Axiom Education

Axiom Education is a local-first, lifelong education platform and education-domain layer for AXIOM-MESH. It is designed to support learning across age groups, jurisdictions, institutions, and life stages without making any one curriculum, grade band, school system, or delivery experience the product boundary.

Current rebuild: 0.5.0-dev.0
Status: active development; not production-ready
Canonical repository: Zoverions/Axiom-Education
Canonical branch: main
First supported jurisdiction: Ontario, Canada
Active Ontario tracks: Kindergarten / Grades 1-8 curriculum-capsule foundation and Secondary curriculum, with MTH1W as the first deep course vertical slice

The historical product and repository name OntarioEdAI is deprecated. Ontario remains the first supported jurisdiction because its source corpus and active rebuild work are present; it is not the identity or architectural boundary of the platform. Elementary, secondary, post-secondary, apprenticeship, professional, reskilling, civic, hobby, and later-life learning can share the same governed learner and evidence substrate while using different curriculum packs and experience layers.

See the High-School Foundation Strategy, Ontario Elementary Readiness, Repository Rename Record, Branch Hygiene Policy, and Deprecations.

Product boundary

Axiom Education application / experience layer
  -> AXIOM Gateway
  -> policy, consent, risk, and plan evaluation
  -> short-lived capability grant
  -> approved education, curriculum, or provider capsule
  -> bounded execution
  -> encrypted learner state and evidence in Grid

The Flutter application remains independently releasable. AXIOM-MESH supplies the policy, consent, bounded execution, evidence, portability, and synchronization substrate for governed effects.

A curriculum pack, grade band, school program, or learning experience is a configured domain surface inside Axiom Education. None of those surfaces defines Axiom Education as a whole. Installing a curriculum or education capsule does not grant access to learner records, activate a provider, or authorize execution.

Current usable surface

The runnable application now opens on a learner-facing local-first home rather than exposing the curriculum database as the product. From that home, a learner can enter the Grade 9 Math learning hub, start bounded offline practice, browse the local Ontario secondary curriculum library, or open family/home-learning tools. These routes remain intentionally narrow: current practice feedback is temporary, no learner progress is persisted by default, and the learner-facing experience does not claim a grade, mastery result, credit, transcript, enrolment, or Ministry approval.

In parallel, the Ontario Elementary track has a complete base source-acquisition layer for its current composed source set: 16/16 bounded metadata-only C1 snapshots, 8/8 English-language Grades 1-8 required-program-family source coverage, 8/8 French-language-school required-program-family source coverage, and separate Kindergarten 2026 C1 evidence. Five source surfaces are strict exact-byte PDFs and eleven DCP HTML surfaces remain observational. Captured Ontario source bytes are not committed and every historical C1 lock remains review-required for redistribution.

Elementary source capture is not curriculum readiness. Current human source identity/scope review is 0/16 approved, licensing review is 0/16 resolved, and the canonical Elementary records-v2 directory contains 0 C2 records. Deterministic source-review and licensing-review plans, a reviewer dossier, and a fail-closed C2 intake gate are executable; C2 candidate eligibility remains 0/16 until current human source approval and compatible licensing evidence exist. The C2 intake v1 supports reference-only candidates only and additionally requires operator-supplied source bytes to match the historical C1 SHA-256 exactly.

The currently runnable secondary surface includes:

• a learner-facing home with direct routes to Grade 9 Math learning, quick practice, curriculum reference, and family tools;
• recoverable first-run initialization and clear local loading, retry, and empty states;
• a searchable curriculum library with course and expectation counts plus course, strand, expectation, and tag browsing;
• pull-to-refresh and explicit retry behavior for the bundled curriculum;
• automatic restoration of a missing or invalid bundled curriculum database;
• a read-only curriculum database and a separate local settings store;
• a Grade 9 Math learning hub that puts four short foundation lessons and mixed practice ahead of the larger draft-unit explorer;
• a four-lesson Grade 9 math foundations path with explicit goals, prerequisites, direct instruction, worked examples, multiple reasoning routes and representations, misconception checks, and reflection prompts;
• source-mapped offline MTH1W draft content for all 9 planned units and all 43 primary lesson slots, with 86 worked examples, 473 guided/independent/retrieval practice items, nine delayed-feedback 10-item quizzes with correction attempts, and nine transparent performance tasks;
• repo-bounded split-unit loading for Units 8 and 9 so geometry/measurement and financial-literacy lessons remain individually reviewable while still passing through the same canonical content contract as Units 1 through 7;
• lesson-selected focused practice plus a mixed-practice route, both available without AI;
• deterministic offline practice for MTH1W-A1, MTH1W-A2, MTH1W-B2, and MTH1W-B4;
• actual answer entry with exact rational and slope-intercept verification, with blank or whitespace-only submissions blocked;
• an ephemeral on-screen summary of total attempts, distinct questions checked, and distinct questions answered correctly; repeated checks of the same question cannot inflate the distinct-success count, and nothing is saved to a learner record;
• a three-different-question stopping cue that is explicitly not a grade or mastery result;
• an in-app and written home-learning routine for two learners sharing a device, with a direct route from the guide into mixed practice;
• item-linked scaffolded hints, derived local topic references, uncalibrated difficulty disclosure, and digest evidence, with technical identifiers kept behind expandable detail controls;
• learner cards that allow explanatory text to grow rather than intentionally truncating it, improving behavior under larger system text sizes;
• fail-closed behavior when the practice configuration, item integrity, or verifier is unavailable.

The source-mapped unit drafts remain AI-free, and the Claw foundations preview remains fully usable without AI. The preview now has an experimental Socratic path that is exposed only when a reviewed host explicitly injects a governed execution binding. That binding is limited to the exact learner, Socratic task, target competency, current learner input, allowed egress/retention, budget, and eligible provider set; it performs one bounded invocation, treats returned text as instructional content rather than authority, records minimized usage metadata, and falls back to the reviewed non-model explanation when execution is denied or unavailable. No model or provider is activated by default, and the Socratic path cannot create mastery, a grade, credit, credential, or learner-record write. The repository also contains an experimental, host-injected AXIOM Gateway client and governed learner write/self-read runtime, but it has no default host binding, local learner-record fallback, production provider, or authority-expanding path. The authored nine-unit milestone is not a complete MTH1W course, credit, grade, transcript, Ministry-approved resource, school enrolment, or replacement for an authorized education provider. Each authored unit has a draft quiz and performance task, while reviewed complete-course assessment, production learner progress, delegated authority, pack activation, portfolio export, and classroom synchronization remain experimental, disabled, specified, or adapter-required in config/capabilities.json.

The branch also contains an experimental Axiom-authored cross-cutting computational-literacy competency family: 18 non-jurisdictional nodes spanning Intent, Specification, Verification, and Systems Understanding, with no competency edges in v1. The bounded executable proofs are one deterministic MTH1W verifier-selection experience and one optional governed-Socratic Claw verification-reasoning experience. This is not official Ontario curriculum, is not validated psychometrics, creates no AI-literacy score, has no model requirement, and has no automatic mastery or grade effect. Model output cannot establish correctness or learner evidence; MathAnswerVerifier remains the sole correctness authority for the bounded MTH1W answer-checking path, and both proofs retain non-generative/no-AI behavior.

The source conflict and blocked completion claim are documented in the MTH1W Source Audit and enforced by config/curriculum-readiness.json. The source-pinned official inventory contains all 57 expectation references without redistributing their verbatim descriptions; the MTH1W Coverage Ledger records the difference between authored coverage and reviewed coverage. The Home Learning Guide describes safe current use. Product sequencing is MTH1W first, then every remaining Grade 9 course, then later grades, as defined in the Course Completion Roadmap. The Global Instructional Methods Baseline requires multiple valid routes without assigning fixed learning-style labels. The scope and negative-path gates for the current MTH1W implementation are frozen in MTH1W Phase 0 and Phase 1.

First five minutes

Supported rebuild toolchain:

• Flutter 3.41.1
• Dart 3.11.x
• Python 3.12 for curriculum and claim verification
• OpenSSL 3.x for external-key Ed25519 curriculum-pack signing

python tools/verify.py
flutter devices
flutter run -d <device-id>

The verification command validates the supported toolchain, installs the exact Python test dependencies and locked Dart dependencies, checks capability and curriculum-readiness claims, verifies Ontario Elementary source-review and licensing-review evidence plus the deterministic reviewer dossier and fail-closed C2 intake gate, checks both monolithic and split authored-unit content, checks formatting, runs static analysis, and runs the complete Python and Flutter test suites. Use an activated virtual environment if you do not want the Python development dependencies installed into your user environment.

For an Android installation smoke build:

flutter build apk --debug

The APK is written to build/app/outputs/flutter-apk/app-debug.apk. The protected GitHub workflow also builds this APK and publishes it as a short-lived workflow artifact after formatting, analysis, and tests pass.

Desktop runs require the corresponding Flutter desktop toolchain to be enabled. For example:

flutter config --enable-windows-desktop
flutter run -d windows

Axiom Education contract

The generic cross-repository contract is contracts/axiom-education.v1.json.

• Brand: Axiom Education
• Contract ID: axiom.education
• Version: 1.0.0
• Curriculum-pack profile: jurisdictional
• Gateway: POST /v1/intents
• Installation grants authority: false

The contract defines bounded curriculum inspection, staging, activation, querying, tutoring, learner-event, progress, and portfolio actions. Learner-data actions require exact purpose-bound consent. High-risk activation and export require explicit confirmation and independent approval.

Mesh compatibility and forward roadmap

The governed runtime requires the machine-verified config/axiom-mesh-compatibility.v1.json profile in addition to a host-injected relative Gateway requester. The profile records the current reviewed AXIOM-MESH main checkpoint and full Gateway contract digest as provenance, not as repository-head runtime authority. Runtime binding is scoped to the exact Education contract, learner-memory profile, the semantic Gateway intents.submit seam, and the full Gateway -> Hypervisor -> Sandbox -> Grid authority path. A missing or incompatible runtime seam leaves the runtime unbound before any token or request is used.

This boundary deliberately tolerates unrelated additive Mesh work. New social, agent, observability, or read-only Gateway routes do not silently broaden Education authority and do not, by themselves, invalidate the Education host binding. A change to the Education /v1/intents seam or an authority-bearing Education contract still requires explicit review and a new compatibility profile.

Runtime Capsule, Personal Agent Pack, Compute Node Profile, Local Trust Envelope, and Agent Runtime Adapter surfaces are tracked only as personal-compute/interoperability readiness. Assurance Graph, provider-observation, and checkout-freshness foundations have merged into AXIOM-MESH but remain readiness-only for Education with runtime adoption explicitly disabled. Delegated human authority and the AXIOM Host profile remain draft/non-production inputs. Merging or installing a contract does not promote it: a reviewed Education adoption decision, compatibility evidence, and executable tests are required.

A missing provider, verifier, identity, policy, consent, source, or artifact must produce an explicit unavailable or denied result. It must never produce mock success in a governed or release build.

Current capability status

The authoritative status registry is config/capabilities.json.

Presently:

• curriculum browsing from a schema-verified bundled SQLite database is implemented;
• the current Ontario secondary corpus deterministically builds into 293 canonical records across 21 courses;
• the frozen MTH1W subset is checked against the full corpus and independently rebuilt twice byte-for-byte;
• MTH1W-labelled A1, A2, B2, and B4 practice generation and bounded deterministic local checking are experimental and explicitly limited to four derived topic references whose official mapping is under review;
• the official MTH1W hierarchy is source-pinned at 14 overall and 43 specific expectations, while educator/cultural review, licensing, reviewed course-wide coverage, cumulative assessment review, accessibility, progress, and teacher workflows remain incomplete;
• a machine-verified 9-unit, 43-lesson MTH1W blueprint maps every official expectation exactly once within a 110-hour planning envelope, but authored primary lessons total only 63.75 hours and leave 46.25 hours unallocated; all nine unit slots are authored machine-verified drafts, not a complete course;
• Units 8 and 9 use repo-bounded split manifests that are materialized through the same canonical unit validator in Python and the same Mth1wUnitContent runtime model in Flutter;
• the four-lesson Grade 9 Math Foundations Preview is experimental; it compares multiple valid reasoning routes but is not a complete MTH1W course;
• the Axiom-authored cross-cutting computational-literacy capability is experimental: exactly 18 non-jurisdictional nodes, zero v1 edges, one deterministic MTH1W proof, and one optional governed-Socratic Claw proof; it is not official Ontario curriculum, not validated psychometrics, produces no AI-literacy score, requires no model, and has no automatic mastery or grade effect;
• curriculum-pack generation, digest verification, and external-key Ed25519 signing are experimental and do not authorize activation;
• Ontario-derived records and Axiom Education extensions use visibly separate namespaces and official-recognition flags;
• legacy irt_* values are exported only as visibly uncalibrated adaptation heuristics;
• the governed Socratic execution seam is experimental and can call only an explicitly injected eligible provider after learner/task/context/privacy/egress/retention/budget hard filters; it creates no education authority and falls back to reviewed non-model instruction on denial or provider failure;
• local Phi-3 tutor inference remains disabled, fails closed, and no longer returns simulated logits output; no model or provider is activated by default;
• canvas observation is an explicitly experimental bounded classifier and no longer returns mock equations;
• handwriting scoring no longer returns fixed synthetic scores when its model is missing;
• the legacy UDP/TCP classroom mesh is disabled by default, AES-GCM protected when explicitly enabled for development, and is not a trusted authority path;
• governed learner records, selective portfolio export, accessibility gates, pack activation, and AXIOM classroom synchronization remain specified or adapter-required;
• Ontario Elementary base source capture is complete for the current composed source set: 16/16 metadata-only C1 snapshots, 5 strict PDF and 11 observational HTML monitoring surfaces, with 8/8 English and 8/8 French required-program-family source coverage plus Kindergarten 2026 evidence;
• Ontario Elementary human source review is 0/16 approved, licensing review is 0/16 resolved, canonical C2 records remain 0, and the reference-only C2 intake gate therefore has 0/16 eligible sources. None of those incomplete gates may be inferred from base source capture.

Ontario curriculum packs

Ontario is the first jurisdiction family, not the product boundary. Elementary and secondary content are separate curriculum tracks that share generic pack, provenance, jurisdiction, learner, and governance infrastructure.

Build the current deterministic unsigned Ontario secondary pack into an empty directory:

python tools/curriculum_pack.py build \
  --input assets/curriculum/ontario_curriculum_full.json \
  --ledger curriculum/source-ledger.v1.json \
  --output /tmp/axiom-education-ontario \
  --pack-id ontario-secondary \
  --pack-version 1.0.0

Build the frozen MTH1W phase-one subset:

python tools/curriculum_pack.py build \
  --input curriculum/slices/mth1w.v1.json \
  --ledger curriculum/source-ledger.v1.json \
  --output /tmp/axiom-education-mth1w \
  --pack-id ontario-mth1w-phase-1 \
  --pack-version 1.0.0

Verify content integrity:

python tools/curriculum_pack.py verify \
  --pack-dir /tmp/axiom-education-ontario

The protected workflow builds the complete current Ontario secondary pack twice and compares manifest.json and records.jsonl byte-for-byte. It also verifies that the frozen MTH1W subset contains exactly 11 source-identical records and produces byte-identical repeated builds. The complete pack is then signed with an ephemeral Ed25519 key and must pass signature verification.

Retrieval indexes are disposable derived artifacts. They are never the authority for curriculum content.

Curriculum trust boundary

A valid pack signature proves that the exact canonical manifest was signed by the corresponding private key. It does not prove:

• Ontario Ministry approval;
• curriculum correctness or completeness;
• licensing or redistribution permission;
• pedagogical quality;
• safe application activation.

The MTH1W source audit pins the reviewed official PDF digest and documents known identifier conflicts. The broader secondary source ledger still records that upstream official-document digests have not been captured for the entire corpus and that course-by-course source and licensing review remains required.

The Elementary track has a separate evidence chain. Its current composed source set has 16/16 bounded C1 snapshots, but the exact source bytes are not retained in the repository. Five PDF sources have strict exact-byte monitoring and eleven DCP HTML sources remain observational response surfaces. Human source review, licensing review, canonical C2 normalization, deterministic full-pack evidence, staging, and governed activation are independent later gates. The deterministic reviewer dossier packages metadata, plans, and blank human-review templates but explicitly excludes captured Ontario source bytes. The C2 intake gate additionally requires operator-supplied bytes to match the historical C1 digest and cannot currently admit any source because review evidence is still incomplete.

Platform evidence

The protected repository gates exercise Linux-hosted verification and Android packaging plus a dedicated Windows quality lane. The active integration branch also contains a bounded Apple lane using the exact Flutter 3.41.1 lock to run formatting, analysis, Flutter tests, a macOS debug build, and an iOS debug build with code signing disabled.

Apple CI is build-compatibility evidence only. It does not establish signing, notarization, App Store readiness, physical-device deployment, production Apple support, or accessibility approval. Native/plugin-sensitive dependency upgrades remain separately reviewed because Android-only evidence is not sufficient to certify platform behavior.

Canonical documents

• Product definition
• Ontario Elementary readiness
• Changelog
• Evidence-gated requirements
• MTH1W Phase 0 and Phase 1
• MTH1W source audit
• MTH1W coverage ledger
• Home learning guide
• Course completion roadmap
• Education model integration
• Curriculum Pack v1
• Repository rename record
• Branch hygiene policy
• Deprecations
• Current curriculum source policy

Historical research is preserved under docs/archive/. When documents conflict, executable behavior and config/capabilities.json control. Archived or deprecated documents are never current claim authority.

Security and privacy boundary

The project targets local processing and data minimization, but local-first architecture alone is not a compliance or security claim. Minor-related data defaults prohibit advertising, behavioural targeting, covert attention or emotion monitoring, diagnosis inference, raw prompts in evidence, and raw student work in logs. Human review and appeal remain mandatory design requirements.