DOOM · The Avocado Post Doom Index

The probability that the present arrangement does not continue

Composite weekly index, eight components, rebased to 1 January 2026 = 100. Published every Friday at 07:00 and revised without notice.

53.0%

▲ +10.6 on the week

Previous	42.4
4-week mean	44.5
12-week mean	44.7
52-week range	38.9 – 53.0
Std. error, weekly Δ	±2.2
Observations	36

This week's movement of +10.6 points exceeds the standard error of the weekly change and is reported as significant at the 95 per cent level.

The series

Weekly reading Four-week mean

This week. A province changed its name without incident, which the panel accepted as evidence that consultation is optional at any scale.

Components and contributions

The index is the weighted mean of eight sub-indices, each on 0 to 100 where higher is worse. Weights are fixed between rebasings and sum to 100. Contribution is the share of this week's reading attributable to each component. Impulse is the number of index points that component added or removed since last week.

Component	Weight	Level	Δ week	Contribution	Impulse
CAP Capability Overhang	22%	54.5	+18.2	12.00	+4.0
ALN Alignment Deficit	19%	60.6	+16.8	11.50	+3.2
GOV Institutional Response Lag	16%	41.8	+1.3	6.70	+0.2
CON Compute Concentration	12%	57.8	+13.8	6.90	+1.7
DIS Discourse Degradation	9%	50.2	−5.6	4.50	−0.5
EQU Public Equanimity inv.	8%	49.1	−13.8	4.10	+1.1
EXE Executive Calm	7%	46.3	−5.3	3.20	−0.4
SUB Substitution Rate	7%	57.8	+19.2	4.00	+1.3
Composite	100%		+10.6	53.0

What each component measures

CAP Capability Overhang 22%

The gap between what the systems can already do and what anyone has got round to noticing.

Construction. Measured as the lag, in months, between a capability first appearing in a paper and it appearing in a procurement document. Scaled so that a lag of zero scores 0 and a lag of eighteen months scores 100.

Note. Our largest single weight, and the one we would most like to reduce.

ALN Alignment Deficit 19%

The distance between published safety commitments and the shipping schedule.

Construction. Word count of stated principles divided by the number of those principles for which a named individual is accountable. Log-transformed, because the raw series is not usable.

Note. The denominator has been zero on four occasions. The series is undefined for those weeks and has been interpolated.

GOV Institutional Response Lag 16%

How long it takes a body created to respond to something to respond to something.

Construction. Median elapsed time between a committee being convened and the same committee being reconvened. Winsorised at the 95th percentile to keep one particular committee from dominating the index.

Note. Seasonally adjusted for the summer recess, which has been extended twice.

CON Compute Concentration 12%

Who owns the machines, and how few of them there are.

Construction. A Herfindahl-Hirschman index across owners of frontier-scale capacity, normalised to 0 to 100.

Note. Rises when capacity concentrates into fewer hands. Also rose in the one quarter when it did not, for reasons the model has not been able to explain and we have not removed.

DIS Discourse Degradation 9%

The share of the argument now being conducted by people who have not read the thing.

Construction. Estimated from citation depth in public commentary. A citation to a summary of a summary is counted at one quarter weight.

Note. This component is the only one that has never fallen.

EQU Public Equanimity 8%

How calm everybody is about all of this.

Construction. A survey series, entered into the model inverted. High equanimity lowers the reading; very high equanimity raises it again above the 80th percentile.

Note. The reversal above the 80th percentile is not a modelling error. Calm past a point is the finding, not the comfort.

EXE Executive Calm 7%

The composure of the people who would actually know.

Construction. Coded from public appearances on a five-point scale by two independent raters. Inter-rater reliability is 0.91, which is higher than we expected and than we would like.

Note. Historically our single best predictor at a four-week horizon. We do not enjoy that and we have twice attempted to remove it. It is still here.

SUB Substitution Rate 7%

Work performed this week by something that does not eat.

Construction. Share of measured task-hours in the reference basket completed without a person present. Reported one week in arrears.

Note. The reference basket was last revised in March and no longer contains three of its original occupations.

Method

The index is a fixed-weight arithmetic composite. For components i with weights w_i summing to 100 and levels x_i on 0 to 100:

DOOM_t = ( Σ w_i · f_i(x_i,t) ) / 100

where f_i is the identity for all components except Public Equanimity, which is inverted, and which reverses again above its 80th percentile. That reversal is intentional and is discussed under Known limitations.

The result is rounded to one decimal place. It is not rounded to two, because two would imply we could tell the difference.

Revision policy

Readings are provisional for four weeks and final thereafter, apart from the ones that are revised later, which are also final. Revisions are not announced. A revised observation is marked in the underlying data and is not marked on the chart, in line with practice elsewhere.

Backtesting

The model has been backtested against every previous end of the world for which a dated record exists. It correctly identified all of them as not happening, with no false positives and a mean absolute error of 0.0. Performance on the remaining case is untested and, by construction, will be tested once.

Known limitations

The index is unbounded above 100 in principle. It has not gone above 100 and we have not modelled what we would do.
Six of the eight components are correlated with one another at above 0.7, which means the index has fewer degrees of freedom than it appears to. We have retained the eight-component structure because reducing it made the number go up.
The Executive Calm component leads the composite by roughly four weeks. We have twice attempted to remove it on the grounds that it is unfalsifiable and twice restored it on the grounds that it works.
No component measures anything that has actually happened. All eight measure how the people who might prevent it are behaving, which is a decision about the model and not an oversight.
The index has never been validated against an outcome. Neither has any other index of this kind, which we mention not as a defence but as context.

Recent observations

Week ending	Reading	Δ	CAP	ALN	GOV	CON
8 September 2026	53.0	+10.6	54.5	60.6	41.8	57.8
1 September 2026	42.4	+0.3	36.3	43.8	40.5	44.0
25 August 2026	42.1	+1.7	47.4	35.9	52.7	46.7
18 August 2026	40.4	−0.9	31.0	45.3	45.4	45.2
11 August 2026	41.3	−1.1	45.5	39.9	32.0	50.8
4 August 2026	42.4	−1.0	39.0	41.0	36.6	51.5
28 July 2026	43.4	−1.1	48.0	31.7	50.0	41.8
21 July 2026	44.5	−1.1	39.6	52.5	44.2	37.7
14 July 2026	45.6	−2.0	49.0	54.2	45.7	31.7

Model card

Model	Avocado Composite Doom Index
Version	2.4
Base period	1 January 2026 = 100
Frequency	Weekly, Fridays, 07:00
Components	8, fixed weights
Inputs	Public sources, one survey series, two coded raters
Owner	Statistics Desk, The Avocado Post
Peer review	None sought. None offered.
Reproducibility	The full series is published as JSON.

The Doom Index is a work of fiction and is not a forecast. It is constructed rigorously from figures that are invented, which distinguishes it from a number of indices constructed loosely from figures that are not. The idea of pricing this at all is taken, with thanks, from Skynet Countdown, which does it sincerely and in real time. Theirs is earnest. Ours is not.