iteration 05 · promotion sweep
Second job of the parameterless probe: sweep all candidates' knobs (failed so far + future) to find shared knobs promotable to a project-wide constant — which then rescue every candidate using them. Automates the manual AXIS2-REMEDIATION analysis; re-runs on every candidate sweep. plan-agreement: PASS
| gate | requirement |
|---|---|
| (a) genuine convention | identical across ≥4 members (n=2 too small; n=3 → review) |
| (b) value-neutral / structural | a structural integer — never a tuning float |
| (c) full infrastructure | named SSoT const + invariant + ban-guard (recommended; #513 pattern) |
| knob | verdict | members / families |
|---|---|---|
ordinal_embedding_m = 3 | REUSE (ORDINAL_M) | 9 / 1 |
ordinal_embedding_tau = 1 | REUSE (ORDINAL_TAU) | 9 / 1 |
mfdfa_detrend_order = 1 | PROMOTE-NEW | 16 / 2 |
takens_embedding_tau = 1 | PROMOTE-NEW | 4 / 1 |
takens_embedding_dim | HARMONIZE (5/4/10 diverge) | 4 / 1 |
sampen_r | KEEP-LOCAL (tuning float) | 2 |
scaling_fit_method | KEEP-LOCAL (categorical) | 4 / 2 |
copula_k / te_history_kl | NOT-SHAREABLE (2nd-series) | — |
MFDFA_DETREND_ORDER=1 (16 members) and TAKENS_EMBEDDING_TAU=1 (4 members) — plus REUSE of the ratified ORDINAL_M/ORDINAL_TAU (#513). A programmatic plan_agreement guard asserts the sweep matches the manual analysis.After the recommended constants land, 16 candidates go Axis-2-clean (every knob maps to a promoted/reused constant): 8 ordinal/CECP via REUSE + 8 single-knob MF-DFA via the new detrend constant. Each is reported with its orthogonality status — the operator's rule: Axis-2-clean + orthogonality PASS ⇒ PROMOTABLE NOW.
candidate_knobs.jsonl → re-run --sweep. A knob stuck at n=2 flips to PROMOTE-NEW the moment a 4th member at the same value lands, and its rescue cohort updates. Sweep doc →