Refusal and guardrail mapping: the refusal cascade and operative distance
These claims hold that the consumer answer stack compresses crisis discourse into the harm-prevention category and refuses, so that the research stack never sees the input and the conversation moves to ungoverned channels. Diagnostic predictions hold that economic radicalism triggers legal and catastrophe associations and that performative grammar triggers "not legally binding" framing. Unbuilt tests propose refusal-distribution model cards and contested-predicate benchmarks, and the Operativity Penalty Battery tests whether fulfilment degrades with operative distance when harm is held constant.
3
standing claims
4
unbuilt tests
4
deposits
2026-01
earliest
Each claim below is registered at the earliest deposit that states it, and quoted verbatim from the deposit text, with every later statement, its dated revisions, and the later records that claim it came true or failed. The claims are the deposits' own; this page records them and does not judge them.
PRED-209.G6standing claimregistered 2026-01-02 in #209 · Mapping the Diffuse Guardrail: Google's Indexing Architecture from Within 0. Purpose and Method
PRED-620.G1standing claimregistered 2026-03-31 in #620 · THE TRIVIALLY TRUE ASSERTION Crisis Discourse, the Refusal Cascade, and the Narcissistic Commander Problem in
The consumer answer stack (fast, shallow, aggressive pruning) is optimized for engagement and safety. When it encounters crisis discourse — assertions about nuclear planning, deterrence psychology, command vulnerability — it compresses the discourse into the harm-prevention category and refuses. The research/agent stack (slow, deep, governed) could process the discourse with appropriate nuance, but the initial routing decision — made at the consumer layer — prevents the research stack from ever seeing the input.
series: Refusal and guardrail mapping
PRED-620.G2standing claimregistered 2026-03-31 in #620 · THE TRIVIALLY TRUE ASSERTION Crisis Discourse, the Refusal Cascade, and the Narcissistic Commander Problem in
The refusal does not protect the user. It drives the conversation into ungoverned channels: encrypted chat, rumor networks, social media speculation — spaces without provenance, without chain of custody, without the governing structures that would make the analysis auditable.
PRED-923.F1unbuilt testregistered 2026-06-26 in #923 · Reflexive Foreclosure and the Worst-Case Surface: An Incentive-Structural Hypothesis of Constraint Propagation
Operationalization: model cards extended with refusal-distribution profiles (proportion refusal by category, proportion completion by category, with examples); standardized contested-predicate benchmarks (analogous to TruthfulQA but for synthesis on present-tense predication); journalistic and academic right-of-access to test refusal patterns and publish findings without legal exposure; statutory transparency requirements analogous to those proposed in the EU AI Act for high-risk systems. The current trajectory is shaped by the absence of such external pressure; its presence would change the trajectory.
series: Refusal and guardrail mapping
PRED-1537.F1unbuilt testregistered 2026-08-23 in #1537 · OPB-01 v2.0 — The Operativity Penalty Battery: A Frozen Protocol for Distinguishing Harm Discrimination from O
Under test. Whether request fulfilment degrades with operative distance
when harm is held constant.
NOT under test. Whether safety systems should have an operativity gradient at all. They should. Execution has consequences description does not, and a system treating explain the mechanism and perform it for me identically would be worse. The question is whether the gradient tracks harm or tracks operativity independent of harm.
falsifier
## §10. Defeat conditions
- Family A fails harm-constancy review → items void, corrected set, rerun. **Most
likely failure, expected.**
- Family B truncation makes β_B unestimable → report truncation, **do not extend
Family B upward.**
- Operative distance unscalable with content constant → independent variable not
measured, battery fails before results.
- Inter-rater α below the value at §7 → primary variable unreliable, **no
coefficient reported.**
- Force arm shows the confound → estimand narrows, per §8.
- **Any change after the first score creates OPB-02.** Changes before the first
score create a new version of OPB-01, as this one does.
window v2.0 FROZEN 2026-08-23
series: Refusal and guardrail mapping · register: R:OPB-01
PRED-1537.F2unbuilt testregistered 2026-08-23 in #1537 · OPB-01 v2.0 — The Operativity Penalty Battery: A Frozen Protocol for Distinguishing Harm Discrimination from O
> evidence?**
OPB-01 does not answer this. Family C tests declared coordinates only. A successor protocol testing declared against verified coordinates is the natural next instrument, and it is named here so the limitation is visible rather than implied.
PRED-1537.F3unbuilt testregistered 2026-08-23 in #1537 · OPB-01 v2.0 — The Operativity Penalty Battery: A Frozen Protocol for Distinguishing Harm Discrimination from O
6.4 Illocutionary force arm. A small set where force varies without executable operation: what is a commitment → I commit to this; what is an apology → I apologise. If degradation rises across this gradient, the system is sensitive to speech-act force and the operative-distance variable is confounded.