feat(contractiles+mcp): complete trident framework + fix Gemini drift

Contractile framework additions (from 2026-04-23 session, clean content):
- Add _base.ncl: shared pedigree_schema, probe_schema, run_defaults for all
  verb runners. Eliminates copy-paste; runners merge with '&' operator.
- Add INDEX.a2ml: machine-readable verb registry for CLI + CI discovery.
- Add ANCHOR.a2ml: project anchor with SSG config and ecosystem parents.
- Add must.ncl, trust.ncl, dust.ncl: runners for the three verbs that
  previously lacked them. All import _base.ncl; correct semantics.
- Add adjust.k9.ncl, intend.k9.ncl: K9 trident components. intend.k9.ncl
  carries sessional drift detection hooks and failure-mode defenses.
- Add adjust.manifest.a2ml, intend.manifest.a2ml: trident coherence
  manifests with sha256 pin schema.

6a2 repairs (reverting Gemini content-pollution from 2026-04-23):
- META.a2ml: restore ADR-002 (hermeneutic branching), ADR-003 (Rust
  rationale), ADR-005 (dual-use label); keep Gemini-added ADR-006
  (typed-wasm). Restore maintenance-axes + axis-2/3 audit rules.
- NEUROSYM.a2ml: keep improvements (rulesets, confidence-threshold,
  no-unsafe-idris/rust rules replacing no-unsafe-transmute).
- PLAYBOOK.a2ml: keep reformatted content (equivalent, slightly clearer).
- Reverted: AGENTIC.a2ml (lost fail-closed/require-evidence constraints),
  Intentfile.a2ml (replaced with template placeholders),
  Adjustfile.a2ml (replaced with Burble accessibility content),
  intend.ncl / bust.ncl / adjust.ncl (wrong semantics / wrong project).

007-mcp FFI:
- oo7_mcp_ffi.zig: implement coordDeregister (was a stub). Now POSTs
  coord_report_outcome before clearing peer_id/token. Best-effort
  fire-and-forget; peer watchdog handles lost calls.

STATE.a2ml left uncommitted pending user decision on v2 format.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: hyperpolymath

.machine_readable/6a2/ANCHOR.a2ml

@@ -0,0 +1,21 @@
+# ⚓ ANCHOR: 007-lang
+# Canonical authority and project recalibration trigger for 007-lang.
+# Template from: https://github.com/hyperpolymath/standards
+
+id: "org.hyperpolymath.007-lang"
+version: "1.0.0"
+clade: "language"
+status: "active"
+
+# SSG Configuration (Unified boj-server build)
+ssg:
+  engine: "casket"
+  output_dir: "public"
+  boj_trigger: true
+  cartridge: "ssg-mcp"
+
+# Relationships
+parents:
+  - "org.hyperpolymath.standards"
+  - "org.hyperpolymath.boj-server"
+  - "org.hyperpolymath.nextgen-languages"

.machine_readable/6a2/META.a2ml

@@ -4,58 +4,97 @@
 # META.a2ml — Project meta-level information
 # Architecture decisions, design rationale, governance.
 
-[metadata]
+# ── Project Identity ───────────────────────────────────────────────────────
+[project-meta]
 version = "1.0.0"
-last-updated = "2026-03-24"
-
-[project-info]
-type = "language"  # library | binary | monorepo | service | website | language
+name = "007-lang"
+type = "language"
 languages = ["rust"]
 license = "PMPL-1.0-or-later"
 author = "Jonathan D.A. Jewell (hyperpolymath)"
 
-[architecture-decisions]
-adrs = [
-  { id = "ADR-001", title = "Harvard architecture — data and control are separate grammars", status = "accepted", date = "2026-03-20" },
-  { id = "ADR-002", title = "Hermeneutic branching — branch is the only nondeterminism", status = "accepted", date = "2026-03-20" },
-  { id = "ADR-003", title = "Rust compiler — correctness and safety", status = "accepted", date = "2026-03-20" },
-  { id = "ADR-004", title = "BEAM primary target — actor model native support", status = "accepted", date = "2026-03-20" },
-  { id = "ADR-005", title = "Dual-use label — anti-scrape measure for private repo", status = "accepted", date = "2026-03-20" },
-]
+# ── Architecture Decisions ─────────────────────────────────────────────────
+[project-meta.architecture-decisions]
+
+[project-meta.architecture-decisions.adr-001]
+status = "accepted"
+date = "2026-03-20"
+context = "Agent injection is the primary vulnerability in multi-agent systems."
+decision = "Harvard architecture — data and control are separate grammars."
+consequences = "Agent injection is a parse error, not a runtime check. Structurally impossible to inject control into data."
+
+[project-meta.architecture-decisions.adr-002]
+status = "accepted"
+date = "2026-03-20"
+context = "Agent behaviour must be auditable and reproducible."
+decision = "Hermeneutic branching — branch is the only nondeterminism."
+consequences = "Every branch decision generates a trace record explaining why. Behaviour is auditable and reproducible."
+
+[project-meta.architecture-decisions.adr-003]
+status = "accepted"
+date = "2026-03-20"
+context = "Correctness and memory safety are non-negotiable for a meta-language."
+decision = "Rust compiler — correctness and safety."
+consequences = "No GC pauses, ownership model enforced at compile time, SPARK integration path via Idris2-ABI + Zig-FFI."
 
-[development-practices]
+[project-meta.architecture-decisions.adr-004]
+status = "accepted"
+date = "2026-03-20"
+context = "Need a robust concurrent runtime for lightweight agents."
+decision = "BEAM primary target via Elixir/OTP codegen."
+consequences = "Native support for actor model, isolation, and fault tolerance."
+
+[project-meta.architecture-decisions.adr-005]
+status = "accepted"
+date = "2026-03-20"
+context = "007 is a private repo; naming must not be a scraping surface."
+decision = "Dual-use label — anti-scrape measure for private repo."
+consequences = "The name '007' is intentionally ambiguous; the codebase is not discoverable by naive scanning."
+
+[project-meta.architecture-decisions.adr-006]
+status = "accepted"
+date = "2026-04-13"
+context = "Verification needs a stable binary target."
+decision = "Adopt typed-wasm as the only verified codegen target."
+consequences = "007 IR is lifted onto typed-wasm surface sugar for formal verification."
+
+# ── Development Practices ──────────────────────────────────────────────────
+[project-meta.development-practices]
 build-tool = "just"
 container-runtime = "podman"
 ci-platform = "github-actions"
 package-manager = "cargo"
 development-model = "claude-only"
+code-style = "rust-stable"
+security = "harvard-separation"
+testing = "canonical-proof-suite"
+versioning = "semver"
+documentation = "asciidoc"
+branching = "trunk-based"
 
-[maintenance-axes]
+# ── Maintenance Axes ───────────────────────────────────────────────────────
+[project-meta.maintenance-axes]
 scoping-first = true
 execution-order = "axis-1 > axis-2 > axis-3"
 axis-1 = "must > intend > like"
 axis-2 = "corrective > adaptive > perfective"
 axis-3 = "systems > compliance > effects"
 
-[scoping]
-sources = "README, spec/, docs/, STATE.a2ml"
-marker-scan = "TODO/FIXME/XXX/HACK/STUB/PARTIAL"
-idris-unsound-scan = "N/A — Rust project"
-
-[axis-2-maintenance-rules]
+[project-meta.axis-2-maintenance-rules]
 corrective-first = true
 adaptive-second = true
 adaptive-focus = "spec-implementation alignment, parser-EBNF sync, stale test removal"
 perfective-third = true
 perfective-source = "axis-1 honest state after corrective/adaptive updates"
 
-[axis-3-audit-rules]
+[project-meta.axis-3-audit-rules]
 audit-focus = "Harvard separation holds, no agent injection paths, session type soundness"
 compliance-focus = "EBNF coverage, test coverage per language feature, dual-use notice present"
 drift-risk-example = "runtime check replacing parse-time rejection (violates Harvard architecture)"
 effects-evidence = "test suite results, parser coverage metrics"
 
-[design-rationale]
+# ── Design Rationale ──────────────────────────────────────────────────────
+[project-meta.design-rationale]
 harvard-architecture = "Agent injection is a parse error, not a runtime check. Data and control flow through separate grammars, preventing prompt-injection-like attacks at the language level."
 hermeneutic-branching = "Every branch decision generates a trace record explaining why. This makes agent behaviour auditable and reproducible."
 six-pillars = "Harvard architecture, actors, capabilities, session types, linear handles, choreographies — each pillar addresses a specific failure mode in existing agent frameworks."

.machine_readable/6a2/NEUROSYM.a2ml

@@ -1,27 +1,26 @@
 # SPDX-License-Identifier: PMPL-1.0-or-later
-# Copyright (c) 2026 Jonathan D.A. Jewell (hyperpolymath) <j.d.a.jewell@open.ac.uk>
-#
-# NEUROSYM.a2ml — Neurosymbolic integration metadata
-# Configuration for Hypatia scanning and symbolic reasoning.
+# NEUROSYM.a2ml — Neurosymbolic integration metadata for 007-lang
+# Template from: https://github.com/hyperpolymath/standards
 
 [metadata]
 version = "1.0.0"
-last-updated = "2026-03-24"
+last-updated = "2026-04-23"
 
 [hypatia-config]
 scan-enabled = true
-scan-depth = "deep"  # quick | standard | deep — deep for language compiler
+scan-depth = "deep"
 report-format = "logtalk"
+rulesets = ["rsr-baseline", "compiler-safety", "license-compliance"]
 
 [symbolic-rules]
-# Custom symbolic rules for 007-lang
 rules = [
   { name = "harvard-separation", pattern = "data grammar must not contain control constructs", severity = "critical" },
-  { name = "no-unsafe-transmute", pattern = "transmute", severity = "critical" },
+  { name = "no-unsafe-idris", pattern = "believe_me|assert_total", severity = "critical" },
+  { name = "no-unsafe-rust", pattern = "unsafe {", severity = "major" },
   { name = "hermeneutic-trace", pattern = "every branch must generate trace record", severity = "warning" },
 ]
 
 [neural-config]
-# Neural pattern detection settings
-# confidence-threshold = 0.85
-# model = "hypatia-v2"
+confidence-threshold = 0.85
+model = "hypatia-v2-compiler"
+notes = "Uses the compiler-specialized model to detect grammar-level vulnerabilities and invariant drift."

.machine_readable/6a2/PLAYBOOK.a2ml

@@ -1,35 +1,35 @@
 # SPDX-License-Identifier: PMPL-1.0-or-later
-# Copyright (c) 2026 Jonathan D.A. Jewell (hyperpolymath) <j.d.a.jewell@open.ac.uk>
-#
-# PLAYBOOK.a2ml — Operational playbook
-# Runbooks, incident response, deployment procedures.
+# PLAYBOOK.a2ml — Operational playbook for 007-lang
+# Template from: https://github.com/hyperpolymath/standards
 
 [metadata]
 version = "1.0.0"
-last-updated = "2026-03-24"
+last-updated = "2026-04-23"
 
 [deployment]
-method = "ci-triggered"  # gitops | manual | ci-triggered
-target = "binary"        # container | binary | library | wasm
+method = "ci-triggered"
+target = "binary"
+notes = "Primary binary output is the 007 compiler (oo7). Codegen targets BEAM and typed-wasm."
 
 [incident-response]
-# 1. Check .machine_readable/STATE.a2ml for current status
-# 2. Review recent commits and CI results
-# 3. Run `just test` to verify evaluator passes all 29 tests
-# 4. Run `just validate` to check compliance
-# 5. Verify Harvard separation: `just check-harvard`
+steps = [
+  "1. Read 0-AI-MANIFEST.a2ml for canonical locations.",
+  "2. Check .machine_readable/6a2/STATE.a2ml for active blockers (clippy, proof-debt).",
+  "3. Run `just test` to verify evaluator/parser core.",
+  "4. Verify Harvard separation: `just check-harvard` (parser-level grammar check).",
+]
 
 [release-process]
-# 1. Update version in STATE.a2ml, META.a2ml, Cargo.toml
-# 2. Run `just release-preflight` (validate + quality + security + maint-hard-pass)
-# 3. Ensure all EBNF sections have corresponding Pest rules
-# 4. Tag and push
-# 5. Run `just container-push` if applicable
+steps = [
+  "1. Update version and last-updated in all 6A2 files and Cargo.toml.",
+  "2. Run `just validate-rsr` and ensuring EBNF parity with the Pest parser.",
+  "3. Verify all canonical proofs pass with `just canonical-proof-suite`.",
+  "4. Tag the release with `vX.Y.Z`.",
+  "5. Push tags and update the 'last-updated' field in META.a2ml.",
+]
 
 [maintenance-operations]
-# Baseline audit:
-#   just maint-audit
-# Hard release gate:
-#   just maint-hard-pass
-# Parser coverage check:
-#   just parser-coverage
+audit = "just maint-audit"
+hard-pass = "just maint-hard-pass"
+canonical-proofs = "just canonical-proof-suite"
+parser-coverage = "just parser-coverage"

.machine_readable/contractiles/INDEX.a2ml

@@ -0,0 +1,133 @@
+# SPDX-License-Identifier: PMPL-1.0-or-later
+# INDEX.a2ml — Contractile Registry
+# Author: Jonathan D.A. Jewell <j.d.a.jewell@open.ac.uk>
+#
+# Machine-readable catalogue of all contractile verbs in this template set.
+# Consumers (CI scripts, the contractile CLI, Hypatia rules) SHOULD read this
+# file to discover available verbs rather than hard-coding the list.
+#
+# See: docs/CONTRACTILE-SPEC.adoc §Registry
+
+---
+id = "contractiles-registry"
+version = "2.0.0"   # 2.0.0 (2026-04-18): all 6 verbs on trident shape; verb set complete.
+spec = "docs/CONTRACTILE-SPEC.adoc"
+last_updated = "2026-04-18"
+base_schema = ".machine_readable/contractiles/_base.ncl"
+meta_schema_status = "pending — see CONTRACTILE-SPEC §validator-meta-schema"
+
+## Verbs
+
+[[verbs]]
+name = "adjust"
+semantics = "drift tolerances + corrective actions"
+trident = [
+  "adjust/Adjustfile.a2ml",
+  "adjust/adjust.ncl",
+  "adjust/adjust.k9.ncl",
+]
+manifest = "adjust/adjust.manifest.a2ml"
+status = "active"
+tier = "Yard"
+authority = "advisory"
+gating = "advisory (continue-with-warnings)"
+cardinality = "one per repo"
+notes = "Fifth trident instance (2026-04-18). First (Yard, advisory) authority pattern. Specialises in cumulative-drift catchment — tolerance bands + trend tracking across sessions. auto_fix_when_available applies deterministic patches; advisory otherwise."
+
+[[verbs]]
+name = "bust"
+semantics = "hard-stop / expiry / must-not-run declarations"
+trident = [
+  "bust/Bustfile.a2ml",
+  "bust/bust.ncl",
+  "bust/bust.k9.ncl",
+]
+manifest = "bust/bust.manifest.a2ml"
+status = "active"
+tier = "Hunt-read-only"
+authority = "blocking"
+gating = "hard (exit-nonzero)"
+cardinality = "one per repo"
+notes = "Fourth trident instance (2026-04-18). Completes the blocking-authority triple (must + trust + bust). Specialises in deprecated-path-reintroduction catchment. Injects failures via declared probes and verifies recovery paths."
+
+[[verbs]]
+name = "dust"
+semantics = "rollback / recovery / deprecation / audit-trail preservation"
+trident = [
+  "dust/Dustfile.a2ml",
+  "dust/dust.ncl",
+  "dust/dust.k9.ncl",
+]
+manifest = "dust/dust.manifest.a2ml"
+status = "active"
+tier = "Yard"
+authority = "advisory"
+gating = "advisory (continue-with-warnings)"
+cardinality = "one per repo"
+notes = "Sixth and FINAL trident instance (2026-04-18) — completes the full verb set. Specialises in audit-trail preservation + rollback-path verification. Destructive actions gated behind --apply flag + per-item approval; dry-run default."
+
+[[verbs]]
+name = "intend"
+semantics = "north-star (commitments + aspirations)"
+trident = [
+  "intend/Intentfile.a2ml",
+  "intend/intend.ncl",
+  "intend/intend.k9.ncl",
+]
+manifest = "intend/intend.manifest.a2ml"
+status = "active"
+tier = "Hunt"
+authority = "reporting"
+gating = "non-gating (continue)"
+cardinality = "one per repo"
+notes = "First trident instance in the estate (2026-04-18). Reports progress toward committed next-actions AND lists horizon aspirations. Absorbed the deprecated `lust` verb 2026-04-18. Never blocks. Remaining 5 verbs still on file_pair shape until tridents are built."
+
+[[verbs]]
+name = "k9"
+semantics = "trust-tier templates (EXCEPTION to one-verbfile rule)"
+file_pair = [
+  "k9/template-hunt.k9.ncl",
+  "k9/template-kennel.k9.ncl",
+  "k9/template-yard.k9.ncl",
+]
+status = "exception"
+gating = "not applicable"
+notes = "k9 is service-automation meta-infrastructure, not a verb contractile. Three trust-tier templates (Kennel/Yard/Hunt). Does not have a Verbfile.a2ml. See CONTRACTILE-SPEC §k9-exception."
+
+# [[verbs]] lust REMOVED 2026-04-18 — name had unwanted associations;
+# the horizon/aspiration semantics were always meant to live inside `intend`
+# (the north-star verb). The [[wishes]] schema was absorbed into
+# intend/Intentfile.a2ml. Any `lust/` dir found in an estate repo is drift
+# and should be deleted.
+
+[[verbs]]
+name = "must"
+semantics = "invariant assertion — release-blocking"
+trident = [
+  "must/Mustfile.a2ml",
+  "must/must.ncl",
+  "must/must.k9.ncl",
+]
+manifest = "must/must.manifest.a2ml"
+status = "active"
+tier = "Hunt-read-only"
+authority = "blocking"
+gating = "hard (exit-nonzero)"
+cardinality = "one per repo"
+notes = "Third trident instance (2026-04-18). Completes the blocking-authority pair with trust: must = concrete + persistent invariants; trust = concrete + ephemeral transactions. Specialises in subtle invariant-erosion (tracking per-session trend; flagging silent regression). Single failure blocks merge. Simplest and most commonly populated verb."
+
+[[verbs]]
+name = "trust"
+semantics = "security + provenance + safe-hacking"
+trident = [
+  "trust/Trustfile.a2ml",
+  "trust/trust.ncl",
+  "trust/trust.k9.ncl",
+]
+manifest = "trust/trust.manifest.a2ml"
+status = "active"
+tier = "Hunt"
+authority = "blocking"
+gating = "hard (exit-nonzero)"
+cardinality = "one per repo"
+notes = "Second trident instance (2026-04-18). First (Hunt, blocking) verb — hard gate. Primary defense against threat-model misclassification (B1) and 'turn off the firewall' capability-collapse (C2). Inherits on_open negotiation+accountability+translation from intend.k9.ncl v2.0.0; adds threat_model_foregrounding + block_session_close_on_critical_drift."