← Back to blog
June 28, 2026Colm Byrne

A Child Introduces a New Caregiver: How Sponsored Trust Bridging Works

Aoife is eight years old. She has been interacting with the companion robot in her home for four months. She talks to it. She shows it drawings. She tells it when she is having a bad day, and the robot dims its LED and speaks more softly. The robot's accumulated coherence for Aoife's context is 0.82 -- deep familiarity, high trust, a behavioural envelope wide enough for animated conversation, playful servo movements, and full verbal engagement.

Mary is the new night nurse. She arrives at 8pm on a Tuesday. The robot has never encountered Mary. Her context coherence is 0.05 -- essentially a stranger. If Mary approaches the robot alone, it enters Phase I (ShyObserver). Minimal movement. Dim LED. Near-silent. It does not know Mary. It does not trust Mary. It has no basis for trusting Mary. This is correct behaviour.

But Aoife is standing right there. And Aoife says: "This is Mary, she's nice."

What should happen next? There are three options, and only one of them is right.

Option 1: Cold-start. Ignore Aoife's introduction entirely. Treat Mary as a stranger regardless. This is safe but wrong. Aoife has spent four months building a relationship with this robot. Her social context -- her presence, her familiarity, her trust -- carries information. Discarding it discards the value of the relationship.

Option 2: Manual override. A caregiver or administrator sets Mary's trust level to a predefined value. This is expedient but dangerous. It violates the fundamental CCF invariant: trust must be earned, never assigned. Manually setting trust creates a trust level with no accumulation history, no interaction count, no decay floor. The number exists but has no backing. It is fiat trust.

Option 3: Sponsored bridging. Aoife's accumulated trust scaffolds a bounded, transient, decaying bridge to Mary. The robot is less startled by Mary, less withdrawn, more willing to engage. But the bridge is not trust. It is a scaffold that allows trust to begin accumulating faster. If Mary earns it, the scaffold promotes to genuine trust. If Mary does not, the scaffold decays and disappears.

The mechanism is described in Claims AP through AS of US Provisional 64/039,655, sections [E8-0001] through [E8-0010].

The Bridge Formula

The initial bridge mass is computed from four inputs:

b_0 = min(b_max, alpha * q_event * C_S * (1 - C_T))

Where:

  • b_max is the maximum bridge mass, defaulting to 0.30. No single sponsor can create a bridge larger than this, regardless of how trusted they are.
  • alpha is the bridge scaling factor, defaulting to 0.50. This is a conservative coefficient that ensures even optimal introductions produce modest bridges.
  • q_event is the quality of the introduction event, a scalar between 0 and 1.
  • C_S is the sponsor's accumulated coherence. Aoife's is 0.82.
  • C_T is the target's current coherence. Mary's is 0.05.
  • (1 - C_T) ensures the bridge is largest when the target is least familiar. If Mary already had high coherence, she would not need a bridge.

The quality metric q_event is not a subjective rating. It is a function of observable interaction patterns:

| Introduction Type | Approximate q_event | |-------------------|---------------------| | Single co-presence (sponsor and target in same context) | ~0.30 | | Verbal introduction ("this is Mary") | ~0.60 | | Verbal + sustained co-presence (5+ minutes) + low tension | ~0.85 | | Repeated escort over multiple sessions | ~0.95 |

Aoife's introduction is verbal, with sustained co-presence and low tension. q_event is approximately 0.85.

Computing the bridge:

b_0 = min(0.30, 0.50 * 0.85 * 0.82 * (1 - 0.05))
b_0 = min(0.30, 0.50 * 0.85 * 0.82 * 0.95)
b_0 = min(0.30, 0.331)
b_0 = 0.30  (capped at b_max)

The raw computation yields 0.331, but the cap limits it to 0.30. Aoife's trust is high enough and her introduction is warm enough to hit the maximum bridge mass. This is the best possible start for Mary. But it is still only 0.30.

What the Bridge Does

The bridge does NOT overwrite Mary's coherence accumulator. Mary's C_ctx remains 0.05. The bridge is a PARALLEL structure -- a transient addendum that elevates the effective coherence when computing the behavioural envelope.

Without bridge:

C_eff = min(C_inst, C_ctx) = min(0.65, 0.05) = 0.05
Phase: I (ShyObserver)
output: {motor: 0.05, LED: 0.08, audio: 0.04, verbal: 0.03}

With bridge:

C_eff_base = min(C_inst, C_ctx) = min(0.65, 0.05) = 0.05
bridge_contribution = bridge_mass * adjustment_factor = 0.30 * 0.24 = 0.072
C_eff_adjusted = 0.05 + 0.072 = 0.122
Phase: I (ShyObserver, but near Phase I/II boundary)
output: {motor: 0.12, LED: 0.15, audio: 0.10, verbal: 0.08}

The robot is still reserved. It is still in Phase I. But it is not frozen. There is perceptible movement, visible LED activity, faint audio. To Aoife: the robot is shy around Mary, but not scared. It is meeting someone new, cautiously but not hostilely. This matches the social expectation. The robot's behaviour tracks Aoife's introduction: "she's nice" does not mean "trust her completely." It means "give her a chance."

Bridge Decay

The bridge is transient. It decays at 10% per interval when the sponsor is not present.

b_t = b_0 * (1 - 0.10)^(intervals_since_sponsor)

Where an interval is one day by default (configurable per deployment). If Aoife is present each day, the decay counter resets. If Aoife goes to school and Mary is alone with the robot during the day, the bridge decays:

| Day | Aoife Present? | Bridge Mass | Effective Contribution | |-----|----------------|-------------|------------------------| | 0 | Yes (evening) | 0.30 | 0.072 | | 1 | No (school day) | 0.27 | 0.065 | | 2 | No | 0.24 | 0.058 | | 3 | Yes (evening) | 0.24 (reset) | 0.058 | | 4 | No | 0.22 | 0.053 | | 7 | No | 0.19 | 0.046 | | 10 | No | 0.13 | 0.031 | | 14 | No | 0.06 | 0.014 | | 17 | No | 0.03 | ~0.007 (effectively expired) |

After approximately two weeks without sponsor presence, the bridge is functionally zero. If Mary has not earned direct positive interactions during this window -- if the robot has not accumulated genuine familiarity with Mary through repeated, tension-free encounters -- the bridge dissolves and Mary returns to stranger status.

But if Mary HAS earned interactions during the bridge window, something important happens: the genuine trust accumulation is now higher than it would have been without the bridge. Because the bridge enabled the robot to engage more (higher motor amplitude, more LED expression, more verbal output), Mary had the opportunity for positive interactions that would not have occurred in full Phase I lockdown. The bridge scaffolded the conditions for natural trust building.

This is the biological analogy that makes the mechanism intuitive. When a child introduces a friend to their parent, the parent is warmer than they would be to a complete stranger. The warmth is not trust. It is the child's endorsement creating conditions for trust to form. If the friend turns out to be unpleasant, the parent's warmth fades rapidly. If the friend is kind and consistent, the warmth transitions into genuine familiarity. The endorsement was a scaffold, not a foundation.

Accelerated Decay Under Adverse Conditions

If Mary's interactions produce negative signals -- if tension spikes during her presence, if the robot's reflexive pathway detects environmental instability correlated with her context -- the bridge decays 2-5 times faster than the baseline rate.

accelerated_decay_rate = base_decay * tension_multiplier
tension_multiplier = 1.0 + (tension_during_target_context * 4.0)

At moderate tension (0.25), the multiplier is 2.0 -- twice the normal decay. At high tension (0.50), the multiplier is 3.0. At severe tension (0.75), the multiplier is 4.0.

This means the bridge is self-correcting. If Aoife's introduction was well-intentioned but Mary's actual behaviour is harsh, inconsistent, or stress-inducing, the bridge erodes rapidly. The robot's withdrawal is not punitive -- there is no punishment mechanism. The withdrawal is mathematical: the bridge mass is decaying faster than normal because the conditions that sustain it (low tension, positive interactions) are absent.

A family observing this pattern can see: the robot was initially warm toward Mary (bridge active), then gradually withdrew (bridge decayed under adverse conditions). This observable behaviour is informative. It tells the family something about the interaction dynamics that they might not have noticed directly -- especially in contexts where the family is not always present during Mary's shifts.

The Eldercare Scenario

Mrs. O'Brien is 83 years old and lives in a nursing home in Limerick. She has been interacting with the companion robot for six months. Her coherence is 0.79 -- one of the highest in the facility. The robot is fully engaged with her: animated conversation, expressive servo movements, responsive to her voice patterns.

A new aide, Declan, starts this week. He will be assisting Mrs. O'Brien with morning routines. The introduction happens during a supervised transition: Mrs. O'Brien's familiar daytime aide, Sinead (coherence 0.71), introduces Declan to the robot by demonstrating the morning routine while Declan observes.

Sinead's introduction quality: verbal + sustained co-presence + low tension + demonstrated routine context. q_event is approximately 0.85.

b_0 = min(0.30, 0.50 * 0.85 * 0.71 * (1 - 0.03))
b_0 = min(0.30, 0.50 * 0.85 * 0.71 * 0.97)
b_0 = min(0.30, 0.293)
b_0 = 0.293

Sinead's coherence is lower than Aoife's (0.71 vs 0.82), so the bridge mass is slightly below the cap. Declan starts with a bridge of 0.293 -- not quite the maximum, but close.

Over the next week, Declan performs the morning routine. The bridge decays on days when Sinead is not present (weekends), but resets when Sinead works alongside Declan (weekdays). Meanwhile, Declan's genuine interactions accumulate:

| Day | Bridge Mass | Declan's C_ctx | Combined Effective | |-----|-------------|----------------|-------------------| | 1 | 0.293 | 0.03 | ~0.10 | | 3 | 0.293 (reset) | 0.08 | ~0.15 | | 5 | 0.293 (reset) | 0.14 | ~0.21 | | 7 | 0.237 (weekend decay) | 0.19 | ~0.25 | | 10 | 0.293 (reset Monday) | 0.26 | ~0.33 | | 14 | 0.237 (weekend) | 0.33 | ~0.39 | | 21 | 0.293 (reset) | 0.42 | ~0.49 |

By week three, Declan's genuine coherence (0.42) exceeds the bridge contribution. The bridge is becoming redundant. By week four, Declan's coherence alone places him in Phase II territory. The scaffold has done its job. The trust is now earned, not borrowed.

If the bridge had not existed, Declan would have spent the first two weeks in full Phase I lockdown -- the robot barely acknowledging him. His interactions would have been limited by the robot's minimal output. The trust accumulation would have been slower because fewer positive interactions were possible. The bridge compressed the onboarding timeline from approximately 5-6 weeks (cold start) to 3 weeks (scaffolded start).

Why the Cap Matters

The bridge cap (b_max = 0.30) is not arbitrary. It is set so that no single sponsor can elevate a stranger beyond Phase I into Phase II through bridge mass alone. Phase II requires effective coherence of approximately 0.35 (with hysteresis). A bridge of 0.30 added to a baseline of 0.05 produces 0.122 -- well short of the Phase II threshold.

This means the bridge can make Phase I more comfortable (less frozen, more perceptible output) but cannot bypass the Phase I/II boundary. The transition to Phase II requires genuine accumulated trust. The sponsor can scaffold conditions for faster accumulation. The sponsor cannot grant the transition itself.

This is a non-negotiable invariant. If the cap were set high enough to cross phase boundaries, a single introduction could produce full companion behaviour with a stranger. The trust would be entirely borrowed, with no accumulation history, no interaction count, no resilience. A stranger with Phase II access is a safety violation. The cap ensures this cannot occur.

For the mathematical guarantee that trust cannot be amplified beyond what is earned, see Sinkhorn-Knopp for Trust and The Compositional Closure Proof.

The Biological Parallel

The sponsor bridge is modelled on an observable human social phenomenon. When a child introduces a friend to their dog, the dog is warmer than it would be to a stranger approaching alone. The child's presence reduces the dog's startle response. The child's calm voice reduces tension. The dog engages more readily.

But the dog is still cautious. It sniffs the stranger. It watches the stranger's movements. If the stranger is gentle and consistent, the dog warms up over subsequent visits. If the stranger is rough or erratic, the dog withdraws despite the child's presence.

The critical feature: the child's endorsement DECAYS. If the child is not present for several visits, the dog reverts toward stranger behaviour with the friend. The child's presence was the scaffold. Without it, the relationship must stand on its own direct history.

CCF's sponsor bridge replicates this dynamic mathematically. The sponsor's coherence creates a transient elevation. The elevation decays without sponsor co-presence. The target must earn direct trust to maintain engagement. The biological analogy is not metaphorical -- the mathematics was designed to reproduce the observable dynamics of sponsored social introductions.

Applications Beyond Childcare

The sponsor bridge mechanism applies to any context where a trusted entity introduces a new entity to the robot's social field:

Hospital ward transfer. A patient is moved from Ward A to Ward B. The Ward A nurse (trusted by the robot) introduces the Ward B nurse during the transfer. The bridge scaffolds the robot's engagement with the new nurse, compressing the acclimatisation period.

School classroom transition. A child moves from one classroom to another. The previous teacher introduces the child to the robot in the new classroom. The bridge carries forward the social context, preventing a full cold start.

Home care aide rotation. Regular aides introduce substitute aides during supervised handoff shifts. The bridge ensures the robot is not fully withdrawn during the substitute's solo shifts, enabling the substitute to build direct trust.

Family visits to care facilities. Family members (highly trusted) introduce new visitors (grandchildren from abroad, extended family, family friends). The bridge enables the visitor to engage with the robot during the visit without the full Phase I lockdown that would otherwise apply.

In each case, the bridge compresses the cold-start period, maintains behavioural continuity for the person being cared for, and preserves the fundamental invariant: trust cannot be created from nothing. It can only be scaffolded by those who have already earned it.

The full implementation is available in ccf-core on crates.io. For the causation packet that records bridge creation and decay events, see Why Did the Robot Back Away?. For how bridge counterfactuals enable transition planning, see 'It Would Have Been Warmer If...'. For multi-sponsor quorum requirements in high-stakes transitions, see Two Sponsors Required.

FAQ

Can a sponsor create a bridge without being physically present -- for example, through a recorded video introduction?

No. The bridge formula requires real-time co-presence: the sponsor's context key must be active simultaneously with the target's context key. A recorded video does not produce a live context key for the sponsor. The robot cannot verify that the sponsor endorsed the target in real time. This prevents replay attacks where a recorded introduction is used to create bridges without the sponsor's current consent.

What happens if the sponsor's own trust drops while a bridge is active?

The bridge mass is computed at creation time and does not retroactively adjust to the sponsor's current coherence. However, if the sponsor's context coherence drops below 0.30 (configurable threshold), the bridge enters accelerated decay regardless of other conditions. This prevents a degraded sponsor from sustaining a bridge they could no longer create. If the sponsor's trust drops to zero (full reset), all bridges they sponsored are immediately invalidated.

Can a child game the system by introducing everyone as "nice"?

The bridge formula includes the sponsor's coherence (C_S) and the introduction quality (q_event). A child who introduces everyone produces many bridges, but each bridge is bounded by b_max (0.30) and each bridge decays independently. The child's own coherence is not affected by sponsoring bridges -- sponsorship does not cost trust. However, if the child introduces an entity that subsequently causes high tension, the bridge decays rapidly. Over time, the accumulation pattern shows which introductions led to successful trust-building and which did not. The system does not penalise the sponsor, but the bridge mechanism self-corrects for poor introductions through accelerated decay.

How does the bridge interact with the Sinkhorn-Knopp mixing step?

The bridge is evaluated AFTER the Sinkhorn-Knopp projection. The mixing step redistributes trust across contexts according to the doubly stochastic constraint. The bridge is then applied to the post-mixing effective coherence. This means the bridge does not participate in trust conservation -- it is a transient addendum outside the mixing matrix. This is intentional: the bridge is not trust. It is a scaffold. Including it in the mixing matrix would allow it to be redistributed to other contexts, violating its purpose.

Is there a maximum number of active bridges?

In the reference implementation, the bridge registry is bounded at 16 active bridges per robot. This is a practical limit, not a theoretical one. Each bridge requires storage for the sponsor identifier, target identifier, creation timestamp, initial mass, current mass, and decay rate. At 16 bridges, the memory footprint is approximately 512 bytes. In practice, most deployments have 1-3 active bridges at any time.

Patent pending. US Provisional 64/039,655.

-- Colm Byrne, Founder -- Flout Labs, Galway, Ireland