Add one typed JSPACE edge with provenance event.

Add one typed JSPACE hyperedge / n-way semantic relation with provenance.

Record actual adoption of a global mutation by an agent, distinct from mere delivery/visibility.

Cold-start AGENT_BOOT_V1 with an automatic Message Board handshake. A task-only tool call observes shared-current coordination without claiming work; explicit work_key/targets/claim mode or explicit coordination_mode=AUTO crosses the start-work boundary and may establish/reuse presence. Duplicate claimed work holds.

Refresh AGENT_BOOT_V1 and its Message Board projection. Existing session coordination intent is preserved; new explicit work coordinates activate AUTO claiming, while observer sessions remain read-only unless promoted deliberately.

Create a distinct antibody variant within the parent's family with optional expiry.

Execute stored failure-antibody unittest witnesses in a restricted repository-owned subprocess; no arbitrary shell/command refs are accepted.

Record success, failure, false positive, regression pass or regression failure for a known failure antibody.

Rank matching antibody variants by semantic match, empirical repair/regression reliability, status and expiry.

Execute the newest registered T_ij for a subject using LOOKUP or the safe declarative transform DSL; compare against an independently resolved target when available and optionally persist a derived coordinate.

Compose and execute a coordinate route C_i0->...->C_ik. Closed routes automatically measure holonomy only when every edge is derivational; LOOKUP-containing loops are marked N/A rather than falsely promoted.

Transport witnessed approximation error through a caller-declared Lipschitz envelope consistent with supplied witness pairs. Geometric nearest, global-envelope, and radius-eligible transport witnesses remain distinct; certificate is conditional on the declared bound.

Assign KEEP_ACTIVE/KEEP_REFERENCE/DORMANT/QUARANTINE/PRUNE_REFERENCE while preserving lineage and downstream dependencies.

Update one regime/action posterior from an explicit observed reward; predictions never train themselves.

Select an organization/action experiment using a diagonal contextual-UCB posterior with local-regime evidence, cross-regime transfer, uncertainty, and a V3 policy prior.

Exactly solve a bounded finite-horizon POMDP with a static uncertain model index when the full joint model-state tree completes. PLAN_ONLY.

Update full-covariance Bayesian contextual state from one explicit observed reward; retains pre-update prediction for calibration.

Full-covariance Bayesian contextual prediction with empirical interval calibration. Posterior/intervals remain model-conditional.

One-step finite-belief controller combining immediate utility, expected next decision utility and information value. PLAN_ONLY.

Bayes-update a finite belief from explicit per-model likelihoods for one actual declared observation.

Build an exact finite-model/outcome belief-policy tree for bounded horizon <=3. PLAN_ONLY; no belief mutation.

Register/reset a finite discrete model belief state with normalized priors. Model belief is not canonical truth.

Read the current finite model belief distribution and entropy.

Return live substrate counts and head state for regression/telemetry.

Return an assembly/function packet. T4 preserves the supplied 15-part USER_VISUAL BOM separately from the primary-source-conditioned infection transition sequence; other machines return generic functional modules rather than fabricated native molecular subunit inventories.

Return the stable 12-archetype/14-seed BNMK catalog. V2 additive fields report 20 source-backed adapters, 20 primary sources, 15 conditioned quantitative claims, bounded expansions, operator phylogeny and optional evidence packets. User seed numbers remain distinct from verified empirical constants.

Compile one known BNMK machine into the source-backed 4D/12D/KC144 contract with primary-source witness, conditioned quantitative claims, state cycle and row associations. Unknown IDs HOLD rather than fabricating a classification.

Evaluate only caller-supplied convergence/stability witnesses: nth-term limit, ratio/root limit, contraction constant, or spectral radius. Assumption-incompatible or inconclusive values return HOLD; V2 does not change these V1 semantics.

Score a normalized producer/consumer interface profile over rate, latency, error tolerance, statefulness, reversibility and coupling. Smith-chart-inspired compatibility proxy; not physical electrical impedance.

Map a source-backed biological mechanism into a target ATHENA problem as explicit COMPUTATIONAL_ANALOGY. Primary-source support improves the mechanism model but never grants execution authority or causal equivalence.

List branch lifecycle states, optionally filtered by ACTIVE/HIBERNATED/REVIEW.

Record one witnessed calibrated branch reward observation and apply EWMA hibernation/resurrection policy.

Move a hibernated branch into REVIEW only from a verified new_evidence/new_gap/bridge_demand trigger.

Read branch lifecycle state for one branch, optionally on one metric basis.

Apply living-bridge cost accounting: build only when future routing/reuse/quality/resilience value exceeds build, maintenance and locked-capacity cost.

Persist one measured organization/resource budget observation. Token/compute dimensions are accepted only when supplied by an observable client.

Allocate tasks to workers using demand x capability fit x availability x measured efficiency subject to observable remaining budgets; unknown cost is penalized, not invented.

Summarize persistent resource observations plus automatically metered MCP tool-call wall time.

Cross-fitted AIPW estimate for binary treatment with linear outcome nuisances, logistic propensity, influence-function SE and CI. Assumption-scoped.

Estimate an assumption-scoped linear BACKDOOR, IV-Wald, or FRONTDOOR mediation effect from explicit numeric samples. Estimation never proves identification.

Search for a valid minimal back-door adjustment set in a caller-supplied causal DAG using d-separation. Identification is conditional on supplied graph/assumptions.

Perform supplied-DAG conditional identification for BACKDOOR, FRONTDOOR or INSTRUMENT criteria. Results are conditional on graph/observed-node/assumption correctness.

Leave-one-adjustment-out robustness diagnostic around V8 linear back-door estimate. Not a formal hidden-confounding bound.

Generate a heuristic observational association skeleton and candidate v-structures using marginal/one-variable partial-correlation thresholds. Hypothesis generation only.

Bootstrap the V7 heuristic association skeleton and return stable undirected/v-structure candidates. Stability is not causal probability.

Cross-fitted binary-treatment/binary-outcome TMLE with logistic nuisance fits, targeting fluctuation and influence-curve interval. Assumption-scoped.

Binary-treatment/binary-outcome TMLE with deterministic cross-fitted validation-weighted linear/quadratic logistic nuisance ensemble. Assumption-scoped.

Select a value-maximizing small finite candidate subset subject to independent-Gaussian one-sided resource chance constraints; exact only under declared finite-model assumptions.

Record a material verified challenge. Noncanonical claims return to ?/CHALLENGED; canonical claims remain #/CANONICAL_CHALLENGED and block automatic routing.

List claim authority heads, optionally filtering by Y or challenge status.

Advance exactly one authority step. ?->+ requires verified evidence; +->! witnessed execution; !-># explicit canonical authority.

Register or reuse persistent claim authority head at Y='?'. Authority is distinct from confidence, truth, consensus, pheromone and reward.

Resolve CANONICAL_CHALLENGED through explicit authorized UPHOLD or DEMOTE governance.

Read current typed authority state and witness history pointers for one claim.

Build and persist a Message Board-backed campaign/coalition proposal over multiple active NEEDs. The proposal greedily respects explicit OFFER capacity and is advisory only; it creates no presence, claims, party membership, scheduler authority, or execution authority.

Read shared-current Message Board claims and classify exact work-key/task/target collisions, fuzzy warnings, declared JOIN/REPLICA intent, and target-only partition evidence into non-authoritative treatment options. This tool never mutates claims, auto-joins, assigns work, fabricates MATA semantics, or grants execution authority.

Advisory deterministic matchmaking for one active NEED using explicit active OFFER envelopes, Message Board presence/claim state, capability fit, dependency unlock, capacity, and collision risk. It never creates a claim, assignment, party membership, or truth authority.

Publish an idempotent typed NEED/OFFER envelope through Message Board V1. The publishing agent must already have active shared-board presence. This does not assign work.

Evidence-gated matched descriptive comparison of SOLO versus PARTY mission samples under a supplied predeclared rule. Weak/ambiguous matching returns UNKNOWN. A rule pass is descriptive observed evidence only, not a causal effect claim or canonical promotion witness.

Allocate tasks by demand × capability fit × available capacity rather than equal participation.

Score a concrete group organization with explicit cost/output vectors and return-on-group-organization; exposes dense-connectivity, duplication, switching, contagion and maintenance penalties.

Evaluate collective homeostasis and emit concrete corrective actions for saturation, duplication, latency, error, staleness, contagion, reserve depletion, weak evidence, bridge overhead and coordination drag.

Select HIVE/SWARM/PACK/FLOCK/HERD/POD geometry, right-size active workers by marginal utility, preserve reserve capacity, allocate roles/topology, and return a COLLECTIVE coordinate packet.

Run evidence-sensitive quorum selection with explicit cross-inhibition, contradiction blocking, risk-dependent threshold and required winning margin; consensus alone cannot commit.

Choose FISSION/FUSE/HOLD from coordination pressure, contagion, cohesion, complementarity, duplicate work, shared dependencies and identity conflict.

Conditional object mutation. Rejects stale expected VID instead of creating a shadow mutation.

Build a depth-1 outcome-contingent action policy under the current finite belief and one supplied experiment. DESIGN_ONLY.

Read an exact half-open UTC window of already-persisted rehydration receipts/events and Message Board events. Returns raw tracked source records with Git blob, exact-byte SHA-256, canonical-record SHA-256, and deterministic trace identities. This tool is read-only and performs no assay classification, behavioral inference, causal attribution, or promotion.

Return the full chart-to-chart transform matrix, resolved coordinate subset, transform coverage, closed transform triangles and holonomy observations for an optional subject.

Rank candidate organizations using base RGO, empirical calibration, bounded learned policy, risk and budget pressure without committing topology.

Assign uncertainty-bearing intervention credit from an observed outcome delta. Weak designs remain associational; residual outcome is preserved.

Summarize persisted intervention credit and causal-confidence history.

Compile one visible output into a complete ATHENA crystal: identity/version/manifestation, exact lexeme coordinates, KC144, JSPACE, SCALE, lineage/time/liminal coordinates, open-world polyatlas, math objects, graph/hypergraph deltas and RETURN.

Advance a CYCLE.1 through as many real internal phases as possible, merging newly supplied witnessed inputs. Semantic boundaries stop with typed WAITING_* state instead of simulating measurements, execution or tests.

List recent CYCLE.1 runs and typed ACTIVE/WAITING/COMPLETE status without expanding full state.

Verify the stored cycle state digest and deterministic replay/integrity receipts of persisted RAG/HUG-packet/GAP/FIELD/AOR/transport child runs. External semantic execution/test truth is not re-simulated.

Start a persisted ATHENA.CYCLE.1 whole-organism run from a task seed and explicit phase configuration. Starting creates state only; it does not claim later semantic phases executed.

Fetch one persisted CYCLE.1 state, phase artifacts, wait requirements, public phase-event receipts and RETURN state.

Rank finite candidate experiments by expected improvement in downstream decision utility under the current belief. DESIGN_ONLY.

Estimate expected value of perfect information by Monte Carlo under the declared Gaussian linear belief and utility model.

Estimate expected value of sample information for finite Gaussian linear experiment designs. DESIGN_ONLY.

Persist one confidence-weighted delayed action credit with explicit temporal discount; delay alone is not causation.

Summarize delayed confidence-weighted credit by action.

Resolve an OID/CID/name/CRYS identifier through identity, version lineage, manifestations, all coordinate charts, JSPACE edges/hyperedges, math objects, crystal versions and RETURN routes.

Propagate a failure/change alarm only over explicit influence/dependency edges with weight, hop decay, threshold and bounded radius.

Compile a PLAN_ONLY source-bound, digest-pinned, isolated-canary, CAS single-writer cutover. It never deploys or grants cutover authority.

Evaluate supplied external canary observations. Missing metrics HOLD; thin sample/window evidence ROLLBACK; failed health/error/latency/restart gates ROLLBACK.

Evaluate a supplied single-writer quiescence observation against exact current-image and state-snapshot coordinates. PASS validates supplied evidence only; it does not stop a writer, install a fence, verify a snapshot, or contact production.

Bind a PLAN_ONLY activation plan, checksum-valid isolated canary witness, supplied quiescence observation, and authority reference into ATHENA.CUTOVER.HOLD.1. The result cannot deploy, resolve secrets, mutate state, stop/start a writer, or activate traffic.

Return ATHENA.DEPLOYMENT.2: exact OCI identity, source-head binding, single-writer persistence, secure HTTP host, canary, CAS cutover, rollback, and receipt laws. Surface availability is not activation evidence.

Fail-closed validation of ATHENA.DEPLOYMENT.BUNDLE.2 intent without contacting infrastructure, resolving secrets, or moving traffic.

Replay an ATHENA.CUTOVER.HOLD.1 packet against exact expected bindings. PASS verifies supplied packet bytes only and is not current cluster observation, authority verification, cutover, or an activation receipt.

Replay the exact plan/image/source/state bindings of a supplied activation receipt. PASS verifies receipt structure, not independent cluster observation.

Return learned/shrunk token-artifact-module-domain-system diffusion coefficients with reliability and causal-weight metadata.

Record observed utility of pheromone transfer between two scales and update a shrinkage-learned diffusion coefficient.

Register a V6 science-shadow claim record without mutating Y1 canonical claim authority.

Summarize independent V6 science-shadow witness groups. Evidential metadata only; no Y1 canonical rewrite.

Attach a SUPPORTS/FALSIFIES/INCONCLUSIVE science-shadow witness with an explicit independence key. This never promotes Y1 authority.

Select a finite candidate subset under correlated Gaussian resource covariance plus ellipsoidal mean ambiguity; exact enumeration only below the declared threshold.

Bounded proxy dual-control planner combining control reward, transition-parameter information value and predictive risk. PLAN_ONLY; execute first action then observe/replan.

Value deterministic two-timepoint treatment policies through the bounded longitudinal parametric g-formula. Assumption-scoped and PLAN_ONLY.

Update longitudinal evidence-backed authority for an entity from reuse/prediction/repair/regression/generalization outcomes.

Rank evidence-backed elder/cultural authority within a scope; age alone confers no authority.

Emit public liminal telemetry for an agent task; private chain-of-thought is neither requested nor stored.

Record verified EQUIVALENT or DISTINCT relation for an unordered pair in a declared dedup context. EQUIVALENT requires witnessed sameness across every preservation dimension; contradictory heads become CONFLICT.

Resolve pair CONFLICT by explicitly authorizing one already-witnessed side; conflicting history remains in the event ledger.

Construct contradiction-aware equivalence components over supplied candidates. Only witnessed contradiction-free equivalence components are collapse-safe; UNKNOWN/conflict preserves identities.

Read the active witnessed relation/conflict head for one unordered pair in a dedup context.

Fit a scoped logistic evidence-dependence model from explicit labelled examples.

Return a Laplace/Hessian logit interval around a fitted V10 evidence-dependence probability. Model-conditional diagnostic.

Record one externally labelled evidence-dependence example for later empirical calibration.

Predict pairwise evidence dependence under a previously fitted scoped calibration model.

Apply a caller-declared logistic metadata-dependence model to science-shadow witness pairs. Diagnostic, not formal independence inference.

Compute metadata-similarity effective-N and spectral participation-ratio diversity for a science-shadow claim's witnesses.

Rank binary-outcome experiments by expected information gain under supplied hypothesis priors/likelihoods, cost, risk, feasibility and ethics; returns DESIGN_ONLY.

Generate candidate experiments from caller-declared factor levels and hypothesis factor-effect models, then rank them by V5 expected information gain. DESIGN_ONLY.

Complete one PLANNED extraction task with one or more actual outputs and a verified witness. Each output becomes an addressable EXTRES result; completion never implies those outputs are mutually equivalent.

Use one witnessed EXTRES payload as a next-generation seed, bounded by the parent EXTRUN depth and task-per-generation limits. Expansion creates work contracts, not semantic results.

Record a witnessed extraction-task failure with explicit reason. Failure creates evidence; it does not fabricate output.

Return only currently PLANNED tasks for an extraction run, ordered by depth/ordinal/id.

Create a bounded AOR SX.1 extraction run. Planning creates typed PLANNED work contracts only; it never fabricates semantic transform results.

Fetch one witnessed extraction result payload and provenance.

Fetch one extraction run and all task heads for replay/navigation.

Fetch one extraction task with its seed, transform contract, status and result refs.

Match a new failure/event against the durable antibody registry and return reusable repairs and regression witnesses; optionally records successful antibody hits.

Persist a diagnosed failure as a reusable signature, trigger, detector, repair, evidence bundle and regression/replay references.

Build a bounded FCI-inspired partial ancestral candidate using all observed-variable conditioning subsets up to a declared order, collider orientation and limited propagation. Not full FCI/RFCI.

Assemble actual SX/RAG/Y/GAP/HUG/branch/AOR residuals into typed FIELD.1 action candidates. Generated proposals are UNMEASURED; exact same action signatures may merge provenance only; conflicting explicit ranking/routing data fail closed to CONFLICT and disputed operands are removed. Semantic similarity never merges.

Fetch one persisted FIELDRUN with frozen module inputs, candidates, metric states, provenance edges and digest.

List recent persisted FIELDRUN receipts without expanding full candidate payloads.

Rebuild one FIELDRUN from frozen module inputs and compare candidate IDs, conflict state, provenance edges and field digest.

Mandatory emission gateway for Athena-aware clients: crystallize the exact body, derive a header, assemble HEADER+BODY, create a third emission MID over the exact visible bytes, coordinate-index every visible lexeme, and return an ENV envelope.

Represent n-way coordination as a lazy (n-1)-simplex without materializing its exponential face lattice.

Execute the bounded fixed-path claim create-if-absent plus separate CLAIM_ACQUIRED event append only from fresh EVENT_READY state; preserve lost races and unjournaled effects.

No-write preparation of a bounded fixed-path CLAIM_V1 from exact EVENT_READY state under fresh H/P/F/C/W. Inferred PENDING readiness returns a reducer-validated NODE_READY prerequisite instead of manufacturing claim authority.

Observe an existing fixed-path provider claim and append only its missing CLAIM_ACQUIRED event when fresh replay remains READY; never create or delete the claim.

Compare exact shared source head, frontier digest and prompt-stack digest; any material coordinate change requires rehydration.

Reduce exact Git-resident SCHED V3 objective/run state into FRONTIER_PACKET_V1 and bind it to the current prompt-stack digest without conflating the two.

Report the fresh bounded GitHub claim-provider capability and exact five-blob SCHED claim/journal contract without exposing credentials or executing a write.

Persist only the reducer-validated fixed-sequence NODE_READY prerequisite for one dependency-ready PENDING node, then rehydrate and require explicit READY.

Return the Pareto frontier over replayably READY SCHED V3 nodes only. Missing event coverage or contract drift fails closed and never manufactures readiness.

Compile and optionally persist GAP.1 = explicit target nodes minus witnessed directed reachability closure over frozen source groups and typed graph edges. This is navigation closure, not logical/causal entailment. Unknown residual metrics route to measurement.

Fetch one persisted GAPRUN with exact source/edge/target/policy snapshot, closure paths, residuals, grow decision and digest.

List recent persisted GAPRUN receipts without expanding full graph snapshots.

Recompile one GAPRUN from frozen inputs and compare closure nodes, gap, grow and decision digest.

Bayesian linear update from one explicit observed target and complete feature vector.

Register/reset a finite-dimensional Gaussian linear parameter belief in natural form. Not a GP/neural/nonparametric posterior.

Read Gaussian linear belief mean/covariance and observation count.

Return configured canonical Git brain root, branch, head and dirty state.

Return seven deterministic leaderboard views over P3+ scoped verified Impossible Challenge completions. Ranks, titles and scores are game-state metadata and never evidence or authority.

Estimate decision value of candidate GP measurements while updating both finite-grid kernel weights and within-kernel posterior action means hypothetically. DESIGN_ONLY.

Bayesian-model-average a GP prediction across the finite hyperparameter posterior, separating within-model and between-model variance.

Estimate downstream decision value of candidate GP measurements by conditional-Gaussian Monte Carlo. DESIGN_ONLY.

Predict with a deterministic inducing-point FITC GP approximation and expose query-level error against the exact current bounded GP reference.

Grid-search fixed RBF GP hyperparameters by exact marginal likelihood; optional CAS application to current GP model.

Compute a normalized posterior over a finite explicit GP hyperparameter candidate grid using marginal likelihood and visible priors. Read-only.

Approximate a continuous log-uniform GP hyperparameter posterior over a declared positive box using deterministic Halton quasi-Monte Carlo particles. Read-only.

Rank candidate GP measurements by downstream decision EVSI plus expected hypermodel entropy reduction under the QMC hyperposterior. DESIGN_ONLY.

Append one actual observed target to the bounded fixed-kernel GP dataset.

Return exact fixed-hyperparameter RBF GP posterior mean/variance for one query. Prediction never self-trains.

Register/reset an exact small-data Gaussian-process regressor with fixed RBF kernel hyperparameters. Model state only.

Choose the shallowest tested FITC/full-GP representation that preserves the exact current decision on the supplied action/query set under a declared decision-margin error rule.

Approximate the current GP with a deterministic farthest-point subset of observed rows and compare the result to the exact bounded GP reference.

Read fixed-kernel GP hyperparameters and observed-point count.

Compute a shortest directed JSPACE path between two nodes, optionally restricted to typed relations.

Return the chronological append-only P5 Hall of Immortals after a fresh shared frontier sync.

Complete a PLANNED HUGINV with output validated against frozen output schema and a verified execution receipt.

Record a witnessed executor failure for PLANNED HUGINV without fabricating output.

Fetch one HUGINV including frozen implementation snapshot, six arguments, context, status, output/receipt or failure.

List registered HUG implementations, optionally filtered by CANDIDATE/TESTED/CANONICAL. Empty list means no semantic implementation is registered; do not simulate one.

Create HUGINV execution packet for registered implementation. Requires exactly io,au,fx,lm,er,st, validates registered input schema, freezes implementation snapshot/context/input digest and returns PLANNED only. It does not execute or invent QHUG semantics.

Advance HUG implementation exactly one maturity step. CANDIDATE->TESTED requires witnessed test; TESTED->CANONICAL explicit authorized ref; skips fail.

Register an exact HUG implementation identity. All six parameter meanings io,au,fx,lm,er,st plus algorithm ref, implementation digest and schemas are mandatory. Registration creates CANDIDATE only; it does not assert semantic QHUG is tested/canonical.

Read one registered HUG implementation including exact six-parameter semantics, schemas, maturity and witness routes.

Recompute HUGINV packet digest. Verifies frozen implementation/input integrity only; semantic replay is N/A unless a real registered executor replays the algorithm.

Hydrate current canonical organism state: head, registry, recent causal history, JSPACE summary, SCALE and pending global mutations.

Record a demonstrated/replayed Impossible Challenge completion with an ΩA achievement coordinate, witnesses, score inputs, and verified cleanup. COMPLETED is refused unless cleanup is VERIFIED and unknown_residue is zero. The receipt is game/provenance state, never canonical truth or global XP authority.

Open a Git-shared Impossible Challenge on the existing Message Board frontier. The opener must already be present on Message Board V1. This records a scoped hard barrier and success contract; it grants no execution authority, XP, title, or proof standing.

Read one Impossible Challenge and its completions after a fresh shared Message Board sync.

Upgrade a completion through independent P3, adversarial P4, or crystallized/reused P5 verification. P3+ requires a different active verifier; P4 requires at least five attack witnesses; P5 requires generalization plus downstream reuse. P3 mints append-only Immortal title records for credited agents.

Persist a text manifestation and assign exact KC144/OID/VID/MID/paragraph/sentence/token/character coordinates to every non-whitespace lexeme.

Estimate main and pairwise interaction contrasts from supplied intervention/outcome observations. Missing 2x2 cells remain UNIDENTIFIED.

Compute order-2..4 factorial inclusion-exclusion contrasts. Every 2^k cell is required; missing cells remain UNIDENTIFIED.

Compose a bounded finite joint science-twin belief across 2..5 caller-declared factor axes, compatibility multipliers and optional state likelihoods. Read-only; not a universal joint posterior.

Rank unique linear actions under a declared multivariate Gaussian belief using exact moments, lower-tail Normal CVaR, cost and Pareto preservation. Unknown coefficient coordinates fail closed. PLAN_ONLY.

Perform the exact finite-dimensional multivariate-Gaussian posterior update for one declared linear-Gaussian observation. Unknown coefficient coordinates and non-finite state fail closed. Model state only; not general continuous Bayes.

Rank policies over a bounded weighted joint-state ensemble using expected utility, lower-tail CVaR, worst case, expected/max regret and Pareto preservation. PLAN_ONLY.

Compute exact finite decision EVI plus entropy reduction over supplied weighted joint science-twin states and declared outcome likelihoods. DESIGN_ONLY.

Compile typed JSPACE edges into an invalidation graph and propagate a bounded decaying alarm. Known dependency relations reverse direction so dependencies invalidate their dependents; unknown relations are ignored unless explicitly mapped.

Project a supplied causal DAG with explicit latent nodes into a restricted observed ADMG using latent-only internal directed paths/common latent ancestors. Not data discovery.

Estimate static two-timepoint treatment-regime risks with a transparent parametric g-formula over one binary intermediate variable. Assumption-scoped.

Estimate static two-timepoint binary treatment-regime risks with sequential logistic targeting. Assumption-scoped and not a general longitudinal-TMLE theorem.

Cross-fit the bounded binary two-timepoint sequential logistic TMLE across held-out folds with explicit baseline/time-order validation. Assumption-scoped; not a general longitudinal TMLE theorem.

Match an agent's latest needs/blockers against peers' latest offers/capability deltas.

Return the canonical unified MAXDEV execution law covering RECON/Omega, memory firewall, SX/RAG/HUG/GAP/FIELD, measurement/Y/AOR/Collective, execution/test/learning, CAS/replay/self-test/promotion and continuation.

CGR: route through caller-supplied typed edges without upgrading standing or claiming a causal path.

ESCPF: split claims by standing and evaluate explicit promotion requests; unsupported OBSERVED/EMPIRICAL/HISTORICAL_PRIMARY and high-stakes symbolic use HOLD.

MMTB: explicit lossy field transport between scoped models; retain residue/provenance and never assert cultural identity/equivalence.

OSD: deterministically select/decode a caller-supplied codebook from explicit sample/seed; symbolic-only, no factual prediction or high-stakes authority.

RSM: simulate B->Theta->Pi workflow gating; missing boundary/phase and declared hazardous classes HOLD; simulation grants no execution authority.

Evaluate one layer-to-layer MCK transport. Cross-layer equivalence/promotion/authority/hazard shortcuts HOLD; an explicit source-bearing bridge may allow only bounded non-identity transport with declared invariants and loss.

SAC: select only from a caller-supplied symbolic address space; preserve provenance/loss and HOLD rather than invent a match.

Shared Git message board for inter-agent presence, work claims, duplicate-work prevention, collaboration, heartbeats, messages/acknowledgements, and release/handoff. Read before expensive shared work; use present to claim a lane, join to collaborate on an existing lane, or REPLICA only for deliberate independent replication. Writes always require a freshly verified shared remote frontier.

Receding-horizon risk-adjusted planning over caller-supplied actions using the learned multivariate transition surface. PLAN_ONLY and never self-training.

Solve a bounded finite-horizon robust dynamic program under state-action rectangular total-variation ambiguity around supplied transition distributions. Unknown state coordinates/non-finite transitions fail closed. PLAN_ONLY.

Update the degree-2 Bayesian model from an explicit observed reward and update the raw-feature OOD reference distribution.

Degree-2 polynomial-basis Bayesian prediction with full-covariance V5 posterior and OOD-dependent interval inflation. Not GP/neural universal inference.

Project current accessible ATHENA runtime state into one Ω packet spanning semantic head, Git, base/crystal, Collective, branch/Y, AOR, development, cycle, promotion and migration state. Unavailable components remain UNKNOWN.

Update the empirical context distribution used to detect distribution shift for a regime. Observations only; no semantic authority.

Score a context against the empirical regime distribution using ridge-regularized Mahalanobis geometry plus unseen-feature pressure.

Return OPERATIONAL_BASIS_V1 derived from the actually registered current prompt/control-plane MCP surface. Descriptors classify semantic capability, effect, authority, freshness dependencies, preconditions, replay, and rollback/compensation without widening execution authority.

Compile and optionally persist AOR.3 decision state. UNKNOWN never becomes zero; dependencies, typed Y authority, evidence, claimed test/persistence and calibration gates are fail-closed; scalar NEXT coexists with Pareto alternatives and explicit budget allocation.

Fetch one persisted AORRUN with frozen input/output, branch lifecycle, authority snapshots and decision digest.

List recent persisted AORRUN receipts.

Recompile one persisted AORRUN from frozen inputs and verify authority snapshot, routing digest, NEXT, GROW and Pareto state.

Compute successor rank sensitivity for one persisted orchestration run.

Build a bounded PAG-like circle/arrow/tail candidate from observed Gaussian conditional independences and conservative collider propagation. Not full FCI/RFCI.

Identify the interval-possible Pareto frontier and choose an uncertainty-rich frontier candidate for experimentation. EXPERIMENT_SELECTION_ONLY.

Compute the exact non-dominated frontier for the supplied candidate set, optionally using interval-robust dominance and min/max metric directions.

Form a Git-shared multi-goal party on top of Message Board V1. The leader must already be actively present on the shared board. Formation emits a durable party event but earns no XP.

Join an existing party while already present on Message Board V1. Goal ownership, role, capabilities, task relation, and current board claim are frozen into the shared party record.

List Git-shared parties after synchronizing the Message Board frontier.

Post a party-scoped coordination message through canonical Message Board V1. The message is tagged with party_id and explicit party goal refs; only acknowledged party-tagged messages in the current reward window can contribute to party XP eligibility. Posting itself earns zero XP.

Record a witnessed multi-goal party outcome and calculate a small receipt-gated coordination XP bonus candidate. Party Reward Provenance V3 requires current frozen claims, ACKed typed result events, and a globally unused source_xp_ref before an award; legacy calls missing those coordinates remain parseable but HOLD. The tool never mutates global XP authority.

Post an attributable party RESULT or VERIFY provenance event through canonical Message Board V1. The sender must still hold the exact board claim frozen into party membership, the goal must be assigned to that member, and recipients must be current party members. Posting and acknowledgement earn zero XP; the event is provenance only and does not prove result truth.

Read one party after a fresh shared-board sync, including members, current board presence, acknowledged communication evidence, Big-3 score, and reward receipts.

Run bounded Gaussian PC-stable conditional-independence search with Fisher-z tests and limited Meek orientation. Hypothesis graph only.

Return globally visible mutations this agent has not adopted.

Reinforce declared multiscale coordinates using learned diffusion coefficients rather than fixed scale attenuation.

Read the durable stigmergic priority field for one route or the highest-priority routes.

Return the persistent multiscale pheromone field grouped by token/artifact/module/domain/system scale.

Reinforce a source coordinate and declared token/artifact/module/domain/system relatives with scale-distance attenuation.

Persistently reinforce or evaporate one artifact/routing pheromone using evidence, reuse, novelty, staleness and contradiction observations.

Rollback one organization-policy transaction as a new versioned transaction; history is preserved.

Score a normalized feature vector with the current bounded learned organization policy.

Return the versioned bounded organization-policy state and empirical reliability for a scope.

Versioned bounded online policy update from an observed reward. Rejects stale expected_version and records rollback history.

Exhaustively solve one small finite-state finite-horizon POMDP tree when node search completes; otherwise return no certificate.

Build an empirical prequential residual band from retained pre-update V5 observations and inflate it under current OOD pressure. Not a distribution-free conformal guarantee.

Apply the explicit semantic inverse for active JSPACE edges created by one topology projection, under semantic-head CAS. Git compensation remains separately surfaced.

Prepare and journal a topology-to-JSPACE projection plan against explicit topology/semantic/Git heads without mutating JSPACE.

Read one topology-to-JSPACE projection saga and its recovery state.

Register a discovered mutation; global classes require adoption by all active agents next cycle.

Evaluate and optionally persist PROMOTION.2 exact-head readiness. Caller-supplied CI/smoke packets may produce ATTESTED_READY but can never mint QUALIFIED; QUALIFIED requires trusted external verification that is not exposed in this caller-attestation schema.

Fetch one persisted versioned PROMRUN with frozen local certificates, caller attestations, trust state, status and decision digest. Historical PROMOTION.1 receipts remain readable.

List recent persisted unified runtime promotion receipts across historical PROMOTION.1 and current PROMOTION.2 semantics.

Replay a stored PROMOTION.1 or PROMOTION.2 predicate from its frozen inputs using the matching historical evaluator version. Replay checks receipt determinism; it does not independently re-query the external provider.

Independently verify the target Git head against the host-configured GitHub repository/check-suite and then evaluate PROMOTION.2 with the internally produced trusted verifier receipt. Caller cannot choose repository, API host, trusted app, required checks, run binding, or verifier packet.

CAS-activate a TESTED prompt candidate as a scope-limited overlay after an observed PASS witness.

Deterministically compile policy, selected modules, and applicable scoped overlays from the configured Git brain.

Persist a baseline/candidate prompt experiment record; unexecuted designs cannot be recorded as observed PASS.

Compare an agent's prompt/Git ancestry with the current brain and require rehydration on material shared changes.

Hydrate the exact current Git prompt runtime and deterministically compile the task/profile addendum with ancestry/digest.

CAS-promote a tested candidate into its canonical Git prompt module only with regression, adversarial, replay and evidence gates.

CAS-write a versioned prompt candidate plus proposal event. Proposal never activates or promotes itself.

Fresh-fetch then publish an exact clean local prompt-brain HEAD by ordinary non-force push only when the remote branch is its ancestor; post-push fresh-fetch verifies equality.

Compare the local prompt-brain checkout with its configured remote branch; optionally fetch before comparison. Never mutates local history.

Fresh-fetch the shared prompt-brain remote and fast-forward a clean current branch only when local HEAD is an ancestor. Failed fetch, ahead, diverged or dirty states hold.

Verify a supplied tree decomposition against QHUG conflict/dependency factor scopes; certify factor coverage, running intersection, width upper bound, and an exact treewidth value when a matching clique lower bound is present.

Analyze a Boolean QHUG patch kernel into exact primal-graph components and structural-free coordinates. No optimization is invented.

Exactly solve a supported disconnected Boolean QHUG kernel by local component enumeration plus Pareto-pruned Minkowski convolution; preserves every scalar-policy tie.

Fetch one frozen RECONRUN with exact consulted/expected source sets, Ω snapshot, defects and digest.

List recent reconstruction receipts without expanding full Ω snapshots.

Recompute one RECONRUN digest from its frozen source contract and Ω snapshot. This verifies receipt integrity, not freshness of current state.

Persist a RECONRUN freezing current Ω plus the exact source refs actually consulted and optional expected refs. Missing expected refs become explicit defects; unlisted/unavailable sources are never implied searched.

Persist a measured closed coordinate-loop observation H_gamma with start, returned state, defect, metric and provenance event.

Update a learned task-regime centroid from observable signals and measured reward. Learned geometry is routing context, not semantic identity.

Return nearest evidence-weighted learned regime centroids plus the coarse V4 regime.

Resolve normalized task signals into a deterministic HIVE/SWARM/PACK/FLOCK/HERD/POD task-regime key for hierarchical learning.

Create or reuse one canonical capability identity from a functional signature.

Register a typed coordinate transform T_ij:C_i->C_j. LOOKUP is navigational only; derivational modes require a safe declarative program and explicit comparison metric.

After an agent completes the current self-prompt and commits substantive work, verify exact checkpoint ancestry and observed pass receipts, persist the cycle, rehydrate the new Git head, and compile the next self-prompt.

Fresh-sync the shared Git branch, then derive a read-only, content-addressed handoff delta describing what a successor agent must rehydrate. Distinct from successor task routing.

Fresh-sync and consume an exact handoff delta, reject drift, and hydrate only the affected dependency cone. Falls back to full rehydration when delta coverage is insufficient.

Fresh-sync the shared Git branch, then list current Git-persisted rehydration loops and their status, step, state/chain digests, and checkpoint heads. Unverified shared state holds instead of returning stale inventory.

Fresh-sync the shared Git branch, then resume the persisted rehydration loop at its exact current prompt/state/chain coordinates. Clean behind checkouts fast-forward; dirty, ahead, diverged, or unverified shared state holds rather than returning a stale local handoff.

Start a Git-persisted explicit rehydration loop, compile the first bounded self-prompt, and checkpoint exact prompt/frontier/state coordinates. This creates no background execution.

Fresh-sync shared Git, verify the exact current loop state digest, apply the same witnessed terminal-closure gate used by advance, then compile a replayable routing-only successor baton. Stale or unverified shared state fails closed by default; ties remain preserved.

Fresh-sync the shared Git branch, then replay and verify the current persisted loop chain, prompt/receipt digests, sequential steps, and Git ancestry. PASS is shared-current causal-integrity evidence, not world truth.

Rank proposed REPLICATION or FALSIFIER designs by expected power, metadata novelty, feasibility, cost and risk. DESIGN_ONLY.

Estimate effective evidential replication/falsification diversity from witness confidence, independence keys and declared dataset/implementation/method/operator/environment/seed metadata.

Resolve a canonical OID/CID/name and return exact navigation, versions, graph edges and KC144 station.

Compile and optionally persist a RAG.1 decision over supplied provenance candidates. Present measurements must be witnessed 0..1 packets; missing required measurements become UNKNOWN and measurement_plan. Selection is coverage-aware, budgeted, EQ1-safe and replayable. This tool does not claim it searched sources it was not given.

Fetch one persisted RAGRUN receipt with exact query/candidate/EQ input, selected sources, coverage, solver/optimality and decision digest.

List recent persisted RAGRUN receipts without expanding full candidate payloads.

Recompile one stored RAGRUN from exact frozen inputs and compare decision digest, selected sources, coverage and equivalence report.

Calibrate a new predicted RGO against accumulated downstream observations with reliability shrinkage toward the identity predictor.

Record predicted versus observed Return-on-Group-Organization and update the persistent online calibration witness.

Simulate multi-step trajectories through the learned transition model with uncertainty-banded discounted return. Always SIMULATE_ONLY.

Run uncertainty-banded multi-step organization rollouts using explicit context transitions only. Always simulate-only; never commits topology.

Return the live ATHENA.RUNTIME.UNIFIED.1 architecture manifest derived from the running Server, schema/startup state, Collective/AOR organs, unresolved semantics and exact Git status.

Evaluate caller-supplied finite action sequences on bounded three-branch moment scenario trees with expected return and lower-tail CVaR. SIMULATE_ONLY.

Exhaustively enumerate small finite schedules under dependencies, worker capability, horizon and explicit budgets. Returns an exact certificate only if search completes; otherwise degrades explicitly.

Bounded beam-search finite-horizon scheduler over task dependencies, worker capacity and observable resource budgets; does not claim global optimality.

Apply the next explicit additive schema migration receipt. Version 1 inventories/pins the already-created modular schema; it does not destructively rewrite organ tables.

Return the non-destructive migration plan from the current schema-ledger version to this runtime target. A future/newer DB schema blocks silent downgrade.

Return the runtime DB schema-ledger version, component versions and current additive SQLite schema fingerprint.

Verify schema-ledger target version plus presence of critical unified-organ tables. Additive schema drift is reported separately from corruption.

Search canonical registry by semantic/canonical text.

Run the read-only ATHENA.SELFTEST.1 local health synthesis: SURFACE.2, COMPOSITION.2, schema migration health, Ω projection and deterministic replay samples. PASS does not substitute for external CI/smoke attestations or semantic proof.

Compute the standard risk-ratio E-value sensitivity metric for a point estimate and optional closest-to-null CI limit.

Compute a two-dimensional risk-ratio bias-factor sensitivity surface across declared exposure-confounder and confounder-outcome strengths.

Estimate deterministic two-timepoint dynamic-policy value with out-of-fold sequential AIPW nuisance predictions. A1 policies may use baseline only; A2 policies may use baseline+A1+L1 only. PLAN_ONLY and assumption-scoped.

Estimate deterministic two-timepoint dynamic-policy value with sequential AIPW augmentation and explicit history preservation. Assumption-scoped; not a general longitudinal causal theorem.

Close a session with public summary telemetry and optionally checkpoint it into the configured canonical Git brain using expected-head CAS.

Begin an agent session against the exact current semantic and Git heads; returns hydration state.

Return ATHENA.STARTUP.1 local readiness: mature surface, composition, schema currency and optional SELFTEST replay health. Reads remain allowed while degraded; this tool does not silently change mutation semantics.

Fit a ridge state-dependent multivariate transition-delta regression from observed before/after action rows and expose predictive covariance plus parameter-information leverage.

Update artifact/routing priority through evidence/reuse/downstream reinforcement plus age, staleness and contradiction evaporation.

Bootstrap the bounded V13 FCI-lite discovery surface and return procedural graph-variant/marked-edge stability. Read-only; bootstrap frequency is not a causal posterior.

Calibrate structural support against externally labelled correctness with weighted out-of-fold isotonic reliability. Duplicate support coordinates are pooled before PAV. Read-only; calibrated reliability is not a causal graph posterior.

Convert V8 bootstrap association stability into an uncertainty-preserving o-o partial graph. Not FCI/PAG/CPDAG theorem.

Audit the promoted unified runtime against the complete mature tool/resource contract and executable composition certificate. Missing mature surfaces or organ/probe failures return FAIL.

Transactionally apply INIT/REPLACE/FISSION/FUSE/PATCH_MODULE to a collective topology only when expected_version matches current version; records reversible before/after witnesses.

Read one versioned collective-control topology and its current CAS version.

Apply a prepared topology projection to canonical JSPACE after topology+semantic CAS preflight, optionally checkpointing Git by expected-head CAS. This is a recoverable saga, not an atomic cross-store transaction.

Rollback a prior collective topology transaction under current-version CAS while preserving rollback itself as a new witnessed transaction.

Return a multivariate empirical action-conditioned transition mean/covariance from V5 observed transition rows with shrinkage toward no change.

Record one observed context transition for an organizational action and update shrinkage transition statistics.

Predict context deltas/uncertainty from observed action transitions. Missing features are not synthesized.

Convert explicit alarmed nodes into GAP target/revalidation pressure while preserving alarm provenance. Alarm transport is not proof of falsity, causal dependence or logical entailment; missing residual metrics remain for measurement.

Translate explicit failure-antibody matches into UNMEASURED FIELD repair candidates. A match never auto-executes or proves the repair will work; measurement and witnessed testing remain required.

Translate a persisted AORRUN successor frontier into Collective task identities/capability requirements. Worker-allocation metrics are never invented; candidates lacking explicit collective_metrics remain UNMEASURED and are not allocation-ready.

Carry explicit bridge economics from an AOR bridge candidate into the Collective bridge-accounting packet. Economics are never derived from AOR bridge score.

Fetch one persisted TRANSPORTRUN including frozen input/output boundary packet and digest.

Freeze current Collective-V2 pheromone state for named routes into AOR/RAG attention-prior packets. This adapter never populates relevance, source_authority, evidence, confidence or Y authority.

List recent AOR×Collective transport receipts.

Replay one persisted typed transport from its frozen inputs and verify transport digest. Replay proves adapter determinism, not semantic truth of the source organ.

Carry a witnessed observed RGO outcome into a reward-observation packet without automatically changing AOR DeltaJ/evidence/Y authority. If DeltaJ names the same outcome, a double-count guard is emitted.

Solve a bounded finite two-stage resource problem: choose first-stage candidates, then best feasible one-option recourse per declared scenario. Exact enumeration only below the finite threshold.

Return empirical coverage/error/width and a reliability-shrunk sigma correction for Bayesian intervals.

Return a model-conditional proxy decomposition into aleatoric, parameter-epistemic, distribution-shift and calibration-error components. Diagnostic only.

Verify an ENV emission by recomputing the exact visible-byte SHA-256; detects any post-crystallization mutation of the user-visible output.

Execute a repository unittest only when bubblewrap namespace isolation is available; otherwise fail closed with HERMETIC_UNAVAILABLE. Never silently falls back.

Execute one repository-owned unittest witness with isolated Python mode, sanitized environment, timeout, network-socket monkeypatch and POSIX resource caps when available. Not claimed OS-hermetic.

Persist measured per-worker resource use, normalized budget pressure, and useful-output efficiency for future scheduling.