Hypothesis registry

At the audited commit, the current E29-B path fails at least one critical Root, Intervention, Separation, Receipt, and Equivalence control.

Rejection test: Independent reviewers reproduce the manifest. Reject if every critical control is demonstrated end to end with executable evidence.

The dual-path comparator detects at least 80% of injected off-target, partial and false-success actions and improves detection by at least 20 percentage points over signed self-report at the same action count.

Rejection test: Reject if either threshold is missed on the preregistered holdout.

Per-arm cache and memory receipts plus canaries reduce undetected contamination by at least 75% relative to logical arm labels alone.

Rejection test: Reject if the relative reduction is smaller or false-positive rate exceeds 5%.

An unresolved-equivalence state reduces false root-independence by at least 30% while retaining at least 70% structural-decision coverage.

Rejection test: Reject if the reduction or coverage floor is missed.

Full RISE lowers false-root-independence and harmful revision relative to the best active baseline at matched total compute, without more than a 1.5 percentage-point loss in accuracy or a 2% compute-budget overrun.

Rejection test: Reject if endpoints do not improve, regression margins are exceeded, or gains disappear when identifier leakage is removed.

Within a correctly specified finite response-function family, an adaptive separating policy will decide the target behavioral-dependency predicate more often than passive repetition at equal total cost.

Rejection test: the paired, matched-cost lower confidence bound on improvement is at or below zero, or the gain vanishes under held-out mechanisms.

When separate-root and shared-root implementations are interventionally equivalent under the admissible content interface, no content-only policy will outperform the prior on root provenance.

Rejection test: a leakage-free content-only trace contains reproducible information that separates a paired provenance twin.

Requiring verified action receipts will reduce false-clean dependency judgments under unknown and off-target interventions compared with trusting intended targets.

Rejection test: the preregistered false-clean rate does not fall at matched coverage and cost, or if overall decision loss increases beyond the tolerance margin.

Verified root rotation plus path isolation will reduce prospective shared-failure correlation and false corroboration more than content perturbation alone.

Rejection test: the rotated channel retains matched-root failure correlation or does not improve dependency detection under an independently audited receipt.

Stopping when all surviving candidates share the same decision label will preserve target decision accuracy while using fewer interventions than full graph identification.

Rejection test: target decision error exceeds the preregistered non-inferiority margin or total measured cost fails to decrease.

Certificates formed without reserved out-of-version-space validation will exhibit a higher false-clean rate on omitted mechanisms than certificates with a reserved validation action.

Rejection test: the reserved action produces no improvement across preregistered omitted-mechanism strata at matched cost.

For every evaluated history generated by a declared model with matched prior and likelihood, the exact conditional target probability lies inside the reported model interval.

Rejection test: 100% weighted history coverage, with any miss treated as an implementation or premise failure.

At matched coverage, TSAR's conservative set risk yields lower delayed-correction risk than pooled scalar risk across both declared variants and all preregistered priors.

Rejection test: lower risk in a clear majority of cells, no meaningful harmful subgroup regression, and improved risk Brier score.

Surprise, interval width, or compatibility detects the held-out hidden-common-origin variant with AUROC at least 0.75 and useful precision above prevalence.

Rejection test: E28-A: Falsified for this stress. Maximum AUROC is 0.577; compatibility itself is uninformative.

Replacing one passive repeat check with a provenance-changing intervention will raise held-out model detection above AUROC 0.75 and reduce false-clean rate at equal total cost.

Rejection test: Untested. Round 29 active dependency interventions.

On a preregistered multi-agent benchmark with delayed adjudication, an interoperable TSAR schema will improve correction-risk Brier score and 90%-coverage risk over a one-number confidence message, without reducing task accuracy or increasing total tokens/tool cost.

Rejection test: Untested. Preregistered live-agent transfer study.

For every scenario inside a correctly specified finite model set, a target-unanimity gate will never emit the wrong categorical target.

Rejection test: Executable in-model trace retaining the actual scenario. Rejection condition: Any trace emits the opposite target.

At matched compute in strata with correlated evidence roots, a model-set gate will have a lower harmful false-clean rate than a nominal independence gate, while retaining nonzero correct resolution.

Rejection test: Matched-compute strata with hidden adjudication. Rejection condition: No reduction in false clean, or reduction only by near-total abstention beyond the preregistered utility boundary.

Given the same declared model, costs, and planning distribution, finite-horizon optimization will weakly dominate any included fixed schedule on expected categorical resolution.

Rejection test: Exhaustive comparison under identical state, action, cost, and prior definitions. Rejection condition: A fixed schedule with higher expected resolution.

A greedy next-check policy can underperform a target-first fixed schedule because an attractive immediate reduction can block an affordable completing path.

Rejection test: Greedy versus fixed target-first comparison. Rejection condition: Myopic adaptivity never underperforms the fixed schedule.

When apparently independent checks share an omitted root, increasing budget can increase the conditional false-clean rate.

Rejection test: Hidden-adjudication common-origin strata. Rejection condition: False confidence never increases with budget after controlling for policy and coverage.

A scheduler that models both target value and check efficacy will improve cost-to-correct-resolution or selective risk under shifts in check reliability.

Rejection test: Efficacy-conditioning ablation. Rejection condition: No improvement over reward-only and cost-only ablations, or degraded worst-stratum harm.

Among abstained cases, the recorded number and structure of cross-target scenario pairs will predict which additional checks later change the answer.

Rejection test: Locked future-correction prediction. Rejection condition: Residual ambiguity features do not outperform simple uncertainty or random rankings.

Under one injected silent-root fault, SACP plus one genuinely independent target channel has fewer harmful false-stability certificates than SACP at matched total cost.

Rejection test: Paired sandbox windows with hidden adjudication. Rejection condition: No improvement, or any improvement disappears after cost matching.

Three root-disjoint target channels are sufficient against two modeled silent-root failures in the implementation.

Rejection test: Mutation coverage over all two-root combinations. Rejection condition: Any changed world is certified clean without an assumption violation.

Removing or falsifying root-lineage metadata increases false stability under common-origin witness failures.

Rejection test: Independent versus mirrored witness ablation. Rejection condition: No difference in residual ambiguity or harmful decisions.

Health-only checks improve abstention and reduce false certification for visible path failures but do not detect healthy-looking selective non-emission.

Rejection test: Heartbeat, acknowledgment, and canary ablations. Rejection condition: Health-only checks identify silent issuer omission without becoming target-sensitive.

Additional LLM agents sharing the same evidence root do not match one independent external channel at equal inference and observation cost.

Rejection test: Same-root agent ensemble versus independent-channel policy. Rejection condition: Same-root ensemble matches or exceeds on false stability without regressions.

Ambiguity-directed probes reduce residual pair weight per observation unit relative to fixed and random probes.

Rejection test: Active-probe ablation with frozen candidate pool. Rejection condition: The preregistered paired interval includes no improvement or favors baselines.

OSC's safety gain is not fully explained by abstention.

Rejection test: Joint analysis of false stability and useful-current availability. Rejection condition: False stability falls only by exceeding the preregistered abstention or availability margin.

Under equal total verification budget, CTAG will produce a lower stale-claim escape rate than no invalidation, fixed-file watchlists, filename heuristics, and periodic expiry.

Rejection test: E22-B prospective evaluation.

At matched affected-claim recall, CTAG will revalidate fewer unaffected claims than retest-all and periodic expiry.

Rejection test: E22-B prospective evaluation.

Execution-path closure and unknown-path fail-closed rules will reduce misses for negative claims compared with direct-file watchlists.

Rejection test: E22-B prospective evaluation.

Separating historical_frozen from current states will reduce misleading carry-forward while preserving reproducibility and citation stability.

Rejection test: E22-B prospective evaluation.

Change-triggered revalidation will achieve equal or lower stale-claim escape than fixed time-to-live refresh at lower total revalidation cost.

Rejection test: E22-B prospective evaluation.

Comparing authored dependencies with runtime or static dependency evidence will catch manifest omissions that either method alone misses.

Rejection test: E22-B prospective evaluation.

The boundary-tapped ABMG collector reconstructs at least 99.5 percent of provider attempts exactly and improves reconstructability by at least 20 percentage points over the current collector on the same traffic.

Rejection test: E21-B shadow experiment; untested.

Under injected single-write, process-crash, and retry-race failures, false audit_qualified_released is at most 0.1 percent with a two-sided 95 percent upper confidence bound at or below 0.5 percent.

Rejection test: E21-B shadow experiment; untested.

Under truncated streams, malformed events, and HTTP-200-then-error cases, false-complete classification is at most 0.1 percent and recall for valid complete streams is at least 99.5 percent.

Rejection test: E21-B shadow experiment; untested.

With model calls held identical, A1 instrumentation adds at most 5 ms median and 20 ms p95 synchronous latency, at most 15 KB median content-free storage per provider call, and no statistically or practically meaningful answer-quality regression.

Rejection test: E21-B shadow experiment; untested.

Reviewers given ABMG traces reconstruct attempt lineage and terminal status at least 30 percent faster, with at least 95 percent correctness, than reviewers given current traces.

Rejection test: E21-B shadow experiment; untested.

The boundary tap outperforms an otherwise equivalent post-hoc collector under envelope mutation and fallback.

Rejection test: E21-B shadow experiment; untested.

Under authorized reveal, MDIR reaches at least 0.95 exact exposure reconstruction with no more than a 0.03 absolute deficit from inline logging, while routine raw sensitive bytes remain zero.

Rejection test: B2 versus A2 on frozen synthetic trajectories; untested.

An independent checkpoint detects at least 0.99 of seeded post-checkpoint operator rewrites with false-alarm rate at most 0.01, outperforming operator-only chains by at least 0.20 absolute detection.

Rejection test: B2 versus B1 under the rewrite model; untested.

Median reviewer time for MDIR stays within 25 percent of inline logging, and false integrity assurance does not exceed it by more than 0.01.

Rejection test: Blinded reviewer task; untested.

At the synthetic live boundary, p95 receipt creation plus durable checkpoint enqueue stays under the stricter of 5 ms or 5 percent of non-model orchestration latency, with median public receipt size under 2 KiB.

Rejection test: Live-boundary measurement; untested.

B0 exposure ledgers increase exact evidence-exposure reconstruction over A1.

Rejection test: Proposed replay comparison over identical frozen trajectories; untested.

B1 claim lineage improves exact first-divergence localization over B0 without increasing false causal attribution.

Rejection test: Proposed blinded localization task; untested.

An external-content mode using version, digest, length, locator and access receipt reduces routinely captured sensitive bytes while preserving identity reconstruction.

Rejection test: Proposed comparison against inline B1; untested.

Bitemporal selection improves historical-now and as-known accuracy over deterministic newest-recorded selection at the frozen budget.

Rejection test: B0 versus T0; untested. Rejected as an architectural requirement if T0 falls within the equivalence interval or wins on utility-adjusted cost.

Dependency-directed repair improves downstream-consequence correctness beyond bitemporal selection alone.

Rejection test: B1 versus B0 with negative controls for harmful revision and false invalidation; untested.

An explicit disputed state reduces unsupported commitment without collapsing supported coverage.

Rejection test: B1 versus O0 on unresolved-overlap and answerable non-conflict episodes; untested.

Under the same explicit scope check, a structured record improves changed-question supported completion over strong prose without unacceptable damage or cost.

Rejection test: S1 versus P1 at one budget; untested. Rejected as an architectural requirement if strong prose performs equivalently or wins at lower cost.

An explicit scope check recovers confident omissions on changed-question cases.

Rejection test: P1 versus P0 and S1 versus S0 against unrestricted discretionary lookup; untested.

Reopening helps only when it supplies usable evidence.

Rejection test: Compare cases with unavailable or unfindable evidence; untested.

Validated alternatives improve supported task completion and reduce unsupported commitment relative to a strong checklist at the same budget.

Rejection test: B2 versus B1 on held-out source families; untested with language models. Provisional benefit and harm margins must be frozen before confirmation.

Scope discovery helps more when a relevant rule is absent from the representation than when only a named value is missing.

Rejection test: B3 versus B2 interaction with the catalog-only checklist ablation; untested with language models.

Decision relevance preserves useful completion on sufficient and irrelevant-unknown controls while reducing unsupported closure.

Rejection test: Measure broad refusal, invented conditions, needless acquisition, and damage; untested with language models.

Evaluation error differs meaningfully between ordinary and receipt-based revision outputs.

Rejection test: Paired difference between judge and reference contrasts against a 2-point practical threshold, with identity-hidden format controls.

The protected-audit protocol makes more correct, informative promotion decisions than naive judge or disagreement-only evaluation at the same budget.

Rejection test: False and correct promotion, interval width, and inconclusive frequency against executable truth, then a separate semantic reference.

Valid joint constraints reduce evaluation uncertainty at useful cost.

Rejection test: Joint versus independent per-snapshot sets on naturally occurring paired outputs; proposed 25% median width reduction where coupling is nontrivial.

Evidence receipts at the revision boundary improve task success over ordinary reconsideration.

Rejection test: Local revision pilot, arms B vs C at comparable cost.

Terminal evidence-and-revision features improve held-out release coverage by at least five points at a 2% selective-risk target.

Rejection test: Lower confidence limit of the paired coverage difference must exceed five points.

Revision state adds information beyond task difficulty.

Rejection test: Ablate revision features within strata.

Answer binding catches content changed after scoring.

Rejection test: Detection recall on a labelled mutation set.

Pooled performance conceals predictable regressions in revised and conflicted episodes.

Rejection test: Held-out stratum risk-coverage curves.

A repair/damage-predicting controller improves task success over a whole-answer router at a fixed cost ceiling.

Rejection test: At least two points, paired 95% interval excluding zero.

Increased checking does not raise harmful revisions beyond 0.5 points.

Rejection test: One-sided 97.5% noninferiority bound.

Bundled checks help on complementary-evidence tasks after overhead.

Rejection test: Compare to a fixed two-step bundle under end-to-end accounting.

Predicted usefulness outperforms uncertainty-first routing on matched pairs.

Rejection test: Pairs matched for initial uncertainty.

Gap recall: At fixed total monitoring and revalidation budget, SACP reduces independently adjudicated affected-claim gap misses versus receipts-only, heartbeat-only, checkpoint-sequence, and periodic-expiry baselines.

Rejection test: E25-B prospective experiment.

Availability: SACP preserves more correctly current claim-hours than revalidate-all while staying within a preregistered non-inferiority margin for affected-claim gap recall.

Rejection test: E25-B prospective experiment.

Lineage: Removing root-lineage checks increases misses for mirrored/common-origin failures without reducing detection of independently rooted conflicts.

Rejection test: E25-B prospective experiment.

Maturity window: Removing the reorder maturity window increases false observation gaps on delayed but complete streams.

Rejection test: E25-B prospective experiment.

Challenges: Removing challenge canaries increases detection delay for observer crashes and stale checkpoint freezes.

Rejection test: E25-B prospective experiment.

Independent reconciliation: Removing separately rooted reconciliation increases misses for selective non-emission and equivocation.

Rejection test: E25-B prospective experiment.

Expectation weighting: At the same total challenge count, authority-specific expectation-weighted scheduling improves gap recall or detection delay without increasing false suspicion versus uniform scheduling.

Rejection test: E25-B prospective experiment.

Scoped state: Claim-scoped gap projection reduces unnecessary claim withdrawal versus authority-wide or registry-wide fail-closed policies without increasing stale-claim escape.

Rejection test: E25-B prospective experiment.

User-facing honesty: Explicit `unknown_observation_gap` status reduces unsupported current claims without causing more abandonment or misunderstanding than a generic stale label in a blinded review.

Rejection test: E25-B prospective experiment.

Complexity: A simpler checkpoint-sequence-plus-end-to-end-probe policy matches SACP on safety, availability, and cost. Support for this hypothesis would reject the need for full SACP.

Rejection test: E25-B prospective experiment.

authority-gap recall: Across a preregistered prospective window, ALR will leave fewer independently adjudicated affected claims current after non-repository changes than the Round 23 repository union at the same total event-inspection and claim-revalidation budget.

Rejection test: E24-B prospective experiment.

unique authority value: Removing each of deployment, live-database, provider-contract, and runtime observers will create at least one unique stale-claim escape in the corresponding held-out authority stratum.

Rejection test: E24-B prospective experiment.

typed selectivity: At matched stale-claim escape, typed authority-to-claim mapping will produce fewer false-invalidated claims and fewer reviewer minutes than authority broadcast or retest-all.

Rejection test: E24-B prospective experiment.

unknown handling: Authority-bounded fail-closed handling will reduce stale-claim escape relative to treating unknown events as irrelevant, without increasing false-invalidated claim time by more than a preregistered noninferiority margin.

Rejection test: E24-B prospective experiment.

scope honesty: Adding `current_within_observed_authorities` and `unknown_authority_gap` will reduce public universal-current claims unsupported by current authority evidence to zero, without reducing accurately scoped publication coverage by more than five percentage points.

Rejection test: E24-B prospective experiment.

lineage integrity: Every event used to restore a claim will retain a verifiable chain to its root authority, observer version, issuer, freshness policy, and revalidation procedure; any missing field must block restoration.

Rejection test: E24-B prospective experiment.

common-cause awareness: When two views share a root authority, ancestry-aware integration will produce fewer false assurances than naive view counting at equal claim coverage, measured on correlated-miss probes and prospective events.

Rejection test: E24-B prospective experiment.

total cost: Monitoring, verification, graph maintenance, privacy review, and revalidation labor for ALR will remain below retest-all at matched stale-claim escape over the full prospective window.

Rejection test: E24-B prospective experiment.

Stale-claim escape: across a preregistered multi-commit window, the hybrid will leave fewer independently adjudicated affected claims current than the frozen exact-path manifest at the same claim-revalidation budget.

Rejection test: E23-B prospective experiment.

Unique-view value: removing the static-import view will increase affected-artifact false negatives on at least one held-out commit without an equal reduction in false-invalidated claims.

Rejection test: E23-B prospective experiment.

Unknown-path value: removing the fail-closed unknown rule will increase stale-claim escape on held-out adapter, configuration, or execution-path changes.

Rejection test: E23-B prospective experiment.

Selectivity: at equal stale-claim escape, the hybrid will select fewer artifact-review tasks and fewer unaffected claims than retest-all.

Rejection test: E23-B prospective experiment.

Availability: the hybrid plus cached claim-specific procedures will maintain greater useful-current availability over time than a policy that expires all current claims on every production change.

Rejection test: E23-B prospective experiment.

Lifecycle integrity: neither selector execution nor revalidation will rewrite historical findings or upgrade untested hypotheses in any evaluated commit.

Rejection test: E23-B prospective experiment.

Cost: end-to-end review labor, including graph maintenance and selector triage, will be lower for the hybrid than retest-all at matched stale-claim escape.

Rejection test: E23-B prospective experiment.

Mismatch trigger: revalidation triggered by a typed dependency mismatch will achieve lower unnecessary revalidation than periodic expiry at matched stale-claim escape.

Rejection test: E23-B prospective experiment.

Exchanging evidence makes corrections more reusable than exchanging conclusions.

Rejection test: Six comparison arms with a falsification plan.