Exact blockers
Reason-coded backlog
Missing locators, rights notes, publication dates, replication assessments, and integrity errors remain individually inspectable.
Phase 1 · publication gateway · maha-epistemic/1.0
The public page is the result of a passed gate.
Maha’s underlying graph may contain drafts, hypotheses, conflicts, and incomplete records. Only records with explicit epistemic axes, source rights, locators, uncertainty, boundaries, and an approving review event become crawlable Knowledge pages.
Iceberg boundary
Publication is a controlled state transition.
Graph records
290
Canonical and withheld pilot records.
Public canonical
2
Records that generate pages and sitemap rows.
Withheld
288
Visible only as gate decisions, not public claims.
Non-transfer rule
Peer review does not mean independent replication. A formal model does not establish hardware readiness. A cell-system result does not establish clinical benefit. Each proposition retains its own evidence axis.
Persistent ingestion and scoped review
Five legacy systems now meet the same gate.
The semiconductor, mathematics, astronomy, religion, and neuromorphic adapters preserve 154 source records as immutable candidates. Each batch survives deployments, and every review decision binds a versioned identity, declared review method, one scope, and the candidate’s frozen content hash. External expert, internal editorial, and automated verification methods remain distinct.
Adapters
5
Imported candidates
154
Withheld pending review
154
Phase 2 · operational review
Withheld no longer means invisible or unstructured.
A private work queue now classifies every gate reason into source completion, scoped review, or release control. Operators can triage and assign source work, attach bounded evidence to exact blockers, and route each frozen target into four method-declared review queues.
Append-only work
Evidence without mutation
Assignments and evidence submissions bind to the old target hash; they cannot silently rewrite the ingested candidate.
Scoped handoff
Review readiness, not approval
Corrected content must be re-ingested under a new hash before reviewer decisions can count toward the publication gate.
Controlled re-ingestion compiler
A ready target can now become a new immutable revision through a restricted compiler. The server derives each editable field from an exact blocker, verifies its prior evidence event, emits a before/after diff, assigns new candidate and review digests, clears earlier review events, and resets publication controls to draft. Declared source-to-claim mismatches can be refined, replaced, or split only through a typed, evidence-bound alignment correction with explicit claim remapping. Arbitrary JSON patches and automatic promotion are not supported.
Phase 3 · controlled publication
Approval and release authority are different decisions.
A candidate can become canonical only when every required scope approves the same frozen target hash and a separately authenticated human release authority accepts the publication-state decision. Each decision discloses whether it came from an external expert, an internal editorial protocol, or an automated verifier. The operations credential is rejected by the release surface.
Exact target
Hash-bound approvals
A changed claim, source, boundary, section, or bridge changes the target digest and makes earlier decisions inapplicable.
Separate authority
No inherited power
Ingestion, source-completion, re-ingestion, and scoped-review credentials cannot publish a canonical route.
Append-only history
Supersede or withdraw
A correction creates a new release event. Older records remain in the public provenance ledger with an explicit status.
Public projection
Only the active release generates a canonical database-backed page and sitemap row. The machine-readable registry excludes credentials, operational fingerprints, private reviewer profiles, and non-consented authority identity fields.
Phase 4 · canonical corpus operations
The protocol now has a frozen operating sample.
A public manifest fixes 20 records across 5 domains before expert decisions are collected. One-time invitations bind a versioned reviewer identity to one required scope on one latest target digest; plaintext credentials are never persisted.
Selection fixed first
No success-only substitution
The selected record IDs, rationales, initial hashes, and blocker counts are public before the review lifecycle runs.
Least authority
One reviewer, scope, and hash
An invitation cannot open another record, review another scope, operate the corpus, re-ingest content, or authorize release.
Atomic terminal event
Consume or revoke
The ledger permits one terminal event. Review creation and invitation consumption occur in the same database transaction.
Phases 5–8 · internal scale
Automated volume stops before authority.
The noncanonical factory compiles immutable audits, source/claim matrices, and reviewer-ready packets across bounded batches. Public-authority calculation fixtures run in CI, while candidate packets remain private, noindex, absent from the sitemap, and incapable of satisfying expert review.
Open the publishing factory protocolEvidence-clearing guide
Source identity for agent provenance
How should source identity be established for agent provenance?
Evidence-clearing guide
Conflict and uncertainty for DOI version relationships
How should conflict and uncertainty be recorded for DOI version relationship?
Evidence-clearing guide
Conflict and uncertainty for machine-generated claims
How should conflict and uncertainty be recorded for machine-generated claim?
Evidence-clearing guide
Locator sufficiency for DOI version relationships
What exact locator is sufficient for DOI version relationship?
Independent axes
No single confidence label can carry all meanings.
Claim kind
What kind of proposition is being made, independently of how much evidence supports it.
observationempirical claimtheoretical modelformal propositioninterpretationhypothesis
Evidence maturity
What the compiled evidence establishes about testing, replication, conflict, or formal verification.
not assessednot applicablesingle studymulti studyindependently replicatedcontestedhistorical attestationformally verified
Review state
Where the record sits in Maha’s editorial lifecycle; this is not a scientific evidence score.
draftin reviewpublished canonicalwithdrawn
Bridge type
Whether a cross-domain link is exact, mechanistic, statistical, analogical, instrumental, or strategic.
mathematical equivalenceshared instrumentationmechanistic dependencystatistical associationstructural analogystrategic dependency
Adversarial pilots
Two domains, two different failure modes.
Phase 1 pilot
Quantum systems and advanced energy
Physical architectures, control systems, measurements, models, and readiness boundaries kept separate from formal possibility claims.
Formal mathematics can describe a valid architecture without establishing that a physical implementation is manufacturable, fault tolerant, or commercially ready.
Open governed domain →
Phase 1 pilot
Synthetic biology and cellular engineering
Molecular tools, cell systems, protocols, measurements, and safety boundaries represented with explicit experimental scope.
Performance in a cell line, organoid, animal model, and human intervention are different evidence states and cannot be silently collapsed.
Open governed domain →
Phase 1 pilot
Fusion and plasma systems
Confinement, magnets, plasma control, heat exhaust, fuel cycles, diagnostics, and inertial target systems separated by physical and engineering evidence state.
A plasma-physics result does not establish net electricity, component lifetime, tritium self-sufficiency, maintainability, or commercial plant economics.
Open governed domain →
Phase 1 pilot
Advanced materials
Two-dimensional materials, moiré systems, topological phases, wide-bandgap substrates, interfaces, fabrication, and metrology under specimen-specific contracts.
A property observed in one flake, stack, temperature range, or device geometry cannot be transferred to wafer-scale yield or product performance.
Open governed domain →
Phase 1 pilot
Biomolecular engineering
Protein design, directed evolution, cell-free expression, RNA control, enzyme cascades, and assay provenance separated by experimental system.
A computational design or in-vitro assay result does not establish folding, function, delivery, organism-level safety, or therapeutic benefit.
Open governed domain →
Phase 1 pilot
Longevity and metabolism
Autophagy, mitochondrial respiration, senescence, NAD+ metabolism, nutrient sensing, and intervention endpoints with assay-specific boundaries.
A molecular marker, cell assay, or animal lifespan result cannot be silently converted into a human healthspan or treatment claim.
Open governed domain →
Phase 1 pilot
Neurotechnology and BCI
Neural probes, cortical arrays, optical control, decoding, stimulation, chronic interfaces, telemetry, and translation constraints.
A decoding result in a named participant, task, session, implant, or animal model does not establish durable general-purpose communication or clinical benefit.
Open governed domain →
Phase 1 pilot
Mechanistic interpretability
Features, circuits, interventions, sparse representations, causal tests, and faithfulness criteria represented separately from explanatory confidence.
A human-readable feature label, probe, attention pattern, or ablation effect does not by itself establish a complete or faithful causal explanation.
Open governed domain →
Phase 1 pilot
Agentic systems and MCP
Tool protocols, capability boundaries, sandboxing, context degradation, coordination, deadlock, memory, injection, and evaluation traces.
A successful tool call or benchmark trajectory does not establish least authority, resistance to prompt injection, reliable coordination, or safe autonomous operation.
Open governed domain →
Phase 1 pilot
Critical supply chains
Mineral occurrence, refining, precursor chemistry, material substitution, trade concentration, controls, and data uncertainty represented by stage.
Resource abundance, mine output, refining capacity, trade share, and qualified semiconductor supply are different quantities and cannot be silently substituted.
Open governed domain →
Single-source outputs
One record, several synchronized surfaces.
canonical record
├─ public HTML page (only after gate passes)
├─ Schema.org JSON-LD
├─ domain registry entry
├─ provenance.json + SHA-256 content hash
├─ sitemap row
└─ llms.txt orientation link