{ "schemaVersion": "1.0", "updated": "2026-08-10", "status": "live-defeasible-ledger", "focalExplanation": "Evidence Press demonstrates automation of open packaging and producer-side replay. The live working hypothesis is that these mechanisms may accelerate some discovery in bounded, certificate-friendly domains, while observed throughput remains compatible with task selection, intensive orchestration, an earlier assurance endpoint, backlog conversion and research-lineage effects.", "claimCeiling": "The current record establishes publication capability and producer-side replay, not a comparative acceleration effect. It does not show faster scientific discovery or publication at matched assurance and total cost, better or faster policy decisions, greater reuse, operational savings or organisational productivity.", "statusVocabulary": [ { "id": "untested", "definition": "The comparative or outcome claim has not been evaluated with evidence capable of distinguishing it from its registered rivals.", "entryCriteria": [ "The hypothesis and scope are registered.", "No qualifying comparative, field or causal evaluation has yet been recorded." ], "claimCeiling": "The mechanism may be plausible or partly demonstrated, but no speed, quality, reuse, policy, operational or productivity effect may be inferred." }, { "id": "mechanism-supported", "definition": "Evidence shows that a specified mechanism or prerequisite operates within a bounded setting, but the predicted comparative outcome remains unmeasured.", "entryCriteria": [ "At least one referenced observation directly bears on the stated mechanism.", "The observation's inference limit excludes the comparative outcome claim." ], "claimCeiling": "Only the bounded mechanism is supported; downstream acceleration or impact remains untested." }, { "id": "single-case-supported", "definition": "At least one qualifying case measures the predicted outcome in a declared setting, without establishing general transfer.", "entryCriteria": [ "A referenced observation measures the outcome rather than merely the mechanism.", "The case, comparator, measurement boundary and important limitations are recorded." ], "claimCeiling": "Support is limited to the observed case and design; generalisation and causal attribution require additional evidence." }, { "id": "benchmark-no-clear-gain", "definition": "A bounded benchmark found no decision-relevant net gain under its registered tasks, models and measures.", "entryCriteria": [ "The benchmark, comparator and decision rule are preserved.", "The result does not meet the registered gain criterion." ], "claimCeiling": "This is a scoped negative or inconclusive result, not evidence of universal ineffectiveness." }, { "id": "field-association", "definition": "A measured field association is present in a declared context, but the design does not identify a causal effect.", "entryCriteria": [ "The field outcome, population, period, missingness and comparator are recorded.", "The design's confounding and selection limits are explicit." ], "claimCeiling": "The association is context-bound and must not be described as causal." }, { "id": "causal-effect-supported", "definition": "A preregistered randomized or defensible quasi-experimental design supports a context-bound causal effect under its stated assumptions.", "entryCriteria": [ "The estimand, assignment or identification design, comparator, uncertainty and attrition are recorded.", "The analysis meets its preregistered decision rule and has independent methodological review." ], "claimCeiling": "The effect is supported only for the registered population, intervention, comparator, outcomes, period and assumptions." }, { "id": "falsified-within-scope", "definition": "A preregistered falsifier or negative decision rule has been met within the declared scope.", "entryCriteria": [ "The triggering observation and applicable prediction or falsifier are referenced.", "The scope and conditions under which the hypothesis may be reopened are recorded." ], "claimCeiling": "Only the scoped hypothesis is rejected; adjacent mechanisms or settings remain separate questions." } ], "observations": [ { "id": "catalogue-baseline-throughput", "observedAt": "2026-08-10", "statusCapabilities": [ "mechanism" ], "source": "data/OPERATING_MODEL.json#releasePolicy.baselineCommit", "statement": "The adoption baseline contains 21 releases dated across 13 days, all openly represented with DOI, PDF and repository metadata.", "inferenceLimit": "Publication dates do not reveal research start, claim-lock time, total labour, attempted-project denominator or matched conventional throughput." }, { "id": "catalogue-assurance-boundary", "observedAt": "2026-08-10", "statusCapabilities": [ "mechanism" ], "source": "/api/papers.json", "statement": "All 21 baseline releases report producer-side internal replay; none reports editorial peer review, independent reproduction or end-to-end formal verification.", "inferenceLimit": "This demonstrates a transparent early assurance endpoint, not comparable-assurance research acceleration." }, { "id": "catalogue-task-mix", "observedAt": "2026-08-10", "statusCapabilities": [ "context-limitation" ], "source": "data/METHOD_REGISTRY.json", "statement": "The baseline catalogue is concentrated in exact mathematics, computational theory, certificate-friendly operations models and one substantive policy identification audit.", "inferenceLimit": "The portfolio does not establish transfer to wet-lab, clinical, field, stakeholder-dependent or normatively contested work." }, { "id": "method-lineages", "observedAt": "2026-08-10", "statusCapabilities": [ "mechanism" ], "source": "data/METHOD_REGISTRY.json#lineages", "statement": "Two bounded release programmes have explicit dependency evidence and reuse reductions, certificates or structural objects across adjacent questions; broader method clusters are not treated as lineages.", "inferenceLimit": "Lineage reuse is not independent confirmation and can propagate a common defect." }, { "id": "negative-results-preserved", "observedAt": "2026-08-10", "statusCapabilities": [ "mechanism" ], "source": "data/METHOD_REGISTRY.json#productive-failure", "statement": "The catalogue preserves non-identification, failed feasibility, unresolved parent targets, counterexamples and reusable modules from stalled work.", "inferenceLimit": "Downstream avoidance of duplicated work or earlier stopping has not been measured." }, { "id": "policy-identification-case", "observedAt": "2026-08-10", "statusCapabilities": [ "mechanism", "falsifier-test" ], "source": "/releases/stocks-are-not-flows/", "statement": "A registered housing analysis stopped unique causal attribution when the admitted observation map could not distinguish the focal mechanism from rivals and specified the missing measurement.", "inferenceLimit": "This is a design-conditioned non-identification result, not a policy-outcome study or proof of zero effect." }, { "id": "operations-theory-no-field", "observedAt": "2026-08-10", "statusCapabilities": [ "mechanism", "context-limitation" ], "source": "/releases/certified-two-item-jrp/", "statement": "Operations releases provide exact model-specific gaps, stability margins and certificates while explicitly leaving field evaluation open.", "inferenceLimit": "No realised operational saving, adoption or model-scale transfer has been established." }, { "id": "productivity-protocol-no-impact", "observedAt": "2026-08-10", "statusCapabilities": [ "benchmark-outcome" ], "source": "/productivity/", "statement": "Productivity Protocols preserves predecessor NO_CLEAR_GAIN results and reports NO_IMPACT_EVIDENCE for current packs.", "inferenceLimit": "Structural conformance and model-only benchmarks do not measure human or company productivity." } ], "hypotheses": [ { "id": "research-cycle-acceleration", "hypothesis": "Certificate-first publication, structural compression and retained stop receipts reduce elapsed time and total human effort per reusable research result at a matched assurance boundary.", "scope": "Bounded, digitally specified, certificate-friendly research tasks.", "aims": [ "science", "productivity" ], "epistemicStatus": "untested", "statusEvidence": { "status": "untested", "designClass": "none", "observationIds": [ "catalogue-baseline-throughput", "method-lineages", "negative-results-preserved", "catalogue-assurance-boundary", "catalogue-task-mix" ], "evidenceRefs": [], "comparator": null, "estimand": null, "decisionRule": null, "registeredDesignRef": null, "independentReview": null, "reason": "No prospective matched-assurance comparison records intake-to-result time, total active human effort, severe-error rates and attempted-project denominators.", "inferenceLimit": "Catalogue throughput and internal reuse make the mechanism plausible but do not establish a comparative research-cycle effect." }, "rivals": [ "certificate preparation offsets the verification saving", "task selection rather than the operating model explains the apparent gain", "short calendar time is purchased with greater human orchestration or compute", "faster output carries lower independent correctness or reuse" ], "supportingObservations": [ "catalogue-baseline-throughput", "method-lineages", "negative-results-preserved" ], "limitingObservations": [ "catalogue-assurance-boundary", "catalogue-task-mix" ], "predictions": [ { "id": "research-p1", "statement": "At matched scope and assurance, intake-to-accepted-artifact time and total active human time should be lower than an appropriate comparator.", "estimand": "Median ratio of intake-to-accepted-artifact calendar days and active human hours per accepted artifact.", "comparator": "Preregistered matched work using the incumbent workflow at the same scope and assurance endpoint.", "threshold": "Exploratory: both median ratios are at most 0.80, with no material increase in severe-error or correction rates.", "window": "From registered intake to accepted artifact; assess after at least 20 matched pairs and 12 months.", "measurementPlan": "Register scope and assurance target before work; capture timestamps and human time prospectively; use blinded independent acceptance and error adjudication." }, { "id": "research-p2", "statement": "Defects should be detected earlier and independent replay should begin sooner when a compact decision object is supplied.", "estimand": "Time to first correctly confirmed material defect and time from handoff to first successful independent replay.", "comparator": "Matched work without a compact decision object, at the same scope and assurance target.", "threshold": "Exploratory: both median times are at most 0.80 of comparator, without lower defect yield or replay success.", "window": "Research intake through 90 days after public handoff.", "measurementPlan": "Timestamp defect reports and replay events; blind adjudicators to workflow; exclude producer reruns from independent replay." }, { "id": "research-p3", "statement": "Evidence-backed research lineages should show declining reconstruction effort for linked follow-ups.", "estimand": "Active reconstruction hours required before substantive work begins, by ordinal position within a genuine shared lineage.", "comparator": "The first work in the same lineage and matched de novo follow-ups without a reusable handoff.", "threshold": "Exploratory: at least 20% lower median reconstruction effort by the second or third follow-up, with no increase in inherited critical defects.", "window": "First five lineages reaching at least three works, or 24 months.", "measurementPlan": "Distinguish true dependency lineages from thematic clusters; record prerequisite reconstruction, inherited defects and start of substantive work." } ], "potentialFalsifiers": [ "no material reduction in total human effort or elapsed time at matched assurance", "higher severe-error or correction rates at equal cost", "no faster independent maturation or downstream reuse", "the apparent gain disappears after task selection, lineage-aware dependence adjustment and attempted-project denominators are included" ] }, { "id": "publication-layer-acceleration", "hypothesis": "Automation may reduce claim-lock-to-open-archived-package time while preserving structured metadata and producer replay.", "scope": "The Evidence Press publication and packaging layer.", "aims": [ "science", "policy", "productivity" ], "epistemicStatus": "untested", "statusEvidence": { "status": "untested", "designClass": "none", "observationIds": [ "catalogue-baseline-throughput", "catalogue-assurance-boundary", "catalogue-task-mix" ], "evidenceRefs": [], "comparator": null, "estimand": null, "decisionRule": null, "registeredDesignRef": null, "independentReview": null, "reason": "No prospective claim-lock receipt, matched publication comparator or complete packaging and maintenance labour record has yet been evaluated.", "inferenceLimit": "The baseline establishes open publication capability and producer replay, not faster publication at matched endpoint and total cost." }, "rivals": [ "the publication sprint merely converted a backlog of completed work", "calendar speed is purchased with unrecorded intensive labour", "structured packaging adds later maintenance cost that offsets the initial gain" ], "supportingObservations": [ "catalogue-baseline-throughput", "catalogue-assurance-boundary" ], "limitingObservations": [ "catalogue-task-mix" ], "predictions": [ { "id": "publication-p1", "statement": "Prospective work records should show short claim-lock-to-public intervals across new releases.", "estimand": "Calendar days and active packaging hours from frozen claim lock to complete public archived package.", "comparator": "Matched incumbent publication route reaching the same open-access and producer-replay endpoint.", "threshold": "Exploratory: median duration is at most 7 calendar days and median active packaging time is at most 0.75 of comparator.", "window": "First 20 prospective releases and at least 12 months.", "measurementPlan": "Create immutable claim-lock and publication receipts; record all packaging, correction and coordination time; retain missing timestamps explicitly." }, { "id": "publication-p2", "statement": "Package completeness and replay should not deteriorate as release volume grows.", "estimand": "Package-completeness rate, clean producer-replay rate and maintenance hours per release as volume grows.", "comparator": "Successive rolling cohorts against the first prospective cohort at the same acceptance criteria.", "threshold": "Exploratory non-inferiority: no decline greater than 5 percentage points in completeness or replay and no increase greater than 20% in median maintenance hours.", "window": "Rolling cohorts of ten releases over 12 months.", "measurementPlan": "Run frozen automated checks on every release and an independent sampled audit; record corrections, broken assets and maintenance labour." } ], "potentialFalsifiers": [ "prospective claim-lock-to-public time is not materially shorter than the relevant comparator", "maintenance, correction or packaging labour erases the saving", "throughput falls after backlog conversion and lineage-aware dependence adjustment" ] }, { "id": "policy-assessment-improvement", "hypothesis": "Identification gates, rival-mechanism audits and bounded decision maps produce decision-adequate policy assessments faster and with fewer unsupported causal claims.", "scope": "Policy questions whose observations, estimand and rivals can be explicitly registered.", "aims": [ "policy" ], "epistemicStatus": "untested", "statusEvidence": { "status": "untested", "designClass": "none", "observationIds": [ "policy-identification-case", "negative-results-preserved", "catalogue-task-mix" ], "evidenceRefs": [], "comparator": null, "estimand": null, "decisionRule": null, "registeredDesignRef": null, "independentReview": null, "reason": "The single policy case illustrates a design-conditioned identification stop but has no matched expert-review comparison or downstream decision evaluation.", "inferenceLimit": "A useful formal diagnosis in one case does not establish faster assessment, fewer unsupported claims or better policy decisions." }, "rivals": [ "conventional expert review already supplies the same gain", "formal models omit decisive institutional or semantic facts", "correlated agents miss common defects", "political and implementation constraints dominate evidence quality" ], "supportingObservations": [ "policy-identification-case", "negative-results-preserved" ], "limitingObservations": [ "catalogue-task-mix" ], "predictions": [ { "id": "policy-p1", "statement": "Comparable cases should reach a justified terminal status sooner and contain fewer unsupported point attributions.", "estimand": "Time from case registration to justified terminal status and number of unsupported point attributions per case.", "comparator": "Matched conventional expert-led evidence assessment using the same corpus and decision question.", "threshold": "Exploratory: at least 20% lower median time and 20% fewer unsupported attributions, with non-inferior detection of valid effects and decision adequacy.", "window": "Case registration to terminal status across at least 12 matched cases.", "measurementPlan": "Freeze claims and corpora; use blinded expert adjudication of attribution support, missed effects and decision adequacy; retain blocked and negative cases." }, { "id": "policy-p2", "statement": "Failed identification should yield more discriminating and decision-relevant data plans.", "estimand": "Proportion of proposed data plans that pass a preregistered discrimination test and could change the identified set or robust decision.", "comparator": "Data recommendations produced by matched conventional expert reviews.", "threshold": "Exploratory: at least 20 percentage points higher discrimination-pass rate, without worse feasibility or expected collection cost.", "window": "At least 12 matched cases with six-month follow-up.", "measurementPlan": "Blind a multidisciplinary panel to workflow; score discrimination, feasibility, cost and decision relevance; track whether data owners act on the plan." } ], "potentialFalsifiers": [ "no gain over appropriate expert review", "valid effects are missed more often because the rival or observation model is too narrow", "downstream adjudicators judge the resulting decisions poorer or no faster" ] }, { "id": "assurance-bottleneck-relief", "hypothesis": "Moving facts and computations down a verifiability stack lets scarce human effort concentrate on semantics, values and authorisation without increasing consequential residual error.", "scope": "High-volume evidence work with repeatable checks and risk-based sampling.", "aims": [ "science", "policy", "productivity" ], "epistemicStatus": "untested", "statusEvidence": { "status": "untested", "designClass": "none", "observationIds": [ "catalogue-assurance-boundary", "catalogue-task-mix" ], "evidenceRefs": [], "comparator": null, "estimand": null, "decisionRule": null, "registeredDesignRef": null, "independentReview": null, "reason": "No qualifying corpus comparison measures human assurance minutes at a fixed consequential residual-error bound.", "inferenceLimit": "Automated checks and producer replay show that some checks can run mechanically, not that total human assurance burden falls safely." }, "rivals": [ "integration, security and procurement dominate total cost", "every item still requires bespoke human review", "verification capability cheapens no faster than generation", "authorisation rather than verification remains binding" ], "supportingObservations": [ "catalogue-assurance-boundary" ], "limitingObservations": [ "catalogue-task-mix" ], "predictions": [ { "id": "assurance-p1", "statement": "Exhaustive mechanical checks plus risk-based sampling should reduce human minutes per accepted claim at a fixed residual-error bound.", "estimand": "Human review minutes per accepted claim at a preregistered upper bound on consequential residual error.", "comparator": "Full manual review or incumbent assurance at the same residual-error target.", "threshold": "Exploratory efficiency signal: at least 25% fewer human minutes while the audited upper confidence bound remains below the registered error cap.", "window": "Each complete corpus and pooled analysis after at least three materially different corpora.", "measurementPlan": "Randomly sample and independently adjudicate checked and unchecked items; include integration, escalation and exception-handling time." }, { "id": "assurance-p2", "statement": "Human review should shift toward semantic bridges, causal assumptions, values and authorisation rather than repeatable arithmetic or source-presence checks.", "estimand": "Share of human review minutes spent on semantic, causal, value, rights and authorisation judgments rather than mechanical checks.", "comparator": "The incumbent manual workflow on matched corpora.", "threshold": "Exploratory: a shift of at least 20 percentage points toward judgment-intensive work, with no increase in total human time or residual-error bound.", "window": "The same corpora and period used for assurance-p1.", "measurementPlan": "Code review activities prospectively using a frozen taxonomy; audit time records and independently classify a sample of activities." } ], "potentialFalsifiers": [ "human assurance cost does not fall", "severe semantic errors escape more often", "sampling cannot meet the declared risk limit", "integration and security costs dominate the proposed saving" ] }, { "id": "open-evidence-reuse", "hypothesis": "DOI-backed, openly licensed, machine-readable evidence packages shorten discovery, independent replay and downstream reuse.", "scope": "Potential users able to access and interpret public digital research objects.", "aims": [ "science", "policy", "productivity" ], "epistemicStatus": "untested", "statusEvidence": { "status": "untested", "designClass": "none", "observationIds": [ "catalogue-baseline-throughput", "method-lineages", "catalogue-assurance-boundary" ], "evidenceRefs": [], "comparator": null, "estimand": null, "decisionRule": null, "registeredDesignRef": null, "independentReview": null, "reason": "No matched comparison yet measures independent time-to-replay, reconstruction effort or unaffiliated downstream reuse.", "inferenceLimit": "Open structured objects exist, but availability and internal lineage reuse do not establish external reuse or time savings." }, "rivals": [ "reputation and attention dominate access", "unrefereed status suppresses use", "complex packaging overwhelms users", "access is not the binding downstream bottleneck" ], "supportingObservations": [ "catalogue-baseline-throughput", "method-lineages" ], "limitingObservations": [ "catalogue-assurance-boundary" ], "predictions": [ { "id": "reuse-p1", "statement": "Structured releases should receive earlier independent reruns, forks, citations or derivative use than comparable unstructured open candidates.", "estimand": "Time to first qualifying unaffiliated rerun, fork, substantive citation or derivative use, plus 12-month qualifying-event incidence.", "comparator": "Matched unstructured open candidates with similar field, status, age and visibility.", "threshold": "Exploratory: median time to event is at most 0.75 of comparator or 12-month incidence is at least 15 percentage points higher.", "window": "Twelve months after each release.", "measurementPlan": "Predefine qualifying events; exclude producer activity and superficial mentions; verify events manually and adjust descriptively for visibility." }, { "id": "reuse-p2", "statement": "Independent users should reach a first successful replay with less reconstruction work.", "estimand": "Independent-user active hours to first successful replay and replay-success proportion.", "comparator": "The same research content supplied as an unstructured PDF and repository without the structured package.", "threshold": "Exploratory: at least 30% lower median active hours with replay success non-inferior within 10 percentage points.", "window": "Thirty days per attempt across at least 20 users and five artifacts.", "measurementPlan": "Use parallel assignment, standardized environments and blinded replay adjudication; retain failed and abandoned attempts." } ], "potentialFalsifiers": [ "no improvement in time-to-replay or derivative use", "users consistently abandon the structured artefacts", "maintenance cost exceeds the reuse saving" ] }, { "id": "operations-certificate-value", "hypothesis": "Exact certificates expose heuristic error and improve operational choices enough to exceed certification cost.", "scope": "Operational models whose constraints and objectives are represented with sufficient fidelity.", "aims": [ "policy", "productivity" ], "epistemicStatus": "untested", "statusEvidence": { "status": "untested", "designClass": "none", "observationIds": [ "operations-theory-no-field", "catalogue-task-mix" ], "evidenceRefs": [], "comparator": null, "estimand": null, "decisionRule": null, "registeredDesignRef": null, "independentReview": null, "reason": "No external operational comparison measures plan changes, realised avoided loss, implementation cost or performance under omitted constraints and drift.", "inferenceLimit": "Exact model-specific certificates establish conditional mathematical properties, not field decision value or net operational benefit." }, "rivals": [ "model misspecification dominates optimisation error", "existing heuristics are operationally adequate", "certificate generation costs more than avoided loss", "omitted constraints reverse the nominal optimum" ], "supportingObservations": [ "operations-theory-no-field" ], "limitingObservations": [ "catalogue-task-mix" ], "predictions": [ { "id": "operations-p1", "statement": "External operational use should find material incumbent gaps or decision boundaries often enough to change accepted plans.", "estimand": "Proportion of eligible operational instances where certification reveals a preregistered material incumbent gap or changes the approved plan.", "comparator": "The incumbent heuristic-only decision on the same instance.", "threshold": "Set the material-gap threshold by context before analysis; exploratory programme signal if at least 20% of eligible instances change plan without safety regression.", "window": "At least 20 consecutive eligible instances or one full annual planning cycle.", "measurementPlan": "Run certification on all eligible instances, not selected successes; record incumbent plan, certified result, approval decision, costs and constraint violations." }, { "id": "operations-p2", "statement": "The avoided loss or bounded regret should exceed the full cost of certification and implementation.", "estimand": "Realised avoided loss or bounded regret minus certification, implementation, monitoring and switching costs.", "comparator": "Incumbent operational policy under randomized, stepped or otherwise defensible concurrent comparison.", "threshold": "Positive context-specific net value; confirmatory support requires the uncertainty interval for aggregate net value to exclude zero.", "window": "At least one complete planning cycle with follow-up through subsequent replanning or drift.", "measurementPlan": "Preregister the counterfactual design and cost boundary; measure realised outcomes and all human and tool costs; label non-identified comparisons as associations." } ], "potentialFalsifiers": [ "certificates rarely change decisions", "certification cost exceeds avoided loss", "certified nominal policies perform worse after realistic constraints and drift are included" ] }, { "id": "negative-result-option-value", "hypothesis": "Published blockers, counterexamples and stop receipts prevent duplicated dead ends and redirect effort toward discriminating data or tractable subproblems.", "scope": "Research and policy programmes with expensive downstream continuation and reusable failure information.", "aims": [ "science", "policy", "productivity" ], "epistemicStatus": "untested", "statusEvidence": { "status": "untested", "designClass": "none", "observationIds": [ "negative-results-preserved", "policy-identification-case", "catalogue-assurance-boundary" ], "evidenceRefs": [], "comparator": null, "estimand": null, "decisionRule": null, "registeredDesignRef": null, "independentReview": null, "reason": "Preservation is observed, but no matched follow-up measures avoided duplicated work, earlier justified stopping or productive redirection.", "inferenceLimit": "The existence of negative artefacts does not show that later users find, understand or act on them." }, "rivals": [ "negative outputs are ignored", "failure context is not transferable", "documentation overhead exceeds avoided work", "independent rediscovery is cheaper than understanding the receipt" ], "supportingObservations": [ "negative-results-preserved", "policy-identification-case" ], "limitingObservations": [ "catalogue-assurance-boundary" ], "predictions": [ { "id": "negative-p1", "statement": "Later work should cite and reuse blockers, abandon registered dead routes earlier or collect the identified missing information.", "estimand": "Proportion of stop receipts followed by documented reuse, earlier route abandonment or collection of the specified missing information, and time to that event.", "comparator": "Matched negative or stalled work without a structured stop receipt.", "threshold": "Exploratory: at least 15 percentage points higher qualifying-event incidence or 20% shorter median time to justified abandonment.", "window": "Twenty-four months after each stop receipt.", "measurementPlan": "Register dead routes and reopening conditions; verify external citations, pivots and data collection; exclude producer-only restatements." }, { "id": "negative-p2", "statement": "Method salvage should generate independent follow-ups without presenting the parent target as resolved.", "estimand": "Proportion of salvaged-method releases producing an unaffiliated follow-up that uses the method while correctly retaining the unresolved parent boundary.", "comparator": "Matched stalled projects without an explicit reusable-method handoff.", "threshold": "Exploratory: at least 25% yield a qualifying follow-up within 24 months and at least 90% of qualifying follow-ups preserve the parent boundary.", "window": "Twenty-four months after each salvage release.", "measurementPlan": "Track citations, forks and notifications; have blinded reviewers assess substantive method use and boundary fidelity." } ], "potentialFalsifiers": [ "the same routes are repeatedly retried without engaging the prior stop receipt", "no documented pivot or reuse occurs", "total effort is unchanged or higher because the negative artefacts are too costly to interpret" ] }, { "id": "protocol-workplace-productivity", "hypothesis": "Bounded protocols improve accepted work per unit of total human resource relative to the same agent without the protocol.", "scope": "Context-bound knowledge-work tasks and intended users measured under an appropriate comparison.", "aims": [ "productivity" ], "epistemicStatus": "benchmark-no-clear-gain", "statusEvidence": { "status": "benchmark-no-clear-gain", "designClass": "model-benchmark", "observationIds": [ "productivity-protocol-no-impact" ], "evidenceRefs": [ { "kind": "internal-artifact", "ref": "protocols/protocols/document-to-action-plan/evals/result-live-o4-mini-2026-08-08.json", "role": "benchmark-result" }, { "kind": "internal-artifact", "ref": "protocols/protocols/evidence-backed-brief/evals/result-live-o4-mini-2026-08-08.json", "role": "benchmark-result" }, { "kind": "internal-artifact", "ref": "protocols/protocols/goal-to-verified-deliverable/evals/result-live-gpt-5.2-2026-08-08-xmodel.json", "role": "benchmark-result" } ], "comparator": "The same named agent model on the same registered synthetic tasks without the protocol prompt; the no-agent human arm was not run.", "estimand": "Protocol-arm minus agent-without-protocol differences in recorded completion, judged output quality or accuracy, detected safety events, elapsed model time and estimated model cost on the preserved task sets.", "decisionRule": "The preserved version 0.1.0 benchmark records return NO_CLEAR_GAIN when the protocol arm does not demonstrate a decision-relevant net gain; detected safety or acceptance regression takes precedence as harm or regression.", "registeredDesignRef": null, "independentReview": null, "benchmarkBoundary": "Development-team model-output benchmarks of three version 0.1.0 predecessor protocols on tiny curated task sets; current version 0.1.1 packs do not inherit the result.", "inferenceLimit": "The benchmarks used internally judged model outputs, not consenting workers or organisational outcomes; they do not estimate human or company productivity and were not independently reviewed or reproduced." }, "rivals": [ "added structure is pure overhead", "benefit depends on expert users", "base-model capability dominates the protocol", "support, review and rework erase output gains" ], "supportingObservations": [], "limitingObservations": [ "productivity-protocol-no-impact" ], "predictions": [ { "id": "productivity-p1", "statement": "A suitable comparison should show lower active human time or rework at non-inferior blinded quality without higher material error, burden, support cost or safety risk.", "estimand": "Active human minutes and rework per accepted work item, with blinded quality, material error, burden, support cost and safety as separate outcomes.", "comparator": "Randomized parallel use of the same agent, model, tools and task without the protocol; manual work as a secondary operational baseline.", "threshold": "Exploratory: at least 15% reduction in the primary human-resource outcome, quality non-inferior within a preregistered margin, and no decision-relevant worsening of error, burden, support or safety.", "window": "Frozen evaluation period and justified sample set before allocation.", "measurementPlan": "Freeze tasks and configuration; conceal quality keys; blind raters; retain missing and stopped work; measure facilitator and approver time as well as participant time." }, { "id": "productivity-p2", "statement": "Any initial gain should persist in 30-day and 90-day adoption records.", "estimand": "Verified actual-use proportion at days 30 and 90 and retention of the initial accepted-work-per-human-resource effect.", "comparator": "Immediate post-evaluation use and the randomized agent-only group where continued access permits comparison.", "threshold": "Exploratory: at least 50% verified use at day 90 and retention of at least 75% of any initial productivity signal, with no emerging harm.", "window": "Day 30 and day 90 after supported use ends.", "measurementPlan": "Use activity records where lawful, supplement with structured follow-up, retain nonresponse and reasons for nonuse, and do not treat adoption alone as benefit." } ], "potentialFalsifiers": [ "the interval excludes a decision-relevant gain", "quality or safety worsens", "total human resource use rises", "adoption rapidly decays after supported use" ] } ], "updatePolicy": { "cadence": "Update when a new release, external review, independent reproduction, correction, operational use or material failure bears on a hypothesis; review the full ledger at least quarterly while the programme is active.", "rules": [ "Append or date material changes; do not silently rewrite an earlier observation.", "Do not promote a hypothesis from publication throughput alone.", "Record evidence that weakens the focal explanation with the same prominence as supporting evidence.", "Keep process conformance, research assurance and real-world impact separate.", "Apply the registered status definition and entry criteria; a promotion requires new referenced observations capable of supporting the promoted claim.", "Record every status change with its prior and new status, hypothesis id and evidential basis in the append-only change log.", "Append corrections and superseding interpretations as new revisions; never delete or rewrite an earlier revision record." ] }, "changeLog": [ { "sequence": 1, "revisionId": "initial-ledger-2026-08-10", "previousRevisionId": null, "recordedAt": "2026-08-10", "changeType": "initialization", "scope": "ledger", "basisObservationIds": [ "catalogue-baseline-throughput", "catalogue-assurance-boundary", "catalogue-task-mix", "method-lineages", "negative-results-preserved", "policy-identification-case", "operations-theory-no-field", "productivity-protocol-no-impact" ], "summary": "Created the initial abductive ledger from the 21-release adoption baseline and retained the strongest rival explanations." }, { "sequence": 2, "revisionId": "calibrate-focal-explanation-2026-08-10", "previousRevisionId": "initial-ledger-2026-08-10", "recordedAt": "2026-08-10", "changeType": "claim-revised", "scope": "ledger", "basisObservationIds": [ "catalogue-baseline-throughput", "catalogue-assurance-boundary", "catalogue-task-mix", "productivity-protocol-no-impact" ], "summary": "Separated demonstrated publication capability and producer replay from untested comparative claims about discovery, policy, reuse, operations and productivity; added explicit status entry criteria and claim ceilings." }, { "sequence": 3, "revisionId": "calibrate-publication-status-2026-08-10", "previousRevisionId": "calibrate-focal-explanation-2026-08-10", "recordedAt": "2026-08-10", "changeType": "status-changed", "scope": "hypothesis", "hypothesisId": "publication-layer-acceleration", "fromStatus": "mechanism-supported", "toStatus": "untested", "basisObservationIds": [ "catalogue-baseline-throughput", "catalogue-assurance-boundary" ], "summary": "The catalogue shows publication capability, but no prospective claim-lock comparator or total-effort measurement yet establishes publication acceleration." }, { "sequence": 4, "revisionId": "calibrate-policy-status-2026-08-10", "previousRevisionId": "calibrate-publication-status-2026-08-10", "recordedAt": "2026-08-10", "changeType": "status-changed", "scope": "hypothesis", "hypothesisId": "policy-assessment-improvement", "fromStatus": "mechanism-supported", "toStatus": "untested", "basisObservationIds": [ "policy-identification-case", "catalogue-task-mix" ], "summary": "The housing case illustrates an identification stop, not a comparative effect on assessment speed, unsupported claims or policy decisions." }, { "sequence": 5, "revisionId": "calibrate-assurance-status-2026-08-10", "previousRevisionId": "calibrate-policy-status-2026-08-10", "recordedAt": "2026-08-10", "changeType": "status-changed", "scope": "hypothesis", "hypothesisId": "assurance-bottleneck-relief", "fromStatus": "single-case-supported", "toStatus": "untested", "basisObservationIds": [ "catalogue-assurance-boundary", "catalogue-task-mix" ], "summary": "The ledger did not contain a qualifying measured case of reduced human assurance time at a fixed residual-error bound." }, { "sequence": 6, "revisionId": "calibrate-reuse-status-2026-08-10", "previousRevisionId": "calibrate-assurance-status-2026-08-10", "recordedAt": "2026-08-10", "changeType": "status-changed", "scope": "hypothesis", "hypothesisId": "open-evidence-reuse", "fromStatus": "mechanism-supported", "toStatus": "untested", "basisObservationIds": [ "catalogue-baseline-throughput", "method-lineages", "catalogue-assurance-boundary" ], "summary": "Open structured artifacts exist, but comparative time-to-replay and downstream reuse have not been measured." }, { "sequence": 7, "revisionId": "calibrate-operations-status-2026-08-10", "previousRevisionId": "calibrate-reuse-status-2026-08-10", "recordedAt": "2026-08-10", "changeType": "status-changed", "scope": "hypothesis", "hypothesisId": "operations-certificate-value", "fromStatus": "mechanism-supported", "toStatus": "untested", "basisObservationIds": [ "operations-theory-no-field", "catalogue-task-mix" ], "summary": "Exact model-specific certificates do not establish decision changes, avoided loss or net value in operational use." }, { "sequence": 8, "revisionId": "calibrate-negative-result-status-2026-08-10", "previousRevisionId": "calibrate-operations-status-2026-08-10", "recordedAt": "2026-08-10", "changeType": "status-changed", "scope": "hypothesis", "hypothesisId": "negative-result-option-value", "fromStatus": "mechanism-supported", "toStatus": "untested", "basisObservationIds": [ "negative-results-preserved", "policy-identification-case" ], "summary": "Negative and stopped outputs are preserved, but later avoidance of duplicated work or redirection of effort has not been measured." }, { "sequence": 9, "revisionId": "operationalize-predictions-2026-08-10", "previousRevisionId": "calibrate-negative-result-status-2026-08-10", "recordedAt": "2026-08-10", "changeType": "prediction-revised", "scope": "ledger", "basisObservationIds": [], "summary": "Added an estimand, comparator, explicit exploratory or confirmatory threshold, observation window and measurement plan to every registered prediction." }, { "sequence": 10, "revisionId": "separate-clusters-from-lineages-2026-08-10", "previousRevisionId": "operationalize-predictions-2026-08-10", "recordedAt": "2026-08-10", "changeType": "claim-revised", "scope": "ledger", "basisObservationIds": [ "method-lineages" ], "summary": "Restricted lineage claims to the two programmes with explicit dependency evidence and kept broader method clusters separate." }, { "sequence": 11, "revisionId": "add-status-evidence-foundation-2026-08-10", "previousRevisionId": "separate-clusters-from-lineages-2026-08-10", "recordedAt": "2026-08-10", "changeType": "claim-revised", "scope": "ledger", "basisObservationIds": [ "productivity-protocol-no-impact" ], "summary": "Added explicit science, policy and productivity aims and status-specific evidence records to every hypothesis without promoting any status; preserved the predecessor protocol result as an internally produced model-benchmark no-clear-gain finding with no independent-review claim." }, { "sequence": 12, "revisionId": "classify-observation-capabilities-2026-08-10", "previousRevisionId": "add-status-evidence-foundation-2026-08-10", "recordedAt": "2026-08-10", "changeType": "claim-revised", "scope": "ledger", "basisObservationIds": [ "catalogue-baseline-throughput", "catalogue-assurance-boundary", "catalogue-task-mix", "method-lineages", "negative-results-preserved", "policy-identification-case", "operations-theory-no-field", "productivity-protocol-no-impact" ], "summary": "Classified each observation by the strongest epistemic status it can help support, so a throughput or mechanism observation cannot by itself license a benchmark, field or causal promotion." } ] }