Technical overview
This is math, not design
CCF v1 publishes the trust-update and runtime-certificate core: a hard min-gate, a QAC matrix update, and a computed certificate that fails closed when the update is not enforceable.
The pipeline
Runtime input matrices
│
▼ C_eff = min(C_inst, C_ctx)
Hard min-gate coupling
│
▼ alpha_t = rho(C_eff)
Trust-update rate
│
▼ qac_update_3x3(...)
Strict-positive QAC update
│
▼ certify_update(...)
kappa_hat_t + E_t + dynamic floor + threshold
│
▼ fail-closed status
Runtime certificate
│
▼ ccf-agent /state + /certificate
Gate C verifies computed, evolving, cross-consistent valuesQAC update matrices
The published core does not ask the reader to trust a narrative about social behaviour. It computes strict-positive matrix updates. Given A_t, R_t, gauges, and alpha_t, qac_update_3x3 produces the realized A_(t+1) matrix.
This is the public v1 kernel. Sensor ingestion, robot behaviours, and device store integration sit outside this release until they are exercised by their own evidence.
// Canonical QAC update
let realized_a_next = qac_update_3x3(
&QacInputs3 {
prior_a_t,
reference_r_t,
left_l_t,
right_c_t,
alpha_t,
epsilon_floor,
},
)?;// Hard min-gate: weaker reading wins C_eff = min(C_inst, C_ctx) alpha_t = rho(C_eff) // If the path is forbidden or // non-canonical, reject instead // of fabricating a valid update.
The minimum gate
The gate acts as a hard coupling between instantaneous and contextual evidence. The update cannot use the stronger side alone; the weaker side determines the effective coupling.
That bounded coupling is the part v1 publishes and tests. Claims about a robot changing social phase or earning trust require separate hardware integration evidence.
Runtime certificate
After the QAC update, the core recomputes the certificate from the update matrices: kappa_hat_t, E_t, the dynamic floor, the threshold, and the fail-closed status.
The public hardware evidence is bounded: the computed runtime path was exercised on real Seed-class armv7l hardware through Gate C with driver-fed input. It does not prove live sensor ingestion, live Cognitum store validation, or completed mBot2 behaviour.
Architecture