mirai-graph 1.4.0 → 2.1.0

package/publications/observable-self-correcting-ai-workflows/README.md

@@ -1,105 +0,0 @@
-# Observable Self-Correcting AI Workflows
-
-Status: article foundation
-
-## Working Title
-
-Observable Self-Correcting AI Workflows with Dynamic Episode Traces and
-Technology Quality Feedback
-
-## Core Question
-
-Can a graph-governed workflow make AI-assisted work auditable and
-self-improving by recording what actually happened, checking whether the
-declared technology was followed and routing findings into correction,
-replay/regression or Kaizen?
-
-## Proposed Contribution
-
-The article introduces an observable correction loop:
-
-```text
-process contract
--> dynamic episode trace
--> technology quality feedback
--> finding classification
--> correction route
--> Kaizen / replay / regression candidate
-```
-
-The loop separates planned technology from actual behavior. It records why a
-path was selected, why alternatives were blocked, which gates were evaluated,
-what result was produced and how findings should be handled.
-
-## Hypotheses
-
-- H1. Dynamic episode traces improve auditability by preserving selected paths,
-  blocked alternatives, gates, evidence and limitations.
-- H2. Technology quality feedback identifies process violations that ordinary
-  test results can miss.
-- H3. Classified findings with correction routes improve recurrence prevention
-  by feeding Kaizen and replay/regression candidates.
-
-## Public Artifacts
-
-- `standard/dynamic-episode-layer.md`
-- `standard/episode-trace.md`
-- `standard/dynamic-layer-code-generation.md`
-- `schemas/dynamic-episode-trace.schema.json`
-- `examples/dynamic-episode-minimal/`
-- `standard/technology-quality-feedback.md`
-- `schemas/technology-quality-feedback.schema.json`
-- `examples/technology-quality-feedback-minimal/`
-- `standard/process-control-contract.md`
-- `releases/1.0.0-rc.4.md`
-- `releases/1.0.0-rc.6.md`
-
-## Experiment Design
-
-Compare ordinary AI-assisted execution with observable graph-controlled
-execution.
-
-Baseline:
-
-```text
-task -> work output -> tests/review
-```
-
-Graph-controlled condition:
-
-```text
-task
--> process contract
--> episode trace
--> selected and blocked paths
--> technology feedback report
--> classified findings
--> correction route / Kaizen / replay candidate
-```
-
-## Candidate Metrics
-
-- episode trace completeness;
-- blocked-alternative coverage;
-- technology-step adherence;
-- unclassified finding count;
-- tests-passed-but-process-violated count;
-- correction-route coverage;
-- repeat-failure rate after Kaizen;
-- replay/regression pass rate;
-- reviewer effort to reconstruct what happened.
-
-## Required Evidence Before Publication
-
-- Sanitized private or public pilot traces.
-- At least one replay/regression run over corrected workflow behavior.
-- Reviewer packet showing whether an external reviewer can reconstruct why the
-  AI acted as it did.
-- Threats-to-validity section covering task difficulty, model variance,
-  selection bias, retention policy and trace privacy.
-
-## Limitations
-
-Current public artifacts validate trace and feedback shape. They do not prove
-that production AI work becomes safe, autonomous or objectively better without
-pilot evidence and independent review.

package/publications/preprint-note.md

@@ -1,11 +0,0 @@
-# Preprint Note
-
-Status: placeholder
-
-Mirai Graph is related to the Evolutionary Object Graph research program.
-
-Public preprint records, DOI links and citation notes should be added here only
-after the public venue page and metadata are confirmed.
-
-Do not describe a preprint as peer reviewed unless the venue record explicitly
-states peer-reviewed publication status.

package/publications/preprint-outline.md

@@ -1,45 +0,0 @@
-# Mirai Graph Preprint Outline
-
-Status: draft outline
-
-## Working Title
-
-Mirai Graph: An Executable Evolutionary Graph Standard For Governed Human-AI Work
-
-## Abstract Shape
-
-This preprint introduces Mirai Graph, an evolutionary graph operating model for
-managing complex systems through typed objects, relations, evidence, generated
-context, readiness states and governance gates. It presents the public alpha
-standard, schemas, validators, synthetic benchmarks and public-safe pilots. The
-paper focuses on reproducible method structure and early evidence, not broad
-claims of effectiveness.
-
-## Proposed Sections
-
-1. Introduction and motivation.
-2. Related work and positioning.
-3. Mirai Graph model.
-4. Evidence, readiness and governance.
-5. Executable process control.
-6. Context-pack generation.
-7. Synthetic benchmark.
-8. Public-safe pilots.
-9. Limitations and threats to validity.
-10. Future work.
-
-## Candidate Contributions
-
-- Implementation-independent standard draft.
-- Public schemas and validators.
-- Executable governance model for launch records and process transitions.
-- Public-safe synthetic benchmark.
-- Evidence package and reproducibility protocol.
-
-## Required Before Publication
-
-- Complete baseline comparison report.
-- Add independent public-safe pilot evidence.
-- Finalize limitations table.
-- Add citations and related-work map.
-- Confirm publication venue and licensing constraints.

package/publications/semantic-intent-goal-vector-control/README.md

@@ -1,99 +0,0 @@
-# Semantic Intent And Goal Vector Control
-
-Status: article foundation
-
-## Working Title
-
-Semantic Intent and Goal-Vector Control for Reducing Drift in AI-Assisted
-Workflows
-
-## Core Question
-
-Can a graph-governed workflow reduce AI drift by controlling two failure points:
-
-1. wrong start before execution;
-2. false completion or goal drift after execution.
-
-## Proposed Contribution
-
-The article introduces a two-sided control model:
-
-```text
-Semantic Intent Resolution
--> governed work start
--> Goal Vector Quality Control
--> reverse audit before completion
-```
-
-The model treats intent confidence, goal vectors, evidence and reverse audit as
-quality-control artifacts. They do not authorize action, release or canonical
-graph update.
-
-## Hypotheses
-
-- H1. Explicit intent, scale, owner and missing-context decisions reduce
-  wrong-start errors compared with direct execution from a free-form request.
-- H2. Explicit batch-to-final-outcome vector links reduce unproductive work that
-  looks useful but does not advance the goal.
-- H3. Reverse audit reduces false completion claims by checking required
-  qualities from final outcome back to evidence and actual work.
-
-## Public Artifacts
-
-- `standard/semantic-intent-resolution.md`
-- `schemas/semantic-intent-resolution.schema.json`
-- `examples/semantic-intent-resolution-minimal/`
-- `standard/goal-vector-quality-control.md`
-- `schemas/goal-vector-quality-control.schema.json`
-- `examples/goal-vector-quality-control-minimal/`
-- `releases/1.0.0-rc.5.md`
-- `releases/1.0.0-rc.6.md`
-
-## Experiment Design
-
-Compare baseline AI-assisted work with Mirai Graph-controlled work on matched
-tasks.
-
-Baseline:
-
-```text
-free-form request -> direct execution -> final answer
-```
-
-Graph-controlled condition:
-
-```text
-free-form request
--> semantic intent record
--> goal vector
--> batch vector links
--> evidence
--> reverse audit
--> final answer or correction route
-```
-
-## Candidate Metrics
-
-- wrong-start count;
-- missing-context-at-start count;
-- work-scale mismatch count;
-- batch-without-goal-vector count;
-- false-completion claim count;
-- missing-evidence count;
-- reverse-audit finding count;
-- correction-route coverage;
-- reviewer correction count.
-
-## Required Evidence Before Publication
-
-- At least one sanitized baseline-vs-graph-controlled pilot.
-- Replayable task package with public-safe inputs and outputs.
-- Limitations table covering task selection, reviewer bias, model variance and
-  synthetic fixture limits.
-- Negative examples showing failures caught by the validator or review protocol.
-
-## Limitations
-
-Current public artifacts validate model shape and fail-closed fixtures. They do
-not prove real-world drift reduction, generalize across organizations or replace
-human review.

package/releases/0.1.0-alpha.0.md

@@ -1,45 +0,0 @@
-# Mirai Graph 0.1.0-alpha.0
-
-Release type: initial public foundation
-
-## Summary
-
-This alpha release establishes the public Mirai Graph repository as a starting
-point for standardization, developer adoption and reproducible synthetic
-examples.
-
-## Included
-
-- Public README and roadmap.
-- Repository purpose brief.
-- Concept and terminology docs.
-- Initial standard draft.
-- Object and relation models.
-- Lifecycle/readiness, evidence/provenance, governance and conformance sections.
-- Context-pack generation, profiles and test-suite direction.
-- JSON schemas for core package elements.
-- CLI validator.
-- Minimal graph example.
-- Synthetic context-reduction benchmark.
-- Initial adoption and contribution guides.
-
-## Validation
-
-Run:
-
-```bash
-npm test
-```
-
-Expected:
-
-- minimal graph validates;
-- synthetic benchmark validates;
-- synthetic context-reduction calculation returns `56.4516%`.
-
-## Important Limitations
-
-- This is an alpha foundation, not a stable standard.
-- Synthetic benchmark evidence is not external validation.
-- The project does not yet include an independent implementation pilot.
-- Context-pack generation has a documented model but no automated generator.

package/releases/0.1.0-alpha.1.md

@@ -1,63 +0,0 @@
-# Mirai Graph 0.1.0-alpha.1
-
-Release type: standard, tooling and pilot evidence alpha
-
-## Summary
-
-This release turns the initial public foundation into a more usable
-research-engineering alpha. It adds graph seed/embryo methods, transfer mapping
-from the practical `$graph` skill, alpha tooling, profiles, research protocols
-and the first independent implementation pilot.
-
-## Added
-
-- Practice-and-science program framing.
-- `$graph` to Mirai Graph transfer map.
-- Standard gap report.
-- Graph DNA, Graph Seed and Graph Embryo standard sections.
-- Graph seed and graph embryo schemas.
-- Minimal graph seed example.
-- Seed preview CLI command.
-- Readiness score CLI command.
-- Context-pack generator alpha command.
-- Profile validation command.
-- Context-pack, profile and readiness score schemas.
-- Software specification and project management profiles.
-- Measure-convert-optimize benchmark protocol.
-- Semantic completeness review protocol.
-- Independent implementation pilot plan.
-- Independent implementation pilot 001 for synthetic conference planning.
-- Worked tutorial from graph seed to validated package.
-- Negative validation fixtures.
-- `release:check` script.
-- GitHub Actions CI workflow.
-
-## Validation
-
-Run:
-
-```bash
-npm run release:check
-```
-
-Expected:
-
-- minimal graph validates;
-- synthetic benchmark validates;
-- graph seed validates;
-- profiles validate;
-- seed preview emits a graph embryo;
-- context-pack generator emits a context pack;
-- readiness score returns a capped pilot score for synthetic evidence;
-- synthetic benchmark returns `56.4516%`;
-- negative fixture fails as expected;
-- whitespace check passes.
-
-## Important Limitations
-
-- Context-pack generation is still alpha and includes the full package graph.
-- Readiness scoring is a navigation signal, not approval.
-- Pilot 001 is synthetic; it validates package creation, not real-world
-  productivity.
-- Broader scientific claims require current-mode vs graph-mode comparisons,
-  semantic completeness reviews and cross-domain replication.

package/releases/0.1.0-alpha.10.md

@@ -1,93 +0,0 @@
-# Mirai Graph v0.1.0-alpha.10
-
-Status: GitHub prerelease prepared; npm publication pending registry
-authentication.
-
-## Theme
-
-Adopter Workflow.
-
-Alpha.10 makes Mirai Graph easier to apply without author-side guidance. The
-release turns the current standards, profiles, templates, examples and CLI
-validators into a clearer self-service adoption path.
-
-## Included Scope
-
-### Adopter Journey
-
-- A concise profile-selection guide for developers, researchers, product teams
-  and organization-governance adopters.
-- A runnable path from starter template to validated package.
-- A single practical workflow:
-
-```text
-choose profile -> copy starter -> fill graph package -> validate -> generate report -> launch work -> record evidence -> run feedback/kaizen
-```
-
-### CLI Usability
-
-- `mirai-graph choose-profile` for role/profile/starter selection.
-- `mirai-graph adopter plan <role-or-profile>` for next-command guidance.
-- `mirai-graph adopter report <template-dir>` for a readable starter report.
-- `mirai-graph release state` for non-secret release-state checks across the
-  package version, git tag, GitHub Release, npm registry and npm auth boundary.
-- `validate:adopter-workflow` for command-level regression checks.
-- `validate:release-state` for pre-tag release-state report regression checks.
-- `mirai-graph release state --require-github-release` for post-GitHub-Release
-  verification.
-- Import/export helper skeletons remain a follow-up unless they directly
-  support the adopter happy path.
-
-### Adoption Evidence
-
-- Clearer organization-governance validation report.
-- Updated public-safe adopter examples.
-- Research evidence notes that distinguish synthetic evidence, pilot evidence
-  and scientific proof boundaries.
-
-## Out Of Scope
-
-- Runtime autonomy.
-- Production execution.
-- Automatic canonical updates from evidence, feedback or proposals.
-- Private SIMAI/Larena/skill-federation payloads.
-- Broad scientific proof claims.
-
-## Done
-
-- A new adopter can pick a profile and run a complete local validation path.
-- The CLI has a documented happy path for the starter templates.
-- Validation reports are readable enough to support review without inspecting
-  raw JSON first.
-- Organization-governance adoption has a stronger public-safe example/report.
-- Release-state reporting separates GitHub Release readiness, npm registry
-  state and npm authentication state.
-- `npm test` and `npm run release:check` pass.
-
-## Release State
-
-- GitHub repository: `zabarov/mirai-graph`.
-- Git tag: `v0.1.0-alpha.10`.
-- GitHub Release: `v0.1.0-alpha.10`.
-- NPM package: pending publication until npm registry authentication is
-  available.
-
-## Validation
-
-```bash
-npm run validate:adopter-workflow
-npm run validate:release-state
-npm run validate:cli-unified
-npm run validate:templates
-npm run release:check
-node packages/cli/mirai-graph.js release state --require-github-release --markdown
-```
-
-## Claim Boundary
-
-This release supports public alpha claims for starter-template adoption,
-role/profile selection, readable local reports and release-state visibility.
-
-It does not claim npm publication, peer-reviewed proof, runtime autonomy,
-production safety or automatic canonical updates from generated context,
-evidence, feedback or proposals.

package/releases/0.1.0-alpha.11.md

@@ -1,114 +0,0 @@
-# Mirai Graph v0.1.0-alpha.11
-
-Status: release candidate prepared in repository; tag, GitHub Release and npm
-publish require separate release execution.
-
-## Theme
-
-Instrumentation Layer.
-
-Alpha.11 prepares Mirai Graph to model not only governed process transitions,
-but the instruments used to steer them: development cockpit signals, feature
-implementation traceability and multi-source quality feedback.
-
-It also adds a role-oriented adopter kit that connects starter templates, CLI
-commands, validation checks and public-safety boundaries into a single entry
-point for external evaluators.
-
-The current alpha.11 evidence index is
-`docs/research/evidence-package-v0.3.md`, which connects synthetic benchmarks,
-instrumentation artifacts, public-safe pilots and claim boundaries.
-
-## Included Scope
-
-### Development Cockpit
-
-- New standard: `standard/development-cockpit.md`.
-- New schema: `schemas/development-cockpit.schema.json`.
-- New result validator mode: `development-cockpit`.
-- Synthetic report in `examples/instrumentation-layer-minimal`.
-
-The cockpit is an instrument panel. It can recommend a bounded next action, but
-it cannot authorize execution, acceptance, release or canonical graph update.
-
-### Feature Implementation Traceability
-
-- New standard: `standard/feature-implementation-traceability.md`.
-- New schema: `schemas/feature-implementation-traceability.schema.json`.
-- New result validator mode: `feature-implementation-traceability`.
-
-The traceability layer connects feature intent to implementation references,
-tests, review, documentation and evidence. It keeps `specified`, `mapped`,
-`implemented`, `tested`, `reviewed`, `accepted` and `released` separate.
-
-### Multi-Source Quality Feedback
-
-- New standard: `standard/multi-source-quality-feedback.md`.
-- New schema: `schemas/multi-source-quality-feedback.schema.json`.
-- New result validator mode: `multi-source-quality-feedback`.
-
-The feedback layer combines tests, validators, cockpit signals, traceability,
-human review and runtime observations into classified findings and transition
-routes. Blocking findings stop the transitions they block.
-
-### Synthetic Example
-
-- New package: `examples/instrumentation-layer-minimal`.
-- New npm checks:
-  - `validate:development-cockpit`;
-  - `validate:feature-implementation-traceability`;
-  - `validate:multi-source-quality-feedback`;
-  - `validate:instrumentation-layer`;
-  - `validate:instrumentation-report`;
-  - `test:instrumentation-layer-negative`.
-
-### Readable Reports And Playground
-
-- `mirai-graph report instrumentation examples/instrumentation-layer-minimal`
-  produces a readable instrumentation report.
-- `docs/adoption/adopter-kit.md` gives developers, researchers, AI product
-  teams and organization adopters a one-hour and one-day adoption path.
-- `docs/research/evidence-package-v0.3.md` summarizes public evidence maturity,
-  pilot coverage, current measured signals and reproducibility commands.
-- The deterministic playground now includes:
-
-```text
-seed -> graph -> context -> launch -> transition explanation -> cockpit -> traceability -> multi-source feedback -> evidence -> kaizen -> baseline comparison
-```
-
-### Negative Fixtures
-
-Alpha.11 fails closed on these instrumentation mistakes:
-
-- cockpit next action without required gates;
-- feature implementation mapping without evidence;
-- accepted transition feedback while a blocking finding exists;
-- process improvement without Kaizen route.
-
-## Out Of Scope
-
-- Runtime autonomy.
-- Production execution.
-- Automatic canonical updates from evidence, feedback, cockpit metrics or
-  proposals.
-- Private pilot payloads or private runtime traces.
-- Peer-reviewed scientific proof.
-
-## Validation
-
-```bash
-npm run validate:instrumentation-layer
-npm run validate:instrumentation-report
-npm run test:instrumentation-layer-negative
-npm test
-npm run release:check
-```
-
-## Claim Boundary
-
-This layer supports public alpha claims for observable process steering and
-reproducible instrumentation artifacts.
-
-It does not claim that synthetic cockpit scores predict real production
-outcomes, that tests are acceptance, or that evidence/proposals authorize
-canonical updates.

package/releases/0.1.0-alpha.2.md

@@ -1,37 +0,0 @@
-# Mirai Graph 0.1.0-alpha.2
-
-Release type: context selection and profile documentation alpha
-
-## Summary
-
-This release improves the alpha context-pack generator and documents the first
-project-management profile lessons from independent implementation pilot 001.
-
-## Added
-
-- Token matching and one-hop relation expansion for context-pack generation.
-- Updated generated context-pack artifacts.
-- Project-management profile documentation.
-- Pilot report and adoption guide wording updated to reflect alpha context
-  selection instead of full-graph-only context generation.
-
-## Validation
-
-Run:
-
-```bash
-npm run release:check
-```
-
-Expected:
-
-- all existing package, seed and profile validation checks pass;
-- context generation succeeds;
-- synthetic benchmark remains reproducible;
-- negative fixture fails as expected.
-
-## Important Limitations
-
-- Context selection is heuristic.
-- Selection explanations are not yet detailed enough for semantic review.
-- Semantic completeness must still be reviewed separately.

package/releases/0.1.0-alpha.3.md

@@ -1,61 +0,0 @@
-# Mirai Graph 0.1.0-alpha.3
-
-Release type: profile conformance, semantic review and release-process alpha
-
-## Summary
-
-This release strengthens Mirai Graph as a public research-engineering package.
-It adds a second independent implementation pilot, executable profile
-conformance checks, semantic-completeness review artifacts and a public release
-process.
-
-## Added
-
-- Independent implementation pilot 002 for a synthetic software specification
-  workflow.
-- Software-specification profile documentation.
-- Profile conformance fixture guide.
-- Validator checks for:
-  - relation id consistency with source, type and target;
-  - profile object-kind conformance;
-  - profile relation-type conformance.
-- Negative fixtures for profile and relation-id failures.
-- Semantic-completeness review artifacts for:
-  - synthetic benchmark context pack;
-  - software-specification pilot context pack.
-- Safety-context inclusion policy for context-pack generation.
-- Public release process and GitHub Release notes policy.
-
-## Validation
-
-Run:
-
-```bash
-npm run release:check
-```
-
-Expected:
-
-- package, seed, profile and pilot validation passes;
-- context-pack generation succeeds;
-- readiness score remains reproducible;
-- synthetic benchmark remains reproducible;
-- negative fixtures fail as expected;
-- whitespace check passes.
-
-## Evidence Notes
-
-- Pilot 002 supports the limited claim that a valid Level 1 Mirai Graph package
-  can be created for a small synthetic software specification.
-- Semantic review of pilot 002 improved from `13/16` to `15/16` after
-  safety-sensitive context inclusion.
-- The synthetic benchmark semantic review records `15/16`.
-
-## Important Limitations
-
-- Evidence remains synthetic and public-safe.
-- This release does not prove real-world productivity improvement.
-- Context-pack selection still lacks per-object explanations and relevance
-  scores.
-- Semantic review artifacts are maintained review records, not automated
-  consensus or peer review.