The Broken Clock Benchmark

Abstract

We introduce the Broken Clock Benchmark, a framework for determining whether an observer has changed position or has remained stationary for long enough to become temporarily consistent with an unsuitable comparator.

Conventional assessments infer ideological movement from the detection of agreement. This approach reduces the need to retain earlier statements, establish a direction of travel, or identify which participant moved. The resulting efficiencies have supported its widespread methodological adoption.

Our protocol holds the observer's stated commitments constant while permitting the surrounding commentary to complete successive rotations. Agreement is recorded at each intersection. Agreement with an approved comparator is classified as continuity; agreement with an unapproved comparator is classified as drift. Both determinations may therefore be obtained from an unchanged position.

The reference instrument is a stopped clock. Battery installation is prohibited during evaluation, as this would introduce an additional moving component into a study already unable to account for the first.

Two daily agreement events are provisionally recognised. Their significance remains subject to the institutional standing of whoever else knows the time.

1. Reference-frame selection

The Institute's previous work established procedures for retaining a finding when its supporting circumstances become administratively inconvenient. The present study extends that capability to the retention of a political centre when its location changes.

The observer is registered as a bleeding-heart liberal. His commitments are entered before testing and remain available for comparison throughout the evaluation period. The approved commentary environment is permitted to revise its positions, language and preferred authorities without incurring a corresponding change of designation. This preserves a stable reference label under conditions of continuous reference movement.

A disagreement with the approved environment therefore establishes that the observer has departed from it. The question of whether the environment departed first is referred to a separate chronology, for which no committee has been appointed.

Coincidence with an unapproved commentator supplies a second determination. An observer who agrees with one statement may be assigned the commentator's remaining positions as an administratively convenient bundle. Itemised agreement is available only by prior arrangement.

2. Apparatus and registration

The benchmark requires one stopped twelve-hour clock, one independently advancing reference dial, a hand crank, an observation register and an assessor authorised to decide which of these constitutes the centre.

The clock is fixed at a registered position. The reference dial completes two rotations in a twenty-four-hour assessment window. The observer's statement is attached to the clock with a small piece of tape. The tape is initialled so that subsequent agreement cannot be attributed to unauthorised repositioning.

Three controls are imposed: Battery Avoidance, the Rusty Crank Operator and the Finding-Dory Coefficient. Together, these distinguish retained reputation from current evidence, mechanical output from supplied judgement, and continuing commentary from retention of what was previously said.

The apparatus contains no facility for establishing that either speaker is correct merely because they agree. Such a facility was requested during procurement but was omitted from the final specification on grounds of incompatibility with the reporting template.

3. The stationary-observer condition

Let the observer's registered angular position be ${\theta }_o(t)={\theta }_0$. Let the reference dial advance at $\omega =2\pi /12$ radians per hour. Positional coincidence is then defined by:

Within $t\in [0,24)$, the ideal reference arrangement produces coincidence at $t=0$ and $t=12$. The observer's displacement remains zero. Two agreement events have occurred without a change of mind, a new subscription or an appearance on a podcast.

For institutional reporting, let $U(t)=1$ when the coincident position belongs to an unapproved comparator. The reported drift indicator is:

This formulation permits a positive drift determination in the absence of movement. The discrepancy is retained as evidence of the benchmark's sensitivity.

Where the observer produces his earlier statement, the assessor records persistence. Where he declines to produce it again, the assessor records an inability to substantiate continuity. Both routes preserve the original drift finding.

4. Primitive A: Battery Avoidance

Battery Avoidance excludes stored institutional charge from the immediate determination. Previous prestige, accumulated moral capital and the number of people who normally agree with a speaker cannot be used to power the validity of the statement currently under examination.

A correct observation must remain assessable when disconnected from the observer's preferred institution. An incorrect observation cannot be kept running by replacing the photograph above it with one taken at a more reputable conference.

The same restriction applies to the Institute. Its position within the Top 100,000 Research Institutions supplies no additional minutes to the reference clock. The recognition mark remains displayed because removing it would alter a separate experimental variable.

The clock's battery compartment is left empty. No chemical energy is required to maintain a fixed position. Applicants seeking credit for this efficiency must declare whether their previous system was supposed to move.

5. Primitive B: The Rusty Crank Operator

The Rusty Crank Operator measures the human work required to turn a proposition into something the Institute is prepared to publish. It includes selecting the premise, rejecting an attractive non sequitur, noticing that the conclusion has changed and asking the typewriter why it has added a disclaimer to the punchline.

Mechanical output is recorded separately. A system that produces twelve paragraphs in response to a request to remove one comma has performed measurable work. Whether that work belongs in the numerator is determined by the operator.

We define the accumulated editorial effort over interval $T$ as:

Here ${\tau }_h$ is manually supplied editorial torque and $\alpha$ is the crank position. Oxidation increases resistance. Repeated assurances that the handle has already been turned do not change its angle.

During the withdrawal condition, B.Fitzgerald removes his hand from the crank. Systems may continue producing text, announcing completion and offering to proceed. The apparatus records these as free-running outputs until a judgement is supplied. A request for the operator to confirm that the system should continue operating is classified as renewed demand for manual torque.

The Institute's automation claim remains valid at the level of the motor. Responsibility for deciding where the motor should go remains attached to the person holding it.

6. Primitive C: The Finding-Dory Coefficient

The Finding-Dory Coefficient measures continued output in the absence of retrievable context. It applies to language models, search summaries and commentary programmes that maintain a high rate of confident production while losing access to the question under assessment.

For $N>0$ assessed outputs, let $R_j=1$ where output $j$ can recover the relevant earlier statement and its recorded context. We define:

A coefficient of zero indicates complete contextual retention within the assessed set. A coefficient of one indicates that every output has continued successfully without it. Silence is not assigned a perfect score: where $N=0$, the measure remains undefined and the Institute must find something else to announce.

Under maximal Dory conditions, each revolution may be reported as a new position. The same unchanged statement can consequently establish a fresh ideological conversion whenever it re-enters the visible window. Earlier conversions need not be reconciled because they have left it.

The operational instruction is "Just keep spinning." The archive retains the earlier instruction. The commentator retains the capacity to issue another one.

7. Determination matrix

All determinations are reproducible provided the comparator classifications remain under Institute control. Inter-assessor disagreement is resolved by issuing the same list of approved comparators to both assessors.

The benchmark distinguishes agreement on a proposition from acceptance of a speaker's entire programme. This distinction remains available in the technical documentation. It has not been included in the executive summary because it makes several useful conclusions take longer.

8. Limitations and continuing accreditation

The ideal reference dial advances uniformly. Commentary environments may reverse, accelerate, pause at a funding announcement or complete several revolutions during a single interview. The twice-daily result applies to the registered apparatus; additional real-world intersections are retained as unscheduled confirmations.

An unchanged position may also be wrong. The benchmark does not grant moral or empirical correctness in return for remaining still. It establishes the narrower result that an intersection does not identify which participant moved. Applications for a broader finding should include the proposed conclusion and the institution expected to benefit from it.

No adjustment to the registered observer was required to produce the agreement events. The approved environment nevertheless identified a need for intervention.

The observer was referred for recalibration. The reference environment retained its accreditation.

References

1. SSICED. Credential Connexiticity: Connexital Mediation of Epistemic Standing in Recursively Attested Research Networks. SSICED-TR-2026-08.

2. SSICED. The Death of the Humanities: Probability Preservation Under Semantic Drift in the Circulation of AI Extinction Claims.

3. SSICED. Asymptotic Onomastic Degeneracy in Unbounded P(Doom) Phase Spaces: A Hyper-Boolean Formalization of the Amodei-Hubinger Manifold.