tegwick
f67787017c
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Assistant: claude-code Assistant-Model: opus Assistant-Process: 14913@bnt-lap001 Assistant-Session: add7ef69-b901-4756-9a67-34cf1125d96f
193 lines
8.4 KiB
Markdown
193 lines
8.4 KiB
Markdown
<!--
|
||
Created 2026-10-12 under ASSEMBLY-WP-0004-T07 (INTENT.md next step 7).
|
||
Follows INTENT.md "Metrics Philosophy" and GADM-0.1 § 22: every construct is
|
||
measured by a sensor bundle, and metric disagreement is information.
|
||
-->
|
||
|
||
# Metrics Catalog v0.1
|
||
|
||
**Status:** exploratory. Indicators are *candidates*: their availability,
|
||
validity and data sources still have to be checked before use.
|
||
|
||
Each GADM-0.2 stock gets a **sensor bundle** of four kinds of measure:
|
||
|
||
- **O** objective outcome
|
||
- **St** structural
|
||
- **B** behavioral
|
||
- **Su** subjective
|
||
|
||
Each stock also gets its main **Goodhart risk**: how optimizing the measure
|
||
could hollow out the construct.
|
||
|
||
Rules for using the catalog:
|
||
|
||
1. No stock is reported through a single indicator.
|
||
2. When indicators in a bundle disagree, record the disagreement as a
|
||
finding. Do not average it away.
|
||
3. A stock's normalization to [0, 1] in GADM is a modeling convenience. The
|
||
catalog keeps the raw indicators.
|
||
4. Changes to indicators are governed like model changes: versioned, with a
|
||
rationale (WS-11, governance of metrics).
|
||
|
||
Concrete sources per jurisdiction for each bundle are listed in
|
||
`checkpoints/IndicatorCatalog.md` (v0.2).
|
||
|
||
---
|
||
|
||
## H — Human capability
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | healthy life expectancy; adult skills proficiency (literacy, numeracy, problem solving) |
|
||
| St | coverage of education, care and lifelong-learning provision; mentoring and apprenticeship capacity |
|
||
| B | participation in learning after formal education; skill use at work and outside it |
|
||
| Su | self-assessed capability to learn and to act on one's goals |
|
||
|
||
**Goodhart risk:** credential inflation. Counting certificates instead of
|
||
usable capability. Organism-assisted performance can mask deskilling (link
|
||
#40), so measure unassisted capability as well.
|
||
|
||
## X — Assembly commons
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | reuse rate of open components (dependents per open package, citations of open datasets) |
|
||
| St | share of critical infrastructure under open standards; share of openly licensed publicly funded research and software |
|
||
| B | contribution and maintenance activity in commons projects; maintainer continuity |
|
||
| Su | practitioners' perceived ease of building on existing components |
|
||
|
||
**Goodhart risk:** publishing without maintenance. Counting released artifacts
|
||
rather than reusable, maintained ones. Enclosure through dependence on a
|
||
single steward.
|
||
|
||
## K_phys / K_Ω — Productive capacity and organism capability
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | sustainable output per unit of physical throughput; task coverage of information organisms |
|
||
| St | installed automation and compute capacity; number of independent organism lineages |
|
||
| B | utilization rates; rate at which organisms are adopted into workflows |
|
||
| Su | users' assessment of organism usefulness and reliability |
|
||
|
||
**Goodhart risk:** capacity without access. K can rise while effective
|
||
agency falls (automated-capitalism scenario). Always report K together with B
|
||
(claims breadth) and P.
|
||
|
||
## F — Risk capacity
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | business formation rate; survival-adjusted re-entry rate after failure |
|
||
| St | availability of staged finance, insurance and public risk pools; bankruptcy regime leniency |
|
||
| B | share of first-time founders without inherited wealth |
|
||
| Su | perceived personal downside of a failed venture |
|
||
|
||
**Goodhart risk:** subsidized waste. Raising experiment counts while
|
||
externalizing harm. Pair with externalized-harm measures (constitutional
|
||
floor in GADM-0.1 § 21).
|
||
|
||
## G — Effective agency
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | restartability (time × cost × re-entry restrictions, GADM-0.1 § 12); economic mobility |
|
||
| St | exit options: provider switching costs, data and organism portability, membrane control held by users |
|
||
| B | actual switching, relocation and career-change rates |
|
||
| Su | perceived agency and perceived range of reachable futures |
|
||
|
||
**Goodhart risk:** formal options without real reach. Measures of legal
|
||
freedom that ignore material reachability. Organism "access" counted without
|
||
asking who controls the membrane (link #41).
|
||
|
||
## N — Diversity & composability
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | diversity of firms, products and lineages (entropy measures); interoperability test pass rates |
|
||
| St | standards coverage; market concentration per component class |
|
||
| B | cross-vendor recombination: products built from components of several sources |
|
||
| Su | developers' and founders' perceived lock-in |
|
||
|
||
**Goodhart risk:** fragmentation counted as diversity. Variety without
|
||
composability. Report both halves; GADM treats them jointly for this reason.
|
||
|
||
## Q — Epistemic capacity
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | correction latency (time from error to public correction); forecast accuracy; reproducibility rates |
|
||
| St | independent audit coverage; public access to evidence; concentration of information channels |
|
||
| B | share of AI-mediated decisions with retrievable provenance and explanation |
|
||
| Su | trust calibration: does trust track actual reliability? |
|
||
|
||
**Goodhart risk:** measuring trust instead of trustworthiness. High trust in
|
||
unreliable sources scores well on naive measures. Use calibration, not
|
||
levels.
|
||
|
||
## P — Domination concentration
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | ownership and market concentration (e.g. Herfindahl–Hirschman indices); concentration of lineage ownership |
|
||
| St | infrastructure, discovery and identity dependency; share of membranes controlled by a few operators |
|
||
| B | rate of self-preferencing and exclusion events |
|
||
| Su | perceived ability to contest decisions by dominant actors |
|
||
|
||
**Reporting requirement (GADM-0.3):** report domination per channel
|
||
(market, infrastructure, discovery/attention, membrane, lineage,
|
||
political), not only in aggregate. In GADM-0.3, a single dominant channel
|
||
(discovery at 0.80) coexisted with an aggregate P of only 0.32
|
||
(`models/GADM-0.3.md` § 2.6).
|
||
|
||
**Goodhart risk:** measuring wealth instead of power. GADM models power
|
||
separately from wealth: control of discovery, membranes and infrastructure can
|
||
be extreme at moderate wealth concentration.
|
||
|
||
## E — Ecological headroom
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | position relative to planetary boundaries; material and energy throughput |
|
||
| St | share of circular material flows; energy mix |
|
||
| B | compute energy use per unit of organism output (the metabolic draw of K_Ω) |
|
||
| Su | — (deliberately empty: ecological headroom is not a perception) |
|
||
|
||
**Goodhart risk:** monetary decoupling claimed as physical decoupling.
|
||
GADM-0.1 § 14 requires physical throughput, not value, as the basis.
|
||
|
||
## S — Social cohesion
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | incidence of violent conflict; legitimacy of institutions in contested decisions |
|
||
| St | channels for voice and appeal; plurality of recognition systems |
|
||
| B | civic participation; peaceful resolution rates |
|
||
| Su | generalized trust; expectation of future participation |
|
||
|
||
**Goodhart risk:** suppressed disagreement read as cohesion. GADM defines
|
||
cohesion as continued legitimate participation *under* conflict.
|
||
|
||
## A — Attention headroom
|
||
|
||
| Kind | Candidate indicators |
|
||
|---|---|
|
||
| O | self-directed time from time-use surveys (time not claimed by work, care obligations or engineered engagement) |
|
||
| St | share of attention-facing services funded by engagement; default notification and interruption regimes |
|
||
| B | interruptions per hour; share of attention spent in self-chosen versus recommended activities |
|
||
| Su | perceived control over one's own attention; reported time scarcity |
|
||
|
||
**Goodhart risk (critical):** counting *captured* attention as a resource.
|
||
Engagement metrics measure the opposite of A. Including A in VFAC rewards
|
||
extraction unless A measures **unclaimed, self-directed** attention
|
||
(`models/InformationOrganismsInGADM.md` § 4.1). In GADM-0.2, omitting A
|
||
understates attention harm (`models/GADM-0.2.md`, reading 5).
|
||
|
||
---
|
||
|
||
## Derived diagnostics
|
||
|
||
| Diagnostic | Definition | Caveat |
|
||
|---|---|---|
|
||
| VFAC | geometric mean of H, X, G, N, Q, S, E, 1 − P, A | Research instrument. Equal exponents damp near-collapse of one stock, so it never replaces floor checks. |
|
||
| Claims breadth B | share of output reaching the broad population | Says nothing about agency (dividend-abundance reading). |
|
||
| Floor status | share of periods in which a constitutional floor is violated | Floors are non-tradeable; do not aggregate them into VFAC. |
|