These are study-level defaults for scale, guards, the calm gate, priors, determinism, export, and ranking. Use them as a starting point. If you change any value, publish the change and keep the choice fixed within a study for auditability.
How to use these defaults
- Declare once per study. Do not mix recipes or units across runs in the same study.
- Tune, then publish. If you tune a parameter (e.g.,
beta_or_L), record the final value in the manifest. - Keep names canonical. Use the exact key names shown so manifests remain portable.
Plain ASCII defaults (copy-ready)
Scale & pooling
Unit = 100 # energy unit; name as E_unit (formation) or G_unit (Gibbs) in manifest
c = 0.5 # slope from contrast to alignment
gamma = 1 # stoichiometric weighting exponent
Guards & clamps
eps_a = 1e-6 # clamp margin so |a| <= 1 - eps_a
eps_w = 1e-12 # guard for W_r in RSI denominator
eps_fuse= 1e-6 # lane-weight renorm guard for fused signals
Unified calm gate (g_t)
kappa = 0.5 # weight on Delta_t = |Z_t - A_t|
mu = 0.8 # accumulation gain for Q_t
rho = 0.9 # memory in Q_t (exponential smoothing)
A_t = 1 / ( 1 + Z_t ) # default slow-track recipe
A_recipe = "1_over_1_plus_Z" # publish if using a different recipe
Z_recipe = "Var_EMA_beta" # or "Rolling_L"; publish the chosen recipe
beta_or_L= 0.2 # beta if EMA, L if Rolling; tune per study
Q0 = 0 # initial calm stock for Q_t (initialize Q_prev with Q0)
Priors (rapidity nudges)
alpha_max = 0.5 # cumulative bound on sum(alpha_k * Index_k)
Determinism & export
seed = 42
time_limit = 60 # seconds (tune per workload)
mip_gap = 1e-6 # if an ILP/MIQP solver is used
threads = 1 # fix for reproducibility
deterministic_mode = true
rounding_mode = "bankers" # half-to-even at export only
rounding_decimals = 6
Ranking policy
ranking_policy = "RSI_env" # if no gate is used, rank by RSI
Manifest publication (required fields)
Always publish:
{ Unit, c, gamma, eps_a, eps_w, eps_fuse,
kappa, mu, rho, Z_recipe, A_recipe, Q0, beta_or_L,
alpha_max, seed, time_limit, mip_gap, threads, deterministic_mode,
rounding_mode, rounding_decimals, ranking_policy }
Also publish the chosen lens name for Unit (e.g., E_unit for formation contrast, G_unit for Gibbs lens).
Navigation
Previous – Identities, Neutrality, Inverses (2.10)
Next – Minimal Pseudocode (2.12)
SHUNYAYA 