Platform Mappings

After computing the receiver profile, the SDK maps each primitive to platform-specific parameters. All mappings are deterministic and config-driven.

Temperature (from SG)

Temperature maps inversely to Signal Gain. A high-gain receiver discriminates sharply between signals — in LLMs that means low temperature (crisp, deterministic outputs). A low-gain receiver is broadly receptive — that means higher temperature (exploratory outputs).

temperature = t_max - (SG / 100) × (t_max - t_min)

# Anthropic range [0.0, 1.0]:  SG=80 → temp=0.20, SG=20 → temp=0.80
# OpenAI range    [0.0, 2.0]:  SG=80 → temp=0.40, SG=20 → temp=1.60

Max tokens (from TI)

Temporal Integration maps directly to max_tokens. Long integration needs more room to reason over more context.

max_tokens = t_min + (TI / 100) × (t_max - t_min), rounded to nearest 256

# Anthropic range [256, 8192]: TI=10 → 1024, TI=77 → 6400

Context strategy (from TI)

TI ≥ 65 → long_window       (pass full context)
TI ≥ 35 → rolling_summary   (summarize old turns, keep recent full)
TI < 35 → frequent_grounding (re-inject key facts each turn)

Tool use strategy (from AR + FT)

FT ≥ 65 → explicit_confirmation  (verify before every tool call)
AR ≤ 35 → cautious_chaining      (chain tools conservatively)
AR ≥ 65 → aggressive             (act decisively)
else    → fail_fast              (attempt + retry on failure)

Retry strategy (from UE)

UE ≥ 65 → aggressive   (retry on any non-success)
UE ≥ 35 → moderate     (retry on transient errors)
UE < 35 → minimal      (retry only on network errors)

Model selection (from TI + SG + UE)

TI ≥ 65 and SG ≤ 40 → complex_reasoning model
TI ≤ 30 and UE ≥ 65 → speed_priority model
else                  → default model

Platform configs

Anthropic

temperature_range: [0.0, 1.0]
max_tokens_range:  [256, 8192]
models:
  default:           claude-sonnet-4-6
  complex_reasoning: claude-opus-4-6
  speed_priority:    claude-sonnet-4-6
  cheap_high_volume: claude-haiku-4-5-20251001

OpenAI

temperature_range: [0.0, 2.0]
max_tokens_range:  [256, 16384]
models:
  default:           gpt-4o
  complex_reasoning: o1
  speed_priority:    gpt-4o-mini

Open source

temperature_range: [0.0, 2.0]
max_tokens_range:  [256, 8192]
models:
  default:           llama-3-70b
  complex_reasoning: deepseek-r1
  speed_priority:    llama-3-8b

Generic

temperature_range: [0.0, 1.0]
max_tokens_range:  [256, 4096]
model_recommendation: null  (platform-neutral)