Listening, not obedient
Flagship architecture
The Grievance Watchdog Architecture
A governance pattern for systems that need to hear harm signals, preserve field evidence, and keep vulnerable people inside the decision architecture before human consequence becomes administrative noise. The complaint must keep its shape long enough to matter.
What changes
From passive intake to accountable listening.
The point is not prettier reporting. It is continuity of attention: credible harm signals become structured records, confidential routes, and accountable escalation before the institution teaches itself not to hear them.
Plain-language summary
The Watchdog is the governance architecture behind GrieVoice and related tools. It designs listening, evidence, and escalation so that a raw harm signal does not get smoothed into a category and forgotten before anyone acts on it.
Refusal-stack counterweight
A safety story still has to measure what it removes.
Guardrails can block instructional harm. They can also remove support, context, and continuity, then call the disappearance safety. This page asks the harder measurement question: who is protected, who is abandoned, and how would anyone know?
Evidence keeps its shape
Narrative becomes a structured, traceable record.
Consequence stays visible
Confidentiality, routing, and escalation become accountable architecture.
When to bring in Watchdog architecture
Use it before the audit, not after the complaint that should have happened.
- Suspiciously quiet channel. Grievance volume is low and the team suspects the silence is about access, fear, or distrust rather than absence of harm.
- AI-mediated workflow is sanding the signal. Automated summaries, scoring, or routing are making harm signals harder to locate in their original form after categorisation.
- Audit or board needs evidence of independent listening. The institution has to show that grievance signals were heard and preserved over time, not just that a process existed.
- New vendor or product about to deploy. Listening architecture installed before launch is much cheaper than retrofitting it after the first complaint surfaces in the wrong place.
The canon
Vocabulary for systems that need to keep noticing.
The canon gives practitioners language for what is structurally wrong: not naive resistance to technology, but a precise way to see when systems lose context, smooth away consequence, or route accountability away from power. It names the moment when the map starts replacing the life underneath it.
liability / human-in-the-loop
The Liability Sponge
A human placed in the loop not for control, but to absorb blame when the system fails.
accountability / metrics
Accuracy Theater & The Accountability Gap
How '94% accurate' masks the fact that the failing 6% targets the most vulnerable.
refusal / sentinel
The Watchdog Paradox (Sentinel vs Sensor)
An obedient sensor transmits. A sentinel halts. High-stakes systems need the right to refuse continuation.
opacity / authority
The Clarke Constraint
When a system's reasoning cannot be interrogated, it acquires unquestionable authority.
procurement / contractual
The Calvin Convention
Six contractual mechanisms that encode pre-action constraints and human control as architecture, not policy.
What you typically leave with
Deliverable categories, not promised outputs.
- Architecture assessment. A written read on the current grievance system against the watchdog pattern, naming where the architecture is sound and where it leaks.
- Scoped pilot recommendation. A bounded next step, typically a GrieVoice pilot for one workflow, one or two languages, and one risk tier, with clear criteria for what counts as success.
- Confidentiality and escalation map. A diagram showing where original narratives are preserved, who can locate them, and how escalation reaches institutional authority without travelling through the supervisor of the reporter.
- Reviewer scripts. The questions auditors, board members, or programme owners can use to test whether the channel is actually being trusted by the people most likely to need it.
Products and instances
Where the watchdog architecture becomes usable.
public
GrieVoice
Multilingual grievance intake infrastructure for worker voice, community reporting, whistleblower channels, and scoped pilots.
public
The Experiment Nobody Authorized
Consolidated counter-narrative dashboard for AI companion guardrails, withdrawal risk, and the outcomes nobody is measuring.
member
Grievance Watchdog Blueprint
Architectural principles for accountable listening, escalation, and pause capacity.
public
AI vs IFC Analysis
The thesis piece: when AI outperforms the gold standard of ethics.