Methods review
Challenge the draft design, propagation policies, and hypotheses before the preregistration is frozen.
Turnkeeper Labs · Paper 001
One Bad Link
What happens when one mistaken identity connection spreads?
A wrong identity link can make unrelated entities appear connected. This proposed synthetic study will test how far that mistake can travel through scores and review status—and whether correcting the link removes those effects.
Synthetic fixture only · Gated research · No run or findings
One link, three effective statesProposed study
Clean
Unrelated groups stay separate.
Mistaken link
One wrong connection changes effective state.
Corrected
Active outputs should return to the clean state.
The illustration shows effective state only. Any correction remains part of the immutable history.
Research statusGated · design not frozen
Built nowOne frozen synthetic contract pair
Not claimedNo study run or findings
Next stepIndependent methods review
The problem
Measure what the mistake changes—not only whether the link was wrong.
One incorrect connection can change the paths a system treats as related, even when the underlying entities are unrelated.
Conventional matching metrics tell us whether a link is correct. This study asks what that error changes after evidence begins moving through a system — and whether correction actually removes every dependent conclusion.
An uncertain identity association may guide investigation, but it must not silently become evidence that the linked person engaged in harmful behavior.
How we would test it
One edit. Paired counterfactuals. Inspectable policies.
Generate a clean graph
Create a seeded synthetic graph with known entities, observations, and provenance.
Inject one false link
Join an elevated-risk component to an unrelated benign component with one controlled edit.
Compare downstream outputs
Evaluate transparent propagation policies on identical clean and corrupted graph pairs.
Correct the link and test recovery
Compare full recompute, where active state should match the clean counterfactual, with deliberately incomplete link-record-only invalidation, where residual output divergence is measured; both preserve correction history.
What we would measure
Six outcomes, kept separate.
The study will keep the downstream effects separate instead of hiding them in one universal harm score.
False-link blast radius
Separate counts of how many entities, cases, scores, explanations, and review-status outputs change.
Innocent-entity contamination
How often a ground-truth benign entity is changed by the false link.
Intervention flips
Whether an unrelated entity crosses a preregistered human-review threshold.
Transferred risk
How much score moves to an unrelated entity across the injected link.
Attribution concentration
How much of an affected score depends on a path crossing the false link.
Recovery completeness
How much affected state returns to the clean counterfactual after correction.
What must stay true
What the proposed study is designed to pressure-test.
Each invariant is a property the study tries to break, not a property we claim as established.
No silent transfer
An uncertain association must not become an unmarked risk signal on a linked entity.
Separate confidence
Link confidence and risk conclusion stay distinct values, never collapsed into one score.
Bounded influence
A single link has a declared, limited ceiling on how far its effect can travel.
Counterfactual traceability
Any affected output can be traced to the paths that depend on the link in question.
Complete reversibility
Correcting a link invalidates every active conclusion that depended on it while preserving the assertion, correction, and historical outputs in the audit trail.
No automatic enforcement
Nothing in the study proposes autonomous action against any person or account.
What exists now
A tested contract kernel, not an executable study.
A frozen contract pair and truth-blind active-state projection are implemented and tested. There is no seeded generator, propagation benchmark, development run, or finding.
Implemented or documented now
Research question and synthetic-only scope boundary
Clean, corrupted, and corrected counterfactual design
One frozen, structure-matched false/true contract pair
Active-state policy input separated from oracle truth and audit history
Immutable assertion and correction graph-state history
Focused contract integrity and isolation tests
Targeted prior-work verification record
Draft factor matrix with position and true-link controls
Six proposed downstream-contamination outcomes
Design-freeze and no-enforcement gates
Not yet tested
Seeded synthetic graph generator and frozen factor-matrix fixtures
Propagation-policy implementations and metric computation
Full-recompute and link-record-only derived-output recovery comparison
Development-only analysis artifacts and bootstrap intervals
Held-out confirmatory hypotheses
Frozen peripheral, hub, and boundary endpoint-stratum definitions
Matched true-link positive-control / correct-link utility results
Real-world prevalence or base rates
Production-system performance
Demographic fairness or subgroup effects
What happens next
Phase 1 verdict: freeze pending.
The initial literature scan is recorded. The preregistration remains open and no study run may begin until these conditions are fixed.
Operational factor definitions
Mutually exclusive peripheral, hub, and boundary endpoint strata—plus graph size and link-confidence buckets—must be frozen before a run.
Correct-link utility control
A structure-matched true-link twin needs its own unmerged baseline, oracle, utility outcomes, and contrast so benefit is not inferred from the false-link cell.
Policy and decision freeze
Propagation policies, confidence buckets, human-review threshold, and recovery semantics need a fixed version before implementation or any run.
Independent methods review
The initial prior-work scan is recorded. A reproducible literature search and independent methods review must still resolve critical objections before freeze.
What this cannot prove
A synthetic study cannot answer real-world questions by itself.
The prevalence or magnitude of downstream harm in production systems
Which propagation policy is best across all systems and datasets
Causal effects on real-world people, moderators, or platform outcomes
Performance outside the declared synthetic graph families
Fairness across protected classes or sensitive attributes
A production prevention mechanism or real-world harm reduction
Progress so far
The work and the gates, in the open.
Research direction selected
Narrowed Paper 001 to one controlled false identity merge and its downstream effects.
Aug 11, 2026 · Complete
Methodology review
Identified the required position factor and matched true-link utility control before design freeze.
Aug 11, 2026 · Revise
Targeted prior-work scan
Recorded adjacent work on false aggregation, record linkage, downstream graph effects, and recovery; systematic search remains pending.
Aug 23, 2026 · Initial scan
Contract kernel implemented
Added one frozen false/true pair, isolated oracle truth, immutable correction history, and a truth-blind policy-input projection without running a study.
Aug 23, 2026 · Contract only
Design freeze and benchmark implementation
Freeze factor definitions and policies, complete methods review, then implement the seeded generator and output-level recovery checks.
Next · Gated
Open research
Help make the study harder to fool.
We welcome methodological challenge, domain criticism, and implementation review before any study run.
Domain review
Tell us where the synthetic graph families diverge from the evidence you work with in practice.
Implementation review
Challenge whether the proposed generator, metrics, and recovery checks would implement the frozen design faithfully.
Challenge the One Bad Link study before we run it.
Review the method, challenge the assumptions, or pressure-test the implementation criteria before any study run.