cvprofiles · construct-validity profile

Patience — cells demeaned + LLM

Admissible measurement set under a stated nomological network, and the image of β on that set. Empty sets and wide ranges are scientific features.

Construct-identified range

[L, U] · method = min/max B*

[0.244910, 0.556052]

Width = 0.311142 · 3 survivors · bootstrap percentile band = [0.164722, 0.622665]

Target β = corr_y on outcome gps_patience_cell. Range uses only measures in M*; rejected β values are shown below but never enter [L,U].

Run freeze

run_id02005bf2737740e3b37636e837801f40ef68f3b65a2d59fd7e929d89b8666db8
package3.0.2
scores_hashb1febee0f30cb536903ad84abfa4b24b41192aaf92bbda74a3f066d01d3b4af9
network_hashf0d1582c64237be2b11057f75e055a94ccbf535fcde9d99506fbda76de1e6031
beta_hash93ecc94543944de5860e399f0b91f9f1c5038e5b4d418e5f3444d3b0a8bc23c7
δ0.0
seed20260814
networkpatience_cells_llm_extension

Bootstrap inference · additive to headline range

Units-only Efron resampling; percentile band over non-empty replicates. The headline [L, U] above remains min/max β(M*) on the full sample.

replicates1000
seed20260814
quantiles[0.025, 0.975]
non-empty1000
empty0 (0.000)
degenerate0 (0.000)
band[0.164722, 0.622665]

Uncertainty band · additive to headline range

Per-side α/2 quantiles over non-empty bootstrap replicates (α = 0.1). Heuristic band — not a confidence interval, not a holdout-robustness band. The headline [L, U] above remains min/max β(M*) on the full sample.

α0.1
κ2.0
quantiles[0.05, 0.95]
band[0.178677, 0.614207]
empty rate0.000
degenerate rate0.000

Boundary attribution · |margin| ≤ κ·SE

measuremarginSEboundary
m_wvs_q13 0.051086 0.051007 yes
m_wvs_q14 -0.564248 0.044344 no
m_composite -0.249511 0.057936 no
m_llm_patience_llama 0.321997 0.033693 no
m_llm_patience_phi 0.440249 0.028916 no
m_noise -0.230693 0.046268 no

Admission frequency p̂m

measurep̂m
m_wvs_q13 0.830
m_wvs_q14 0.000
m_composite 0.000
m_llm_patience_llama 1.000
m_llm_patience_phi 1.000
m_noise 0.000

Heuristic uncertainty band over non-empty bootstrap replicates (selection uncertainty on the pooled sample; holdout verdict is a full-sample point finding outside the band). Not a confidence interval; not a holdout-robustness band. Boundary attribution: |margin_m| <= kappa * SE_m.

θ-grid sensitivity · diagnostic viewport

λ scales threshold magnitudes only. λ is never auto-selected; the headline is always λ=1.0. This grid is not part of the freeze preimage.

λ|M*|emptyLU
0.5 3 no 0.244910 0.556052
1.0 3 no 0.244910 0.556052
1.5 2 no 0.416622 0.556052
2.0 2 no 0.416622 0.556052

θ-anchors · pre-data audit

One pre-data anchor per restriction (hash 07533ac8ce87…). The engine enforces completeness + hash; "pre-data" is a process commitment, not a temporal claim.

restrictionkindpre-dataanchor
conv_edu_cell literature yes falk2018_table_v — Same education implication applied within country (cell means). θ=0.20.
mono_edu_cell literature yes falk2018_table_v — Leftover within-country rank-monotone education restriction. θ=0.15.

Menu admission

M* size = 3 / 6. Slack rule: admit if sr(m) ≥ −δ for all r.

measure status β(m) failed restrictions
m_wvs_q13 admissible 0.244910 —
m_wvs_q14 rejected -0.325797 conv_edu_cell
m_composite rejected -0.026645 conv_edu_cell
m_llm_patience_llama admissible 0.416622 —
m_llm_patience_phi admissible 0.556052 —
m_noise rejected -0.002776 conv_edu_cell

Slack matrix sr(m)

Positive / zero ≈ satisfied at δ=0.0. Negative = violation magnitude.

measure conv_edu_cell mono_edu_cell
m_wvs_q13 0.051086 0.060120
m_wvs_q14 -0.564248 -0.498965
m_composite -0.249511 -0.221653
m_llm_patience_llama 0.321997 0.358320
m_llm_patience_phi 0.440249 0.523252
m_noise -0.230693 -0.165338

Network (researcher-owned)

Engine does not invent restrictions. Agent may author oracle R for synthetic DGPs only.