bench(worldmodel): the goal metric is also a defect — rho 0.25 at its best

[aurascoper]

Node 24's bounded check before any design decision. Node 22 showed the true latent cost yields 76%, so Spearman(true latent cost, true position cost) should be substantial.

It is not.

Result

150 episodes × 64 candidates × 3 seeds, true-env rollouts — no predictor involved:

goal latent	median ρ	mean ρ	ρ > 0.5	ρ < 0	node 22 success
one-frame (online goal)	0.2514	0.2735	22.4%	17.8%	76.0%
incumbent (target goal)	0.1570	0.0744	26.7%	41.3%	37.5%
predictor cost (node 24)	−0.018	—	—	—	35.0%

The latent metric is weak even at its best

One-frame reaches only ρ = 0.25. It still yields 76% because the CEM needs only to find a good candidate among 64, and ρ=0.25 enriches the elite set enough — but the metric is not measuring what success measures.

The two-encoder frame inverts the ranking rather than weakening it

With the target-encoded goal, mean ρ collapses to 0.074 and 41.3% of episodes have negative correlation — in four episodes in ten the incumbent cost ranks candidates backwards.

That is the mechanism behind node 22's 76% vs 37.5%, and it is worse than "a weaker signal": it is a signal with the wrong sign nearly half the time.

Two independent cost defects, not one

1. the latent distance is a weak proxy — ρ ≈ 0.25 at best
2. the predictor destroys even that — ρ = −0.018

Node 24 correctly rejected forward-model accuracy and named rank preservation. This refines it: a perfectly rank-preserving predictor on this metric would cap near 76%, not the 93.5% ceiling. The remaining ~17.5 points are the metric's own weakness and no forward-model change recovers them.

The complete measured chain

true position cost                        93.5%
true latent cost, one frame  (rho  0.25)  76.0%
true latent cost, two frames (rho  0.074) 37.5%   <- 41% sign-inverted
predictor cost               (rho -0.018) 35.0%

Each step is a measured loss with a named cause.

Next

Unchanged in kind and now fully specified — the design decision node 13 reserved for the user, with both halves quantified.

A control-aware goal metric must lift ρ from 0.25. A rank-preserving training signal must stop the predictor driving it to zero. They are separable and should be attempted separately, since fixing only the predictor caps at 76% and fixing only the metric leaves the predictor destroying it.

The one-frame goal is a prerequisite for either: free, removes a 41.3% sign inversion, and necessary but not sufficient per nodes 22 and 25.

Ledger-only. Verified: smoke 2 runs; ledger and ADR gates green.