NEXUS

Tuning Studio · v0 · read-only

PLANTED TRUTH — Rates are against planted truth in a synthetic population, not SAR outcomes.

high_value — threshold sweep

Fires when: a single transaction's amount is strictly greater than the threshold
Today: £75,000
Population: 5,000 customers, 200 planted · seed 20260924 · digest 7dc1ddbc19ba…
Grain: customer

Precision (yield)Recall (coverage)F1

Capacity

Capacity: 300 a day at bank scale, TM-SIM read as a 1 in 150 sample = 2 per working day here. Unit of work: one alert per customer-month (30-day dedupe). 261 working days; bank holidays not modelled.

	Today	Operating point
Threshold	£75,000	£177,500
Recall	0.550	0.325
Precision	0.162	0.301
Effective recall, worked in the year	0.135	0.325
Workload, customer-months / year	2,419	498
per working day	9.27	1.91
Backlog at year end	1,897	1
Peak backlog	1,901	17
Wait p50 / p95, working days	105 / 198	3 / 7

Segments

Fitted on the named population and scored on a hold-out (seed 20260925) within a workload budget of 2,419 customer-months a year. A line resting on fewer than 10 planted customers is flagged thin. Planted per segment: business 34 · corporate 9 · correspondent 1 · retail 156.

Option	Lines	Lines to govern	Caught, fitted	Caught, hold-out	Workload, hold-out	Thin
One line	all £75,000	1	110	122	2,601	none
Chosen lines	non-retail £152,500 · retail £30,000	2	151	149	2,595	none
One per segment	business £95,000 · corporate £155,000 · correspondent £200,000 · retail £30,000	4	155	154	2,681	corporate, correspondent

Champion & challengers

Each challenger fitted within the champion's workload of 2,419 customer-months a year, scored on a hold-out (seed 20260925).

relative_spike mirrors how the planted value_spike was generated (a spike relative to the customer's own scale), so planted truth favours it by construction; a win shows the rule can catch what was planted, not that it beats the champion on real behaviour.

Rule	Fires when	Workload, hold-out	Caught, fitted	Caught, hold-out	Per 100 reviews	Only this rule	Negative control lift	Cold start
Champion	payment above £75,000	2,601	110	122	4.69	—	—	—
Amount and velocity	payment above £75,000 with 2+ payments in 7 days	1,519	80	88	5.79	0	0.75	0
Relative spike	payment over 4x the customer's 3-month median	2,253	161	170	7.55	63	0.93	49

Stability & drift

Every rule held fixed over periods drifted by a stated scenario (inflation and a non retail step). Unstable past 25% workload change or a 10-point recall drop; a transient when the busiest month exceeds 2x the median month. Payment amounts against Y1: Y2 PSI 0.02 (stable) · Y3 PSI 0.05 (stable).

Rule	Fires when	Workload Y1	Workload Y2	Workload Y3	Busiest month	Verdict
champion	payment above £75,000	2,419	4,185 (+73%)	5,423 (+124%)	1.07x	UNSTABLE
champion indexed	payment above £75,000, indexed to the median payment	2,419	3,510 (+45%)	4,080 (+69%)	1.06x	UNSTABLE
d3 lines	non-retail £152,500 / retail £30,000	2,353	3,481 (+48%)	4,360 (+85%)	1.12x	UNSTABLE
relative spike	payment over 4x the customer's 3-month median	2,286	2,704 (+18%)	2,293 (+0%)	1.74x	stable

Risk score & calibration

One score from the two logged rules: 50 x (largest payment / £75,000) + 50 x (largest multiple of the customer's own baseline / 4x), per customer-month, worked in order to 2,419 reviews. Ties are broken alphabetically, blind to the labels.

Weights are stated assumptions, not fitted: the score shows what ranking the two logged rules together DOES within a budget, not the best weights for them.

Queue	Workload, fitted	Caught, fitted	Caught, hold-out
Ranked score, 50/50	2,419	166	171
Customer-baseline rule	2,286	161	170
Champion	2,419	110	122

Band	Months	Planted rate, fitted	Hold-out, same score
top 0.5%	285	46%	35%
top 1%	286	5.6%	5.2%
top 2%	571	1.9%	1.5%
top 4%	1,141	0.44%	0.66%
top 8%	2,284	0.13%	0.32%
top 16%	4,566	0.22%	0.18%
top 32%	9,134	0.077%	0.066%
top 100%	38,817	0.046%	0.034%

Cap	Months tied at it	Caught	Cap is
200	330	166	free
100	1,652	166	free
75	3,375	112	binding — the tie-break decides
50	8,480	49	binding — the tie-break decides

Threshold	Alerts	TP	FP	FN	Precision	Recall	F1
£10,000	3469	188	3281	12	0.054	0.940	0.102
£12,500	3144	186	2958	14	0.059	0.930	0.111
£15,000	2852	184	2668	16	0.065	0.920	0.121
£17,500	2624	181	2443	19	0.069	0.905	0.128
£20,000	2401	173	2228	27	0.072	0.865	0.133
£22,500	2191	167	2024	33	0.076	0.835	0.140
£25,000	2022	167	1855	33	0.083	0.835	0.150
£27,500	1873	164	1709	36	0.088	0.820	0.158
£30,000	1738	162	1576	38	0.093	0.810	0.167
£32,500	1630	159	1471	41	0.098	0.795	0.174
£35,000	1535	154	1381	46	0.100	0.770	0.178
£37,500	1460	152	1308	48	0.104	0.760	0.183
£40,000	1382	150	1232	50	0.109	0.750	0.190
£42,500	1297	147	1150	53	0.113	0.735	0.196
£45,000	1213	144	1069	56	0.119	0.720	0.204
£47,500	1139	140	999	60	0.123	0.700	0.209
£50,000	1092	138	954	62	0.126	0.690	0.214
£52,500	1047	136	911	64	0.130	0.680	0.218
£55,000	1001	133	868	67	0.133	0.665	0.221
£57,500	960	133	827	67	0.139	0.665	0.229
£60,000	908	129	779	71	0.142	0.645	0.233
£62,500	862	126	736	74	0.146	0.630	0.237
£65,000	817	124	693	76	0.152	0.620	0.244
£67,500	765	119	646	81	0.156	0.595	0.247
£70,000	730	117	613	83	0.160	0.585	0.252
£72,500	710	115	595	85	0.162	0.575	0.253
£75,000	679	110	569	90	0.162	0.550	0.250
£77,500	652	108	544	92	0.166	0.540	0.254
£80,000	628	107	521	93	0.170	0.535	0.258
£82,500	603	103	500	97	0.171	0.515	0.257
£85,000	583	102	481	98	0.175	0.510	0.261
£87,500	560	101	459	99	0.180	0.505	0.266
£90,000	539	100	439	100	0.186	0.500	0.271
£92,500	524	99	425	101	0.189	0.495	0.273
£95,000	505	98	407	102	0.194	0.490	0.278
£97,500	480	94	386	106	0.196	0.470	0.276
£100,000	464	93	371	107	0.200	0.465	0.280
£102,500	450	91	359	109	0.202	0.455	0.280
£105,000	432	89	343	111	0.206	0.445	0.282
£107,500	415	89	326	111	0.214	0.445	0.289
£110,000	405	88	317	112	0.217	0.440	0.291
£112,500	395	85	310	115	0.215	0.425	0.286
£115,000	386	85	301	115	0.220	0.425	0.290
£117,500	376	85	291	115	0.226	0.425	0.295
£120,000	360	82	278	118	0.228	0.410	0.293
£122,500	354	81	273	119	0.229	0.405	0.292
£125,000	340	79	261	121	0.232	0.395	0.293
£127,500	334	79	255	121	0.237	0.395	0.296
£130,000	329	79	250	121	0.240	0.395	0.299
£132,500	325	77	248	123	0.237	0.385	0.293
£135,000	318	76	242	124	0.239	0.380	0.293
£137,500	304	75	229	125	0.247	0.375	0.298
£140,000	296	73	223	127	0.247	0.365	0.294
£142,500	291	73	218	127	0.251	0.365	0.297
£145,000	284	72	212	128	0.254	0.360	0.298
£147,500	275	72	203	128	0.262	0.360	0.303
£150,000	271	72	199	128	0.266	0.360	0.306
£152,500	264	72	192	128	0.273	0.360	0.310
£155,000	254	69	185	131	0.272	0.345	0.304
£157,500	250	68	182	132	0.272	0.340	0.302
£160,000	247	68	179	132	0.275	0.340	0.304
£162,500	238	67	171	133	0.282	0.335	0.306
£165,000	235	67	168	133	0.285	0.335	0.308
£167,500	229	66	163	134	0.288	0.330	0.308
£170,000	224	66	158	134	0.295	0.330	0.311
£172,500	223	66	157	134	0.296	0.330	0.312
£175,000	221	65	156	135	0.294	0.325	0.309
£177,500	216	65	151	135	0.301	0.325	0.312
£180,000	210	64	146	136	0.305	0.320	0.312
£182,500	209	64	145	136	0.306	0.320	0.313
£185,000	207	64	143	136	0.309	0.320	0.314
£187,500	205	64	141	136	0.312	0.320	0.316
£190,000	199	62	137	138	0.312	0.310	0.311
£192,500	191	60	131	140	0.314	0.300	0.307
£195,000	187	59	128	141	0.316	0.295	0.305
£197,500	184	59	125	141	0.321	0.295	0.307
£200,000	180	59	121	141	0.328	0.295	0.311