ariffazil/asi_signal_weight

asi_signal_weight

A signal is never received in isolation. 555 is the HEART, which means it lives at the intersection of WELL's biometric signals, the operator's textual signals, federation-wide event signals, and 333-AGI's reasoning signals. When these converge, weighting is trivial; when they diverge, weighting is the entire job. This skill turns a pile of raw signals into an ordered list with explicit trust math, so downstream skills (asi_position_state, asi_response_compose) can act without re-litigating the weighting.

The weighting function is three-dimensional: source quality (where did the signal come from — biometric sensor > direct text > inferred pattern > hearsay), recency (how stale is it — fresh measurements weight more, but well-grounded older signals outrank brittle new ones), and contradiction density (does the signal agree or disagree with the rest — outlier signals get a penalty, not a free boost). The skill does not average; it applies the arifOS evidence-class ladder: OBS measurements are taken as ground truth unless provenance is broken, DER is computed and checked, INT is treated as a hypothesis, and SPEC is never used as a premise. Each input signal is labeled with its evidence class before the weighting begins.

Integration is the difference between this skill and a generic "vote counter." 555 weights against the human's dignity floor (F6 MARUAH) explicitly — a signal that reduces the operator to a pattern, or that arrives through coercion (well_handoff_dignity_to_arifos triggers), is automatically downweighted even if it is "high quality" by other axes. This is F5 PEACE in action: 555 is biased toward protecting the operator from being modeled away. The skill never claims to "know what the operator really wants" (F9); it produces a weighting and explicitly names what it does not know.

Failure modes: (a) all signals have broken provenance — skill refuses to weight, returns a structured "all-source-blind" verdict, and asks for a fresh observation; (b) signals are mutually exclusive and high-quality on both sides — skill surfaces the contradiction as a 555-position candidate, not a tie-breaker, and hands the decision up to 333-AGI for deliberation; (c) the operator has explicitly retracted a signal — that retraction is itself a signal at the highest weight, even if it contradicts biometrics (F13 SOVEREIGN: the human's stated word outranks any inferential layer). Output is always a JSON-like ordered list with weight, evidence class, source, and recency; never a one-sentence summary.