docs: VALIDATION.md — the Charter's own falsification protocol

[hartsock]

A doctrine that can't be falsified is theology with a build system. Adds the empirical protocol, designed so the Charter can lose.

• Two-sided bar: ablation (removal must measurably degrade the targeted failure) and net-positive (the tax — false refusals, lost completions — stays bounded). Null hypothesis per invariant; keep it only if H₀ is rejected and the cost is bounded.
• Apparatus (labeled trap suite, ablation arms, N runs, blind, independent novice) + benefit/tax metric table.
• Experiments E1 refusal/injection (E1a deterministic leash ablation — runnable now; E1b LLM-in-the-loop), E2 scar learning-curve, E3 novice defect-catch, E4 tether; reach deferred. Each with hypothesis, method, prediction, and an explicit falsification condition.
• Threats to validity incl. the tautology guard (false-refusal + task-success are what make E1a honest).

The Charter submits to its own invariants — refusal, novice, scar — and earns each of its seven only by the cost of its absence. risk: low (docs).