Panel estimates never enter a score
A panel reviews selected dimensions from the source-backed evidence and audits the registry. Its estimates do not change the indicator score or confidence.
run 2026-10-03T07-39-51-747Z-in_session2 rounds3 panelistsworking sessiondataset 9.0.1348 of 455 cells carried to 9.3.0full runcoverage ≤ 0.5
Each panelist holds a fixed stance
The stance explains why panelists disagree.
In-session one-vendor panel under D154(b) as amended by D155, run on 2026-10-03 by a Claude Code orchestrating session (Claude Opus 5.5). Three panelists, each stance a separate set of contexts: institutionalist on claude-opus-5-5, bottom_up on claude-opus-5-5, wealth_sceptic on claude-sonnet-5-5, all Anthropic, each through Claude Code subagents. Each panelist answered in batches of 25 countries, one fresh context per batch and per round, so no context held two stances and none saw another panelist's text except the anonymised round-1 summary the round-2 prompt carries. Two rounds plus the indicator audit, at the code default scope (maxCoverage 0.5): 455 country-dimension cells across all 125 countries. Prompts are byte-identical to a gateway run's, built by pnpm bench delphi --in-session with prompt version 3; the system rules were given as a file each panelist read first. Every batch confirmed it read every prompt in full and wrote every answer by hand. No call failed and none was rerun into this file; an earlier attempt on prompt version 2 was discarded because panelists condensed prompts. One vendor means the spread measures stance, not vendor, and shared model-family blind spots are hidden by a narrow IQR (D155). An OpenAI panelist through the codex CLI (gpt-5.6-sol) can be added as a follow-up multi-vendor run, which supersedes this one under D155. Estimates are an interpretation layer beside the indicators and never enter score.
| Stance | Model | Panelist id |
|---|---|---|
| Institutionalist | claude-opus-5-5 | institutionalist@claude-opus-5-5 |
| Bottom-up analyst | claude-opus-5-5 | bottom_up@claude-opus-5-5 |
| Wealth sceptic | claude-sonnet-5-5 | wealth_sceptic@claude-sonnet-5-5 |
Where the panel disagreed
A cell is marked when the middle half of the panel spans more than 25 points.
No cell has an interquartile range above 25 points.
What moved between rounds
A negative IQR shift means the panel narrowed.
| Country | Dimension | Median | Median shift | IQR | IQR shift |
|---|---|---|---|---|---|
| United Arab Emirates | Shared Purpose | 33.0 | -5.0 | 2.5 | -5.0 |
| China | Shared Purpose | 33.0 | -5.0 | 2.0 | -2.0 |
| Myanmar | Shared Purpose | 30.0 | -5.0 | 6.5 | -4.0 |
| Sudan | Trust | 20.0 | 5.0 | 4.5 | -3.0 |
| Tajikistan | Adaptability | 36.0 | -4.0 | 4.5 | -3.0 |
| Tunisia | Shared Purpose | 34.0 | -4.0 | 1.5 | -2.5 |
| Burkina Faso | Trust | 32.0 | -3.0 | 1.0 | -2.0 |
| Burkina Faso | Shared Purpose | 31.0 | -3.0 | 1.0 | -1.5 |
| Cuba | Shared Purpose | 32.0 | -3.0 | 1.0 | -1.5 |
| Cyprus | Shared Purpose | 47.0 | -3.0 | 2.0 | -3.0 |
| Sudan | Agency | 18.0 | 3.0 | 5.0 | -3.0 |
| El Salvador | Adaptability | 47.0 | -3.0 | 2.0 | -1.5 |
| El Salvador | Shared Purpose | 42.0 | -3.0 | 1.0 | -1.5 |
| Tanzania | Anticipation | 25.0 | -3.0 | 1.0 | -2.0 |
| Tanzania | Trust | 52.0 | -3.0 | 2.0 | -2.0 |
| United States | Shared Purpose | 47.0 | -3.0 | 2.0 | -2.0 |
| United Arab Emirates | Anticipation | 60.0 | -2.0 | 2.0 | 1.0 |
| Australia | Anticipation | 74.0 | 2.0 | 2.5 | -2.5 |
| Burundi | Trust | 26.0 | -2.0 | 1.0 | -1.5 |
| Burundi | Shared Purpose | 26.0 | -2.0 | 1.0 | -1.5 |
| Belgium | Trust | 70.0 | -2.0 | 1.5 | -1.5 |
| Belgium | Shared Purpose | 60.0 | -2.0 | 2.0 | -4.0 |
| Bulgaria | Building | 28.0 | -2.0 | 2.0 | -1.0 |
| Canada | Anticipation | 72.0 | 2.0 | 2.0 | -2.0 |
| Canada | Coordination | 78.0 | -2.0 | 2.0 | -3.0 |
How the panel rated each indicator
Each indicator gets a class, a validity rating and a wealth-proxy risk rating. The weakest validity comes first.
| Indicator | Registry class | Panel class | Construct validity | Wealth proxy risk | Registry prior | Redundancy votes |
|---|---|---|---|---|---|---|
| Intentional homicide rate | O | O | 0.22 | 0.68 | 0.35 | none |
| Business share of R&D | I | I | 0.22 | 0.57 | 0.70 | none |
| Unemployment rate | O | O | 0.23 | 0.28 | 0.25 | Long-term unemployment share (1) |
| Voice and accountability | P | P | 0.27 | 0.75 | 0.35 | none |
| Scientific articles | O | Idisputed | 0.30 | 0.78 | 0.40 | none |
| High-technology exports | O | O | 0.30 | 0.45 | 0.40 | Economic fitness (3) |
| Income inequality | O | Idisputed | 0.30 | 0.35 | 0.20 | none |
| Regulatory quality | P | P | 0.32 | 0.92 | 0.60 | Government effectiveness (3) |
| Logistics performance | P | P | 0.32 | 0.77 | 0.50 | Border compliance time to export (3) |
| Export diversification | C | O | 0.32 | 0.53 | 0.30 | New export products (3) |
| Manufacturing value added | O | O | 0.32 | 0.28 | 0.20 | none |
| Logistics infrastructure quality | P | P | 0.32 | 0.80 | 0.50 | Electricity connection speed (1) |
| Tax revenue | O | O | 0.33 | 0.63 | 0.30 | none |
| Adults doing digital tasks | C | Odisputed | 0.35 | 0.77 | 0.40 | none |
| Trust in public institutions | P | P | 0.35 | 0.40 | 0.20 | none |
| Resident patent applications | O | O | 0.35 | 0.73 | 0.45 | Resident trademark applications (1), Resident industrial design applications (1) |
| Resident industrial design applications | O | O | 0.37 | 0.57 | 0.30 | Resident trademark applications (3), Resident patent applications (1) |
| University spinouts | C | O | 0.37 | 0.67 | 0.35 | none |
| Procedures to start a business | C | I | 0.38 | 0.35 | 0.15 | Time to start a business (3) |
| Control of corruption | P | P | 0.38 | 0.87 | 0.55 | Rule of law (3), Bribery incidence (1) |
| University-industry collaboration | C | P | 0.40 | 0.72 | 0.40 | Public-private collaboration (1) |
| Research citation impact | O | O | 0.40 | 0.58 | 0.60 | none |
| Economic fitness | O | O | 0.40 | 0.72 | 0.45 | High-technology exports (3) |
| Political polarisation | P | P | 0.40 | 0.33 | 0.05 | none |
| Perceived control over life | P | P | 0.42 | 0.33 | 0.20 | none |
| Government effectiveness | P | P | 0.42 | 0.92 | 0.60 | Regulatory quality (3) |
| Resident trademark applications | O | O | 0.42 | 0.62 | 0.30 | Resident industrial design applications (3), Resident patent applications (1) |
| Long-term unemployment share | O | O | 0.42 | 0.37 | 0.20 | Unemployment rate (2) |
| Time to start a business | C | I | 0.43 | 0.40 | 0.20 | Procedures to start a business (3) |
| Court case clearance rate | O | O | 0.43 | 0.43 | 0.20 | Time to enforce a contract (3) |
| New public software repositories | O | O | 0.43 | 0.58 | 0.70 | none |
| Rule of law | P | P | 0.45 | 0.87 | 0.55 | Control of corruption (3), Government compliance with the courts (1) |
| Time to enforce a contract | C | Cdisputed | 0.45 | 0.43 | 0.20 | Court case clearance rate (2) |
| Human Capital Index | C | O | 0.45 | 0.87 | 0.50 | none |
| Venture deals | C | O | 0.45 | 0.68 | 0.60 | Early-stage entrepreneurial activity (2) |
| Early-stage entrepreneurial activity | C | Odisputed | 0.45 | 0.33 | 0.15 | Venture deals (1) |
| Long-horizon research commitments | C | C | 0.47 | 0.42 | 0.30 | none |
| Border compliance time to export | C | Cdisputed | 0.47 | 0.55 | 0.30 | Logistics performance (2) |
| Tolerance of entrepreneurial failure | P | P | 0.47 | 0.32 | 0.10 | none |
| Electricity connection speed | C | Cdisputed | 0.47 | 0.48 | 0.25 | Logistics infrastructure quality (1) |
Evidence the panel asked for
Panelists name evidence they want to see. The list feeds the collection agenda.
| Evidence | Mentions | Dimensions |
|---|---|---|
| government foresight capacity dataset | 75 | 1 |
| government foresight capacity register | 50 | 1 |
| volunteering rates from open microdata | 49 | 1 |
| hostile polarisation measure | 32 | 1 |
| volunteering microdata | 29 | 1 |
| court case clearance rates | 28 | 1 |
| court case clearance rate | 26 | 1 |
| government foresight capacity record | 24 | 1 |
| hostile polarisation measure separating camps from calm | 23 | 1 |
| multi-year public research grant register | 22 | 1 |
| grant register with multi-year award durations | 21 | 1 |
| grant register with multi-year commitments | 21 | 1 |
| hostile-polarisation measure | 21 | 1 |
| v-dem polarization check value for 2023-2025 | 19 | 1 |
| long-horizon public research grant register | 19 | 1 |
| grant register with award durations | 19 | 1 |
| hostile polarization measure | 18 | 1 |
| grant register with multi-year awards | 17 | 1 |
| belonging item that reads the same across regimes | 17 | 1 |
| belonging measure | 15 | 1 |
| cross-agency delivery records | 14 | 1 |
| court case clearance rate from case records | 12 | 1 |
| long-term unemployment share | 12 | 1 |
| belonging item comparable across regime classes | 11 | 1 |
| long-term unemployment share from ilostat | 11 | 1 |
| generalised interpersonal trust (a165) | 11 | 1 |
| cooperation beyond the in-group (g007_34_b) | 11 | 1 |
| volunteering rates from an inspectable source | 10 | 1 |
| volunteering rates | 10 | 1 |
| budget execution fidelity | 10 | 1 |