Ownership and stewardship

Why does ownership decay fastest, and what should it be derived from?

Ownership is asserted metadata about people, and people change roles faster than schemas change. With ordinary departures plus internal moves of perhaps 20–30% a year, an ownership field left alone for three years is more likely wrong than right.

Worked example

If 25% of owners change each year, the chance that an owner is still correct after three years is 0.75 × 0.75 × 0.75 = 0.42. So there is a 58% chance the field is wrong. At 20% a year it is 0.8³ = 0.51. At 30% it is 0.7³ = 0.34.

RelationshipAsserted metadata decays →

0.75 \times 0.75 \times 0.75 = 0.42

$c$: share of owners who change per year

20% churn a year25%30%

Data table

Chance an untouched ownership field is still right
Years since last update	series	Still correct
0	20% churn a year	100%
0.075	20% churn a year	98.3%
0.15	20% churn a year	96.7%
0.225	20% churn a year	95.1%
0.3	20% churn a year	93.5%
0.375	20% churn a year	92%
0.45	20% churn a year	90.4%
0.525	20% churn a year	88.9%
0.6	20% churn a year	87.5%
0.675	20% churn a year	86%
0.75	20% churn a year	84.6%
0.825	20% churn a year	83.2%
0.9	20% churn a year	81.8%
0.975	20% churn a year	80.4%
1.05	20% churn a year	79.1%
1.13	20% churn a year	77.8%
1.2	20% churn a year	76.5%
1.27	20% churn a year	75.2%
1.35	20% churn a year	74%
1.43	20% churn a year	72.8%
1.5	20% churn a year	71.6%
1.57	20% churn a year	70.4%
1.65	20% churn a year	69.2%
1.73	20% churn a year	68.1%
1.8	20% churn a year	66.9%
1.88	20% churn a year	65.8%
1.95	20% churn a year	64.7%
2.02	20% churn a year	63.6%
2.1	20% churn a year	62.6%
2.17	20% churn a year	61.5%
2.25	20% churn a year	60.5%
2.33	20% churn a year	59.5%
2.4	20% churn a year	58.5%
2.48	20% churn a year	57.6%
2.55	20% churn a year	56.6%
2.63	20% churn a year	55.7%
2.7	20% churn a year	54.7%
2.77	20% churn a year	53.8%
2.85	20% churn a year	52.9%
2.92	20% churn a year	52.1%
3	20% churn a year	51.2%
3.08	20% churn a year	50.4%
3.15	20% churn a year	49.5%
3.23	20% churn a year	48.7%
3.3	20% churn a year	47.9%
3.38	20% churn a year	47.1%
3.45	20% churn a year	46.3%
3.52	20% churn a year	45.5%
3.6	20% churn a year	44.8%
3.67	20% churn a year	44%
3.75	20% churn a year	43.3%
3.83	20% churn a year	42.6%
3.9	20% churn a year	41.9%
3.98	20% churn a year	41.2%
4.05	20% churn a year	40.5%
4.13	20% churn a year	39.8%
4.2	20% churn a year	39.2%
4.28	20% churn a year	38.5%
4.35	20% churn a year	37.9%
4.42	20% churn a year	37.3%
4.5	20% churn a year	36.6%
4.58	20% churn a year	36%
4.65	20% churn a year	35.4%
4.72	20% churn a year	34.8%
4.8	20% churn a year	34.3%
4.88	20% churn a year	33.7%
4.95	20% churn a year	33.1%
5.03	20% churn a year	32.6%
5.1	20% churn a year	32%
5.17	20% churn a year	31.5%
5.25	20% churn a year	31%
5.33	20% churn a year	30.5%
5.4	20% churn a year	30%
5.47	20% churn a year	29.5%
5.55	20% churn a year	29%
5.63	20% churn a year	28.5%
5.7	20% churn a year	28%
5.78	20% churn a year	27.6%
5.85	20% churn a year	27.1%
5.92	20% churn a year	26.7%
6	20% churn a year	26.2%
0	25%	100%
0.075	25%	97.9%
0.15	25%	95.8%
0.225	25%	93.7%
0.3	25%	91.7%
0.375	25%	89.8%
0.45	25%	87.9%
0.525	25%	86%
0.6	25%	84.1%
0.675	25%	82.4%
0.75	25%	80.6%
0.825	25%	78.9%
0.9	25%	77.2%
0.975	25%	75.5%
1.05	25%	73.9%
1.13	25%	72.4%
1.2	25%	70.8%
1.27	25%	69.3%
1.35	25%	67.8%
1.43	25%	66.4%
1.5	25%	65%
1.57	25%	63.6%
1.65	25%	62.2%
1.73	25%	60.9%
1.8	25%	59.6%
1.88	25%	58.3%
1.95	25%	57.1%
2.02	25%	55.8%
2.1	25%	54.7%
2.17	25%	53.5%
2.25	25%	52.3%
2.33	25%	51.2%
2.4	25%	50.1%
2.48	25%	49.1%
2.55	25%	48%
2.63	25%	47%
2.7	25%	46%
2.77	25%	45%
2.85	25%	44%
2.92	25%	43.1%
3	25%	42.2%
3.08	25%	41.3%
3.15	25%	40.4%
3.23	25%	39.5%
3.3	25%	38.7%
3.38	25%	37.9%
3.45	25%	37.1%
3.52	25%	36.3%
3.6	25%	35.5%
3.67	25%	34.7%
3.75	25%	34%
3.83	25%	33.3%
3.9	25%	32.6%
3.98	25%	31.9%
4.05	25%	31.2%
4.13	25%	30.5%
4.2	25%	29.9%
4.28	25%	29.2%
4.35	25%	28.6%
4.42	25%	28%
4.5	25%	27.4%
4.58	25%	26.8%
4.65	25%	26.2%
4.72	25%	25.7%
4.8	25%	25.1%
4.88	25%	24.6%
4.95	25%	24.1%
5.03	25%	23.6%
5.1	25%	23.1%
5.17	25%	22.6%
5.25	25%	22.1%
5.33	25%	21.6%
5.4	25%	21.2%
5.47	25%	20.7%
5.55	25%	20.3%
5.63	25%	19.8%
5.7	25%	19.4%
5.78	25%	19%
5.85	25%	18.6%
5.92	25%	18.2%
6	25%	17.8%
0	30%	100%
0.075	30%	97.4%
0.15	30%	94.8%
0.225	30%	92.3%
0.3	30%	89.9%
0.375	30%	87.5%
0.45	30%	85.2%
0.525	30%	82.9%
0.6	30%	80.7%
0.675	30%	78.6%
0.75	30%	76.5%
0.825	30%	74.5%
0.9	30%	72.5%
0.975	30%	70.6%
1.05	30%	68.8%
1.13	30%	66.9%
1.2	30%	65.2%
1.27	30%	63.5%
1.35	30%	61.8%
1.43	30%	60.2%
1.5	30%	58.6%
1.57	30%	57%
1.65	30%	55.5%
1.73	30%	54%
1.8	30%	52.6%
1.88	30%	51.2%
1.95	30%	49.9%
2.02	30%	48.6%
2.1	30%	47.3%
2.17	30%	46%
2.25	30%	44.8%
2.33	30%	43.6%
2.4	30%	42.5%
2.48	30%	41.4%
2.55	30%	40.3%
2.63	30%	39.2%
2.7	30%	38.2%
2.77	30%	37.2%
2.85	30%	36.2%
2.92	30%	35.2%
3	30%	34.3%
3.08	30%	33.4%
3.15	30%	32.5%
3.23	30%	31.7%
3.3	30%	30.8%
3.38	30%	30%
3.45	30%	29.2%
3.52	30%	28.4%
3.6	30%	27.7%
3.67	30%	27%
3.75	30%	26.2%
3.83	30%	25.6%
3.9	30%	24.9%
3.98	30%	24.2%
4.05	30%	23.6%
4.13	30%	23%
4.2	30%	22.4%
4.28	30%	21.8%
4.35	30%	21.2%
4.42	30%	20.6%
4.5	30%	20.1%
4.58	30%	19.6%
4.65	30%	19%
4.72	30%	18.5%
4.8	30%	18%
4.88	30%	17.6%
4.95	30%	17.1%
5.03	30%	16.7%
5.1	30%	16.2%
5.17	30%	15.8%
5.25	30%	15.4%
5.33	30%	15%
5.4	30%	14.6%
5.47	30%	14.2%
5.55	30%	13.8%
5.63	30%	13.4%
5.7	30%	13.1%
5.78	30%	12.7%
5.85	30%	12.4%
5.92	30%	12.1%
6	30%	11.8%

Principle: Absence is not evidence →

It has the highest decay rate of anything here and, critically, no signal when it becomes false. The field still renders.

The fix is the same as everywhere else: derive it from a fact that is already load-bearing. Examples: who owns the code repository, which deploy identity runs the job, which service principal (a non-human account) writes the table. These are maintained because something breaks if they are wrong, so they stay true as a by-product.

A second structural point: ownership must resolve to a group, never to a person, because a person is a single point of failure with a guaranteed eventual failure. Use an indirection: the asset points to a team, and the team resolves to its members at query time. This is late binding again (as with asset IDs): a stable handle plus a deferred lookup.

Ownership decays faster than anything else you store. Derive it from something that already has to be right.