fewer than 5 units lost, below the search threshold, so no fix was searched. The retainer fingers do not snap over the cell's top edge. First build: check the retainer fingers snap over inserted cells; measure contact force.

4 All fixes in one rerun SIMULATEDUNCALIBRATEDSYNTHETIC PART

Risk	Change	To approve it before tooling ASSUMPTION
R1	base diameter, 29.80 ±0.08 mm to 29.88 ±0.08 mm	A tool-change quote and plan for the new nominal, then a dimensional check
R2	S4 gripper part offset, ±0.10 mm to ±0.05 mm	A capability run that measures the station's real spread
R3	cover cell ribs reach, 1.60 ±0.04 mm to 1.56 ±0.04 mm	A tool-change quote and plan for the new nominal, then a dimensional check
R4	base hooks catch depth, 2.20 ±0.06 mm to 2.20 ±0.03 mm	Supplier capability at the tighter tolerance, and a gauge study

First-pass yield

96.30% → 99.46%

Yield after retries

96.56% → 99.72%

Good units (+158; one at a time they sum to +168)

4,828 → 4,986

Units still lost, by failure

S4 jam / misalignment 13, S3 cell not seated 1

5 What the model does not cover, and where it is uncertain

Uncalibrated. No value here is measured; 5 of the 12 drivers in the tables are ASSUMPTION or HYPOTHESIS values. Counts are not a forecast, and the ranking and direction of each change hold only under these assumptions until a real build calibrates the stations.
Sampling. Over seeds 42 to 46, first-pass yield runs 96.14% to 96.74% and after retries 96.50% to 96.96%; S4 loses the most at 5 of 5 seeds, and R1 is the largest risk at 3 of 5. Ranges are spreads over these baseline seeds, not confidence intervals.
Inspection error. On the first pass the S5 inspect gap camera rejected 5 units whose true gap met both limits and passed 4 whose true gap did not. Its repeatability is an ASSUMPTION until a gauge study.
One change at a time. Each fix is a rerun with one change; fixes interact, so plan with the combined rerun. Candidates follow fixed rules (a nominal moved half or one tolerance toward margin, a halved tolerance or station error), so a better fix may lie outside them.
Physics and timing. Analytic models only; the MuJoCo contact cross-check does not cover these operations yet. The engine is a CPU Monte Carlo (NumPy) with SimPy line flow. Cycle times, transport and retries are ASSUMPTIONS, so the bottleneck and 462 good units per hour are not a forecast.
Not modeled. For this product, re-driving into formed threads, antenna and RF performance, cosmetic defects, drop, water ingress, cell aging. For any product, flex cables, adhesives, foams, cosmetics, reliability, firmware, supply and operator error.

Generated by uv run python -m virtual_build.risk_report --spec specs/tracker_puck.yaml --out examples/build-risk-report from engine commit a65ef230a3b1, spec specs/tracker_puck.yaml digest 800e67cef127. Every number is read from the engine runs in the companion JSON, which also holds each change for vb compare; tests/test_risk_report.py regenerates this page and fails if it drifts. Simulated, uncalibrated, synthetic part. In a blind backtest a page like this is locked before the real failures are opened. Not a physical calibration, a certification or a reason to skip a real build.