{ "schemaVersion": "1.0", "recordType": "evidence-atlas-roadmap", "updated": "2026-08-15", "status": "active-reviewed-roadmap", "claimCeiling": "This roadmap records intended work and review decisions. It does not establish that an undiscovered relationship exists, that a proposed relationship is correct, or that the Atlas improves research outcomes.", "currentBaseline": { "asOf": "2026-08-15", "releaseCount": 28, "methodCount": 13, "clusterCount": 13, "lineageCount": 2, "acceptedRelationshipCount": 175, "publishedProposalCount": 0, "relationshipComposition": { "usesMethodCount": 135, "clusterMembershipCount": 28, "lineageMembershipCount": 7, "directInterReleaseCount": 5, "defaultProgrammeViewCount": 40 }, "methodPrevalenceChecks": [ { "methodId": "adversarial-controls", "releaseAssignmentCount": 20 }, { "methodId": "structural-compression", "releaseAssignmentCount": 18 } ], "sourceRefs": [ "data/RELATIONSHIP_REGISTRY.json", "data/METHOD_REGISTRY.json", "papers/*/meta.json" ] }, "principles": [ "Keep asserted, computed and proposed relationships machine-readably distinct.", "Do not infer correctness, novelty, priority, influence, independence or impact from graph geometry, proximity or similarity.", "Require resolvable evidence, an inference limit and additive review history for every proposal decision.", "Retain rejected and superseded proposals as negative knowledge once the proposal layer is activated.", "Keep machine access independent of the interactive renderer." ], "nextSteps": [ { "id": "atlas-step-proposal-intake", "priority": 1, "state": "ready", "title": "Implement fail-closed proposal intake and review receipts", "purpose": "Give people and agents a safe place to nominate relationships without adding them to the accepted graph.", "dependencies": [], "deliverables": [ "A deterministic proposal validator and submission format.", "Content-derived proposal identifiers and resolvable evidence references.", "Append-only review receipts for awaiting-review, accepted-as-asserted, rejected and superseded states.", "Negative controls for unresolved endpoints, unknown predicates, missing evidence and accidental promotion." ], "acceptanceCriteria": [ "A valid proposal is published only in the proposal register and never in the accepted edge array.", "Invalid, incomplete and self-promoting proposals are rejected deterministically.", "A review decision records reviewer provenance, basis, inference limit, date and predecessor state.", "Rejected and superseded records remain retrievable." ], "stopConditions": [ "Stop if proposal history cannot remain append-only or if review provenance cannot be represented without overwriting prior states.", "Stop if any ingestion route can write directly to the accepted graph." ] }, { "id": "atlas-step-projection-calibration", "priority": 2, "state": "ready", "title": "Calibrate projection weight and taxonomy legibility", "purpose": "Make the first visual impression distinguish direct inter-release evidence from registry-derived navigation and make broad, low-discrimination labels visibly interpretable.", "dependencies": [], "deliverables": [ "A tested direct-inter-release projection alongside the existing programme and method views.", "Visible working-taxonomy status, edge-composition counts and method prevalence.", "A documented lineage eligibility rule requiring a reciprocal lineage identifier and, for a non-root member, an evidential parent link.", "A design note on representing machine-checked reciprocity as an evidence attribute rather than a replacement for knowledge status." ], "acceptanceCriteria": [ "The initial projection does not make registry-derived density appear to be direct inter-release evidence.", "Ubiquitous methods and singleton clusters carry enough prevalence or scope context to prevent over-interpretation.", "Every projection reconciles exactly with the canonical graph and preserves the complete nonvisual register." ], "stopConditions": [ "Do not remove programme or method navigation merely to make the graph look sparse.", "Stop if a projection requires a second relationship ontology or changes accepted graph semantics." ] }, { "id": "atlas-step-discovery-pilot", "priority": 3, "state": "blocked", "title": "Run a bounded relationship-discovery pilot", "purpose": "Test whether multiple transparent retrieval routes can nominate useful hidden connections at a tolerable false-proposal burden.", "dependencies": [ "atlas-step-proposal-intake" ], "deliverables": [ "A frozen-corpus candidate generator using separately reported citation, symbolic, shared-object and semantic-retrieval routes.", "Route-specific evidence rather than one opaque credibility score.", "A proposal-only output with no power to modify accepted relationships." ], "acceptanceCriteria": [ "A frozen hidden-link benchmark and lexically similar negative controls are registered before evaluation.", "Recovery, false-proposal burden and reviewer time are reported separately.", "Every candidate includes the strongest plausible non-relationship explanation." ], "stopConditions": [ "Stop if the benchmark cannot be frozen independently of the generator.", "Stop if candidate generation cannot be isolated from accepted graph production.", "Stop or narrow the route if review burden outweighs useful recoveries." ] }, { "id": "atlas-step-review-workbench", "priority": 4, "state": "blocked", "title": "Add a proposal-review workbench", "purpose": "Let reviewers inspect a proposed edge, its evidence, contrary explanation and full decision history without confusing it with accepted knowledge.", "dependencies": [ "atlas-step-proposal-intake" ], "deliverables": [ "A distinct proposals view with accessible nonvisual records.", "Filters for awaiting-review, accepted, rejected and superseded proposals.", "Links from every decision to source artefacts and review receipts." ], "acceptanceCriteria": [ "Proposal state is visible in text, structure and machine data, not colour alone.", "No proposal geometry or ranking is described as evidence.", "The accepted graph remains usable when the proposal layer is unavailable." ], "stopConditions": [ "Stop if the interface visually collapses proposed and accepted relationships.", "Stop if a no-JavaScript or nonvisual reviewer cannot retrieve the same decision record." ] }, { "id": "atlas-step-agent-queries", "priority": 5, "state": "watch", "title": "Add bounded agent query surfaces when use cases justify them", "purpose": "Reduce token and reconstruction cost for agents seeking neighbourhoods, dependencies, methods or unresolved proposals without replacing the canonical graph export.", "dependencies": [], "deliverables": [ "Documented filtered-query patterns or endpoints with stable identifiers and pagination.", "Responses that always carry source pointers, status and inference limits." ], "acceptanceCriteria": [ "A demonstrated agent task is materially cheaper than retrieving the complete graph.", "Filtered responses reconcile exactly with the canonical graph and proposal register." ], "stopConditions": [ "Defer while the complete graph remains small and cheaper to retrieve.", "Stop if query convenience creates a second semantic source of truth." ] }, { "id": "atlas-step-scale-renderer", "priority": 6, "state": "watch", "title": "Reassess the renderer at catalogue scale", "purpose": "Preserve usable navigation as the graph grows without making a visualization library part of the evidence model.", "dependencies": [], "deliverables": [ "Repeated desktop, mobile, keyboard and performance comparisons against the dependency-free SVG.", "A bounded evaluation of Cytoscape.js or Sigma.js plus Graphology only if thresholds are crossed." ], "acceptanceCriteria": [ "The current renderer remains the default until measured usability or performance degrades.", "Any replacement consumes the same canonical JSON and preserves the nonvisual register." ], "stopConditions": [ "Do not migrate for aesthetic preference alone.", "Reassess when visible nodes exceed 250, interaction latency becomes material, or accessibility regresses." ] }, { "id": "atlas-step-outcome-evaluation", "priority": 7, "state": "ready", "title": "Measure whether the Atlas reduces rediscovery and reconstruction cost", "purpose": "Test the Atlas as a research tool without inferring value from page views, graph size or proposal volume.", "dependencies": [], "deliverables": [ "Prospective records of qualifying discovery, replay, reuse, abandoned duplicate routes and reviewer effort.", "A comparison plan aligned with the open-evidence-reuse hypothesis in the IBE ledger." ], "acceptanceCriteria": [ "Qualifying outcomes, comparator, observation window and missingness are registered before analysis.", "Producer use, unaffiliated use and superficial mentions remain separate.", "Null, adverse and abandoned outcomes remain in the denominator." ], "stopConditions": [ "Do not claim research acceleration from traffic, catalogue growth or internal reuse alone.", "Stop causal interpretation if selection, visibility and correlated lineage reuse cannot be addressed." ] } ], "reviewPolicy": { "cadence": "Review on maintainer request and at least monthly while active; also review after five added releases, the first submitted proposal, a material objection or correction, or a renderer threshold event.", "nextScheduledReview": "2026-09-15", "triggers": [ "maintainer-request", "five-releases-added", "proposal-submitted", "material-objection-or-correction", "renderer-threshold-crossed", "monthly-review-due" ], "checklist": [ "Rebuild the graph and compare release, method, cluster, lineage, accepted-edge and proposal counts with the prior review.", "Confirm every release and accepted relationship still resolves to its declared source record.", "Sample accepted edges across predicates and recheck basis and inference limits against source artefacts.", "Inspect every new, awaiting, rejected and superseded proposal without promoting similarity to fact.", "Review new external citations, objections, corrections and reproductions that may change a relationship basis.", "Assess accessibility, mobile layout, graph performance and agent retrieval cost.", "Record decisions, unresolved questions, observed regressions and one exact next action in a new review-log entry." ], "recordRule": "Append a dated review-log entry. Do not rewrite a prior review, delete a rejected proposal, or change an accepted relationship without the repository's normal additive evidence and publication gates." }, "reviewLog": [ { "reviewId": "atlas-review-2026-08-15-baseline", "reviewedAt": "2026-08-15", "trigger": "initial-roadmap-record", "reviewer": "Evidence Press producer workflow", "provenanceNote": "Producer-side baseline review against the repository source graph and publication gates.", "baseline": { "releaseCount": 28, "acceptedRelationshipCount": 175, "publishedProposalCount": 0 }, "observations": [ "The first source-driven Atlas is live with accepted edges generated from declared repository records.", "The proposal register is structurally present but empty.", "No discovery model or similarity route currently contributes accepted or proposed relationships.", "The current graph remains below the renderer reassessment threshold." ], "decision": "continue", "nextAction": "Implement fail-closed proposal intake, validation and append-only review receipts before running relationship-discovery experiments." }, { "reviewId": "atlas-review-2026-08-15-fable-comment", "reviewedAt": "2026-08-15", "trigger": "user-supplied-external-model-comment", "reviewer": "Anthropic Fable, as identified by the maintainer", "provenanceNote": "The maintainer supplied the review text. The exact model version, prompt and run record were not independently established.", "baseline": { "releaseCount": 28, "acceptedRelationshipCount": 175, "publishedProposalCount": 0 }, "observations": [ "The reviewer independently reconciled 13 methods, 13 clusters, 2 lineages, 28 release assignments and all 175 accepted edges from the public registry and Atlas.", "The current default programme projection already excludes 135 uses-method edges, but 35 of its 40 visible edges are cluster or lineage memberships and only 5 are direct inter-release links.", "The prevalence concern is supported by the source registry: adversarial-controls is assigned to 20 of 28 releases and structural-compression to 18 of 28.", "The proposed Polydegree lineage is not eligible under the registered rule because the successor declares an evidential parent link but its reciprocal lineageId is null; the public explanation should surface that rule.", "A reciprocity check could add useful evidence, but reciprocity should remain separate from the asserted, computed and proposed knowledge-status axis." ], "decision": "change-priority", "nextAction": "Keep fail-closed proposal intake first, then test an evidence-calibrated projection and surface taxonomy status, prevalence and lineage eligibility before the discovery pilot." } ] }