Interpreted groups accumulate here — from obvious/outlier groups (step 1), homogeneous clusters, and manual selections (step 5). Each entry's traces are excluded from "remaining" everywhere in the app once added. Each card now also carries a small fingerprint panel (recent addition) — every case attribute the group has enough data to check, laid out together, with the ones that actually diverge from the whole dataset outlined — so the record captures not just what the group is, but a first-pass account of why. This tab is also where the next subset to investigate gets picked: either the whole "Remaining" pool, or an existing group reopened because it turned out not homogeneous enough — either way, the choice loops back to 2. Encoding.
Groups — 2D arrangement
hover a group for details · click to jump to its card on the right
How this is arranged: each interpreted group is reduced to a simple feature vector (the proportion of its traces' events belonging to each activity type), then projected to 2D via PCA (principal component analysis) — groups with similar behavioral profiles land near each other, dissimilar ones land far apart. Pie size ∝ √(trace count); slices show the outcome mix. This is a rough visual arrangement for orientation, not a precise similarity measure.
All interpreted groups — click a column header to sort; hover a row or a pie above to cross-highlight
| # ▲ | size | outcomes | dominant % | S | O |
|---|---|---|---|---|---|
| 0 | 46,371 | 100% | – | 0.99 | |
| 1 | 20,385 | 100% | – | 1.00 | |
| 2 | 362 | 95% | – | 0.81 | |
| 3 | 635 | 97% | 2.9× | 0.54 | |
| 4 | 589 | 95% | 2.9× | 0.23 | |
| 5 | 594 | 94% | 2.8× | 0.16 | |
| 6 | 56,482 | 100% | 1.9× | 1.00 | |
| 7 | 3,848 | 93% | 1.9× | 0.70 | |
| 8 | 1,523 | 100% | 1.9× | 1.00 | |
| 9 | 257 | 100% | 1.6× | 1.00 | |
| 10 | 4,779 | 98% | 3.2× | 0.92 | |
| 11 | 3,318 | 98% | 3.3× | 0.91 | |
| 12 | 1,471 | 86% | 3.6× | 0.62 | |
| 13 | 993 | 93% | 3.4× | 0.74 | |
| 15 | 1,547 | 92% | – | 0.78 | |
| 16 | 1,018 | 95% | – | 0.84 | |
| 17 | 549 | 96% | – | 0.87 | |
| 18 | 399 | 90% | – | 0.69 | |
| 19 | 229 | 90% | – | 0.69 | |
| 20 | 328 | 72% | 3.9× | 0.45 | |
| 21 | 292 | 84% | 4.2× | 0.61 | |
| 22 | 295 | 75% | 3.9× | 0.38 |
Interpreted groups
22 groups interpreted (+1 reopened for further analysis, shown below with their children) — 146,264 of 150,370 traces (97.3% of the full dataset) explained so far; 4,106 traces (2.7%) remain, still unexplained.
Obvious group: Create Fine → Payment (S – · O 0.99)
STEP 1 · OBVIOUS GROUPACTIVE
paid_full: 46,315 (100%)unresolved: 56 (0%)
notification type
absent 100% (+53pp)
expense
0.00 (0.0×)
trace length (N actions)
2 (0.5×)
duration
10d (0.0×)
max gap
10d (0.0×)
creation year
peak 2000
End-date dynamics — year × month (n=46,371)max 850/mo
Temporal interpretationOnly 13% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Jul 2009 (850 trace ends), falls 4 months from the nearest referral.
Obvious group: Create Fine → Send Fine (S – · O 1.00)
STEP 1 · OBVIOUS GROUPACTIVE
unresolved: 20,385 (100%)
notification type
absent 100% (+53pp)
trace length (N actions)
2 (0.5×)
creation year
peak 2000
duration
97d (0.3×)
expense
12.53 (1.5×)
max gap
97d (0.4×)
End-date dynamics — year × month (n=20,385)max 873/mo
Temporal interpretation16% of its traces end within ±1 month of a referral, close to the 19% baseline — no clear relationship. Its busiest month, Dec 2005 (873 trace ends), falls 2 months from the nearest referral.
Obvious group: Create Fine → Payment → Send Fine (S – · O 0.81)
STEP 1 · OBVIOUS GROUPACTIVE
unresolved: 343 (95%)paid_full: 19 (5%)
notification type
absent 100% (+53pp)
creation year
peak 2002 (+39pp)
duration
64d (0.2×)
max gap
54d (0.2×)
trace length (N actions)
3 (0.8×)
amount
35.82 (0.5×)
End-date dynamics — year × month (n=362)max 55/mo
Temporal interpretationOnly 4% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Jul 2002 (55 trace ends), falls 3 months from the nearest referral.
Topic T1-dominant (K=6, step 4) (S 2.9× · O 0.54)
STEP 4 · TOPICACTIVE
paid_full: 618 (97%)unresolved: 17 (3%)
notification type
absent 100% (+53pp)
creation year
peak 2001 (+19pp)
duration
140d (0.4×)
max gap
110d (0.4×)
trace length (N actions)
3 (0.8×)
expense
10.34 (1.3×)
End-date dynamics — year × month (n=635)max 104/mo
Temporal interpretationOnly 8% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Dec 2001 (104 trace ends), falls 4 months from the nearest referral.
Topic T3-dominant (K=6, step 4) (S 2.9× · O 0.23)
STEP 4 · TOPICACTIVE
paid_full: 558 (95%)unresolved: 31 (5%)
notification type
absent 100% (+53pp)
creation year
peak 2002 (+49pp)
duration
92d (0.3×)
max gap
67d (0.3×)
trace length (N actions)
3 (0.8×)
amount
44.74 (0.6×)
End-date dynamics — year × month (n=589)max 119/mo
Temporal interpretationOnly 7% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Sep 2002 (119 trace ends), falls 4 months from the nearest referral.
Topic T4-dominant (K=6, step 4) (S 2.8× · O 0.16)
STEP 4 · TOPICACTIVE
paid_full: 559 (94%)unresolved: 35 (6%)
notification type
absent 100% (+53pp)
creation year
peak 2002 (+29pp)
trace length (N actions)
3 (0.8×)
max gap
148d (0.6×)
duration
217d (0.6×)
expense
10.40 (1.3×)
End-date dynamics — year × month (n=594)max 78/mo
Temporal interpretationOnly 10% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Oct 2002 (78 trace ends), falls 3 months from the nearest referral.
Topic T0-dominant (K=15, step 4) (S 1.9× · O 1.00)
STEP 4 · TOPICACTIVE
credit_collection: 56,482 (100%)
notification type
P 100% (+47pp)
duration
690d (2.0×)
max gap
527d (2.1×)
trace length (N actions)
5 (1.3×)
creation year
peak 2000 (+6pp)
expense
11.70 (1.4×)
End-date dynamics — year × month (n=56,482)max 8799/mo
Temporal noteTrivial by construction — this entry's outcome is credit_collection, so its 100% concentration within ±1 month of a referral batch (vs. 19% baseline) just restates that its cases end when they're referred.
Topic T4-dominant (K=15, step 4) (S 1.9× · O 0.70)
STEP 4 · TOPICACTIVE
paid_full: 3,587 (93%)unresolved: 257 (7%)credit_collection: 4 (0%)
notification type
absent 0%
trace length (N actions)
6 (1.6×)
expense
11.81 (1.4×)
max gap
156d (0.6×)
amount
95.60 (1.3×)
creation year
peak 2012 (+2pp)
End-date dynamics — year × month (n=3,848)max 133/mo
Temporal interpretation20% of its traces end within ±1 month of a referral, close to the 19% baseline — no clear relationship. Its busiest month, Apr 2002 (133 trace ends), is itself a referral month.
Topic T6-dominant (K=15, step 4) (S 1.9× · O 1.00)
STEP 4 · TOPICACTIVE
credit_collection: 1,523 (100%)
notification type
P 100% (+47pp)
trace length (N actions)
6 (1.6×)
duration
675d (2.0×)
max gap
456d (1.8×)
creation year
peak 2000 (+7pp)
expense
11.30 (1.4×)
End-date dynamics — year × month (n=1,523)max 253/mo
Temporal noteTrivial by construction — this entry's outcome is credit_collection, so its 99% concentration within ±1 month of a referral batch (vs. 19% baseline) just restates that its cases end when they're referred.
Topic T14-dominant (K=15, step 4) (S 1.6× · O 1.00)
STEP 4 · TOPICACTIVE
credit_collection: 257 (100%)
notification type
absent 0%
trace length (N actions)
9 (2.4×)
expense
19.93 (2.4×)
duration
911d (2.7×)
max gap
610d (2.4×)
amount
142.33 (2.0×)
End-date dynamics — year × month (n=257)max 38/mo
Temporal noteTrivial by construction — this entry's outcome is credit_collection, so its 94% concentration within ±1 month of a referral batch (vs. 19% baseline) just restates that its cases end when they're referred.
Topic T0-dominant (K=9, step 4) (S 3.2× · O 0.92)
STEP 4 · TOPICACTIVE
paid_full: 4,700 (98%)unresolved: 78 (2%)dismissed: 1 (0%)
notification type
P 100% (+47pp)
creation year
peak 2004 (+12pp)
trace length (N actions)
5 (1.3×)
expense
12.50 (1.5×)
duration
559d (1.6×)
max gap
378d (1.5×)
End-date dynamics — year × month (n=4,779)max 634/mo
Temporal interpretationOnly 7% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Dec 2005 (634 trace ends), falls 2 months from the nearest referral.
Topic T1-dominant (K=9, step 4) (S 3.3× · O 0.91)
STEP 4 · TOPICACTIVE
unresolved: 3,253 (98%)paid_full: 65 (2%)
notification type
P 100% (+47pp)
creation year
peak 2000 (+15pp)
trace length (N actions)
6 (1.6×)
expense
13.01 (1.6×)
max gap
134d (0.5×)
duration
259d (0.8×)
End-date dynamics — year × month (n=3,318)max 172/mo
Temporal interpretation21% of its traces end within ±1 month of a referral, close to the 19% baseline — no clear relationship. Its busiest month, Nov 2000 (172 trace ends), falls 17 months from the nearest referral.
Topic T3-dominant (K=9, step 4) (S 3.6× · O 0.62)
STEP 4 · TOPICACTIVE
paid_full: 1,260 (86%)unresolved: 211 (14%)
notification type
absent 96% (+49pp)
creation year
peak 2002 (+31pp)
duration
83d (0.2×)
max gap
52d (0.2×)
trace length (N actions)
3 (0.9×)
amount
44.93 (0.6×)
End-date dynamics — year × month (n=1,471)max 209/mo
Temporal interpretationOnly 6% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Sep 2002 (209 trace ends), falls 4 months from the nearest referral.
Topic T6-dominant (K=9, step 4) (S 3.4× · O 0.74)
STEP 4 · TOPICACTIVE
paid_full: 923 (93%)unresolved: 58 (6%)dismissed: 12 (1%)
notification type
P 99% (+46pp)
creation year
peak 2007 (+20pp)
trace length (N actions)
5 (1.3×)
expense
12.50 (1.5×)
max gap
198d (0.8×)
duration
368d (1.1×)
End-date dynamics — year × month (n=993)max 120/mo
Temporal interpretationOnly 9% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Oct 2008 (120 trace ends), falls 5 months from the nearest referral.
Topic T8-dominant (K=9, step 4) (S 3.5× · O 0.48) ↳ superseded — reopened for further analysis
STEP 4 · TOPICSUPERSEDED · reopened
credit_collection: 718 (83%)paid_full: 71 (8%)unresolved: 71 (8%)
notification type
absent 0%
trace length (N actions)
6 (1.6×)
creation year
peak 2003 (+12pp)
duration
710d (2.1×)
max gap
539d (2.1×)
expense
11.97 (1.4×)
End-date dynamics — year × month (n=860)max 128/mo
Temporal noteTrivial by construction — this entry's outcome is credit_collection, so its 86% concentration within ±1 month of a referral batch (vs. 19% baseline) just restates that its cases end when they're referred.
Topic T0-dominant (K=20, step 4) (S – · O 0.78)
STEP 4 · TOPICACTIVE
unresolved: 1,428 (92%)paid_full: 119 (8%)
notification type
P 100% (+47pp)
trace length (N actions)
5 (1.3×)
expense
12.03 (1.5×)
max gap
99d (0.4×)
duration
168d (0.5×)
creation year
peak 2000
End-date dynamics — year × month (n=1,547)max 68/mo
Temporal interpretation23% of its traces end within ±1 month of a referral, close to the 19% baseline — no clear relationship. Its busiest month, Jan 2004 (68 trace ends), is itself a referral month.
Topic T2-dominant (K=20, step 4) (S – · O 0.84)
STEP 4 · TOPICACTIVE
paid_full: 969 (95%)unresolved: 49 (5%)
notification type
P 100% (+47pp)
creation year
peak 2011 (+42pp)
expense
13.93 (1.7×)
trace length (N actions)
5 (1.3×)
amount
111.46 (1.6×)
duration
419d (1.2×)
End-date dynamics — year × month (n=1,018)max 313/mo
Temporal interpretationOnly 6% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Sep 2012 (313 trace ends), falls 6 months from the nearest referral.
Topic T3-dominant (K=20, step 4) (S – · O 0.87)
STEP 4 · TOPICACTIVE
credit_collection: 528 (96%)unresolved: 21 (4%)
notification type
P 100% (+47pp)
creation year
peak 2003 (+15pp)
trace length (N actions)
6 (1.6×)
duration
613d (1.8×)
expense
12.81 (1.6×)
max gap
442d (1.7×)
End-date dynamics — year × month (n=549)max 111/mo
Temporal noteTrivial by construction — this entry's outcome is credit_collection, so its 97% concentration within ±1 month of a referral batch (vs. 19% baseline) just restates that its cases end when they're referred.
Topic T5-dominant (K=20, step 4) (S – · O 0.69)
STEP 4 · TOPICACTIVE
unresolved: 361 (90%)paid_full: 21 (5%)dismissed: 17 (4%)
notification type
P 100% (+47pp)
creation year
peak 2000 (+41pp)
trace length (N actions)
5 (1.4×)
max gap
59d (0.2×)
duration
130d (0.4×)
amount
128.28 (1.8×)
End-date dynamics — year × month (n=399)max 60/mo
Temporal interpretationOnly 6% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Nov 2000 (60 trace ends), falls 17 months from the nearest referral.
Topic T18-dominant (K=20, step 4) (S – · O 0.69)
STEP 4 · TOPICACTIVE
unresolved: 206 (90%)paid_full: 19 (8%)dismissed: 4 (2%)
notification type
P 99% (+46pp)
creation year
peak 2001 (+18pp)
trace length (N actions)
5 (1.3×)
max gap
113d (0.4×)
expense
11.26 (1.4×)
duration
197d (0.6×)
End-date dynamics — year × month (n=229)max 41/mo
Temporal interpretation29% of its traces end within ±1 month of a mass credit-collection referral (vs. 19% baseline) — roughly balanced before/after (7% vs. 6%). Its busiest month, Feb 2002 (41 trace ends), falls 2 months from the nearest referral.
Homogeneous cluster: T15-dominant, 72% dismissed (S 3.9× · O 0.45)
STEP 5 · CLUSTERACTIVE
dismissed: 236 (72%)unresolved: 91 (28%)paid_full: 1 (0%)
notification type
P 100% (+47pp)
trace length (N actions)
6 (1.6×)
creation year
peak 2011 (+8pp)
expense
13.18 (1.6×)
max gap
96d (0.4×)
amount
126.42 (1.8×)
End-date dynamics — year × month (n=328)max 30/mo
Temporal interpretation36% of its traces end within ±1 month of a mass credit-collection referral (vs. 19% baseline) — skewed toward the month before a referral (18% vs. 8%), as if these cases are closed out just ahead of a sweep. Its busiest month, Mar 2007 (30 trace ends), is the month before a referral.
Homogeneous cluster: T17-dominant, 84% unresolved (S 4.2× · O 0.61)
STEP 5 · CLUSTERACTIVE
unresolved: 246 (84%)dismissed: 46 (16%)
notification type
absent 0%
max gap
1,250d (4.9×)
duration
1,417d (4.1×)
trace length (N actions)
6 (1.6×)
creation year
peak 2008 (+9pp)
expense
12.39 (1.5×)
End-date dynamics — year × month (n=292)max 237/mo
Temporal interpretationOnly 7% of its traces end within ±1 month of a referral (vs. 19% baseline) — its timing runs independent of the collection schedule. Its busiest month, Dec 2011 (237 trace ends), falls 3 months from the nearest referral.
Homogeneous cluster: T4-dominant, 75% unresolved (S 3.9× · O 0.38)
STEP 5 · CLUSTERACTIVE
unresolved: 221 (75%)paid_full: 64 (22%)dismissed: 7 (2%)credit_collection: 3 (1%)
notification type
absent 0%
trace length (N actions)
9 (2.4×)
expense
21.88 (2.6×)
creation year
peak 2000
max gap
163d (0.6×)
amount
98.26 (1.4×)
End-date dynamics — year × month (n=295)max 10/mo
Temporal interpretation17% of its traces end within ±1 month of a referral, close to the 19% baseline — no clear relationship. Its busiest month, May 2012 (10 trace ends), falls 2 months from the nearest referral.
Remaining / unexplained
4,106 of 150,370 traces (2.7%) not yet interpreted.
dismissed: 1,737 (42%)unresolved: 1,663 (41%)paid_full: 493 (12%)credit_collection: 213 (5%)
End-date dynamics — year × month (n=4,106)max 273/mo
Temporal interpretation25% of its traces end within ±1 month of a mass credit-collection referral (vs. 19% baseline) — roughly balanced before/after (7% vs. 6%). Its busiest month, Dec 2011 (273 trace ends), falls 3 months from the nearest referral.
Temporal patterns across groups
Beyond the month-of-year rhythm relative to mass credit-collection referrals (below), the committed groups also differ sharply in when across the log's 13.5-year span their traces end — a dimension none of this session's encodings addressed directly, since none of them use absolute calendar time. Three loose patterns show up when the year of each trace's end date is compared against the log's 2000–2013 range.
Concentrated early (2000–2004). G4 (Topic T3-dominant, K=6, 95% paid_full) and G12 (Topic T3-dominant, K=9, 86% paid_full) put 94% and 91% of their traces in the log's first five years; G2 (obvious group, 95% unresolved) does much the same at 88% — cases created early that simply never closed. G11 (Topic T1-dominant, K=9, 98% unresolved) is front-loaded differently: 57% falls in 2000–2004 and none at all in the final five years, so this group's cases stop appearing well before the log ends rather than trailing off gradually.
Concentrated late (2009–2013). G21 (Homogeneous cluster, 84% unresolved) is close to the mirror image of G11: 97% of its traces end in the log's final five years, essentially none in the first five — the natural reading is right-censoring, cases still open when the log was extracted. G16 (Topic T2-dominant, K=20, 95% paid_full) shows a similar late skew (86%) without an obvious censoring explanation, which is a smaller open question rather than a settled one.
Spread across the whole span. G0 and G1, the two largest and earliest-committed obvious groups, split close to evenly between early and late years (30%/29% and 29%/31%) — consistent with the continuous case-creation pattern already visible in the log's overall start-date distribution. Neither has a temporal signature beyond the log's own steady intake.
Outcome alone does not predict which pattern a group falls into: the four unresolved-dominant groups above (G1, G2, G11, G21) show four different signatures — flat, early, front-loaded-but-not-early, and late. That is the same lesson the workflow's structure-and-outcome criterion applies elsewhere in this session: one outcome label can still cover several distinct behavioral populations. See "Outcome dynamics" below for how these same groups relate to the mass-referral schedule specifically.
Concentrated early (2000–2004). G4 (Topic T3-dominant, K=6, 95% paid_full) and G12 (Topic T3-dominant, K=9, 86% paid_full) put 94% and 91% of their traces in the log's first five years; G2 (obvious group, 95% unresolved) does much the same at 88% — cases created early that simply never closed. G11 (Topic T1-dominant, K=9, 98% unresolved) is front-loaded differently: 57% falls in 2000–2004 and none at all in the final five years, so this group's cases stop appearing well before the log ends rather than trailing off gradually.
Concentrated late (2009–2013). G21 (Homogeneous cluster, 84% unresolved) is close to the mirror image of G11: 97% of its traces end in the log's final five years, essentially none in the first five — the natural reading is right-censoring, cases still open when the log was extracted. G16 (Topic T2-dominant, K=20, 95% paid_full) shows a similar late skew (86%) without an obvious censoring explanation, which is a smaller open question rather than a settled one.
Spread across the whole span. G0 and G1, the two largest and earliest-committed obvious groups, split close to evenly between early and late years (30%/29% and 29%/31%) — consistent with the continuous case-creation pattern already visible in the log's overall start-date distribution. Neither has a temporal signature beyond the log's own steady intake.
Outcome alone does not predict which pattern a group falls into: the four unresolved-dominant groups above (G1, G2, G11, G21) show four different signatures — flat, early, front-loaded-but-not-early, and late. That is the same lesson the workflow's structure-and-outcome criterion applies elsewhere in this session: one outcome label can still cover several distinct behavioral populations. See "Outcome dynamics" below for how these same groups relate to the mass-referral schedule specifically.
Outcome dynamics — all traces, pooled
The same trace-end months as the cards above, pooled across every group and broken down by outcome instead of by knowledge-base entry. Left: raw monthly counts. Right: the same counts renormalized to a monthly share (proportions sum to 100% in every month) — separating a shift in outcome mix from a shift in overall case-closing volume. 150,370 traces, Jan 2000 – Jun 2013.
Mass credit-collection referrals. Ten months in this span each closed more than 1,000 traces as
Mass credit-collection referrals. Ten months in this span each closed more than 1,000 traces as
credit_collection in one go — Apr 2002, Jan 2003, Jan 2004, Dec 2004, Feb 2006, Feb 2007, Mar 2009, Oct 2010, Mar 2012, Apr 2013 — against a typical month of a few dozen; these are marked with a dashed gold line on both charts below. Every card above is annotated with how its own end dates relate to this schedule: whether they cluster within a month of a referral (and if so, mostly just before it, just after it, or evenly split), or run on their own rhythm regardless of it. The credit_collection-dominant entries (G6, G8, G9, G17, and the reopened G14) trivially coincide with these months by definition of their outcome; the more informative cases are the non-collection entries that still cluster nearby — G20 (T15-dominant, 72% dismissed) most clearly, skewed to the month before a referral, as if those cases are resolved or dismissed just ahead of the sweep rather than being swept up in it.Outcome counts per month
Credit collectionDismissedPaid in fullUnresolvedmass referral month
Outcome share per month
Credit collectionDismissedPaid in fullUnresolvedmass referral month