v0.3.1.0 readiness

Author: theislampill

.github/workflows/ci.yml

@@ -0,0 +1,40 @@
+name: CI
+
+on:
+  push:
+  pull_request:
+
+jobs:
+  runtime-checks:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Install checker dependencies
+        run: python -m pip install --upgrade pip PyYAML
+
+      - name: Build and verify runtime
+        run: |
+          python tools/build_framework_pipeline.py
+          python tools/build_compiled_runtime.py
+          python tools/check_compiled_runtime_freshness.py
+          python tools/check_compiled_module_boundaries.py
+          python tools/check_stub_integrity.py
+          python tools/check_consolidation_call_budget.py
+          python tools/check_routing_parity.py
+          python tools/check_routing_parity.py --strict
+          python tools/check_recursive_traversal_governance.py
+          python tools/check_render_modes.py
+          python tools/check_frontmatter.py
+          python tools/check_coverage.py
+          python tools/check_framework_pipeline.py
+          python tools/check_recursion_collapse_noetic_frame.py
+          python tools/check_metacompliance_current_canon.py
+          python tools/check_smoke_artifacts.py
+          python tools/check_ir_instance_integrity.py
+          python tools/check_diagnostic_ir_catalogue_integrity.py
+          git diff --check

AGENTS.md

@@ -34,10 +34,14 @@ python tools/check_coverage.py
 python tools/check_recursion_collapse_noetic_frame.py
 python tools/check_metacompliance_current_canon.py
 python tools/check_smoke_artifacts.py
+python tools/check_ir_instance_integrity.py
+python tools/check_diagnostic_ir_catalogue_integrity.py
 ```
 
 Before packaging or pushing, run the full applicable checker suite and confirm the generated
-runtime is fresh. When `smokes/runtime-grounding-v5/` exists, include the smoke artifact gate.
+runtime is fresh. The canonical smoke artifact gate is repo-local
+`smokes/runtime-grounding-v5/`; it is committed regression evidence with trace/verdict
+provenance, and `tools/check_smoke_artifacts.py` defaults to that root.
 Package only when explicitly requested.
 
 ## Skill Architecture
@@ -95,6 +99,9 @@ False-compliance guards are executable, not merely prose:
   render only, while runtime proof and verdict language stay in `trace.md` and `verdict.md`.
 - Hard smoke PASS verdicts below the depth floor, cross-fixture contamination, and repeated generic
   paragraphs across unrelated outputs must fail the smoke artifact checker.
+- Release smoke artifacts must carry provenance in `trace.md` or `verdict.md`: package filename,
+  SHA256, model/host, invocation mode, prompt pointer, timestamp, and live-run versus
+  handcrafted-regression classification.
 - `Operation:` lines must begin with the closed operative verbs named in
   `recursive-state-transitions.md`.
 - Non-operative source-status use requires the operative-warrant sentence with the

CHANGELOG.md

@@ -15,13 +15,22 @@
 - Added `tools/check_metacompliance_current_canon.py` to guard the current docs, root control plane,
   generated default surface, source-status/noetic-frame anchors, and held-release anchors against
   metacompliance drift.
+- Added Diagnostic IR catalogue/source-basis integrity checking via
+  `tools/check_ir_instance_integrity.py` and
+  `tools/check_diagnostic_ir_catalogue_integrity.py`, including ghost-load, omnibus active-module,
+  class-mismatch, source-basis mismatch, and post-render decision-consistency bad samples.
 - Updated README, AGENTS, TODO, and tooling docs so the current checker suite includes the
   metacompliance current-canon guard.
 - Added `tools/check_smoke_artifacts.py` and the `runtime-grounding-v5` hard/bounded smoke gate to
   reject cross-fixture contamination, smoke/test scaffold language in default output, repeated
   generic paragraphs across unrelated smoke outputs, and hard PASS verdicts below the depth floor.
+- Aligned smoke artifacts to the repo-local `smokes/runtime-grounding-v5/` root and required
+  release-smoke provenance fields for package filename, SHA256, model/host, invocation mode,
+  prompt pointer, timestamp, and live-run classification.
 - Hardened render/release governance so output files remain clean user-facing default render
   artifacts while runtime proof and grading stay in trace/verdict artifacts.
+- Clarified package naming: PowerShell packaging emits the canonical RC `.skill.zip`; upload hosts
+  that require `.skill` should receive the same checked payload renamed, not a re-zipped repo root.
 
 ### Current guidance
 
@@ -45,6 +54,9 @@
 > Release-history note: this entry records v0.3.1.0 state. Terms such as `State Refresh`,
 > next eligible door, compact Layer A light surface, or full audit are historical wording when
 > they differ from the current compact DSL/IR + bounded governed response + state/noetic re-read canon.
+> Current mainline canon after v0.3.1.0 adds checker-backed IR source-basis integrity,
+> repo-local smoke artifact provenance, and default render wording that avoids literal
+> STOP/HOLD/RECURSE/PARTIAL governance labels in public output.
 
 v0.3.1.0 is the render-mode governance and release-hygiene patch release. It fixes a regression in
 which the default `/daee-epistemics` invocation was emitting full `[Diagnostic IR]` code-fenced

README.md

@@ -288,6 +288,8 @@ python tools/check_framework_pipeline.py
 python tools/check_recursion_collapse_noetic_frame.py
 python tools/check_metacompliance_current_canon.py
 python tools/check_smoke_artifacts.py
+python tools/check_ir_instance_integrity.py
+python tools/check_diagnostic_ir_catalogue_integrity.py
 ```
 
 The compiled runtime may still name atomized paths such as `references/tactics/M9-predication-mode.md`.
@@ -491,10 +493,16 @@ class CASEOUT,LAYERS,REST,CATALOGUE,FRONTMATTER,PFAUDIT slate;
 
 ## Install / Package (Claude-First)
 
-The distributable artifact for this repository is `daee-epistemics.skill`. 
-Its archive root must contain `SKILL.md`, `references/`, `compiled-module-map.json`, and `build-manifest.json` directly. 
-Do not zip the whole repo root, and do not produce a bundle whose top level is `skill/`.
-Package the contents of the generated `skill/` directory, not the directory itself.
+The canonical user-facing upload name is `daee-epistemics.skill`. The canonical local RC build
+filename is `build/daee-epistemics-RC00001-v0.3.1.0.skill.zip`.
+`package.ps1` emits a local `.skill.zip` archive because it is a zip payload with the skill root
+at archive root. If a host expects `.skill`, rename that checked `.skill.zip` payload to
+`daee-epistemics.skill`; do not re-zip it.
+
+The archive root must contain `SKILL.md`, `references/`, `compiled-module-map.json`, and
+`build-manifest.json` directly. Do not zip the whole repo root, and do not produce a bundle whose
+top level is `skill/`. Package the contents of the generated `skill/` directory, not the directory
+itself.
 
 Before release, regenerate and verify the runtime with the command set in [Source / Runtime Layout](#source--runtime-layout). The checked local packaging command is:
 
@@ -503,7 +511,9 @@ powershell -NoProfile -ExecutionPolicy Bypass -File .\package.ps1 build\daee-epi
 ```
 
 Smoke tests should use the latest explicitly named rc package and hash from the current run.
-Do not use `build/compiled-skill/` or older rc archives as smoke-test inputs.
+Do not use `build/compiled-skill/` or older rc archives as smoke-test inputs. The release smoke
+artifact suite is committed under `smokes/runtime-grounding-v5/` as regression evidence; future
+live runs may regenerate it, but the checker defaults to that repo-local root.
 Current readiness checks and smoke prompts live in [`docs/package-smoke-readiness.md`](docs/package-smoke-readiness.md).
 
 For path fidelity, build the archive from Bash / WSL / Linux rather than Windows zip tooling. This keeps archive entry names slash-safe for skill hosts that inspect the bundle structure directly.

TODO.md

@@ -18,6 +18,16 @@ Completed historical gaps, coverage passes, routing parity work, and recursive t
 
 No active unresolved release-blocking technical pass is currently assigned.
 
+## Completed Mainline Audit Patch Catalogue (post-v0.3.1.0)
+
+- Diagnostic IR ghost-load/source-basis enforcement is checker-backed by
+  `tools/check_ir_instance_integrity.py` and
+  `tools/check_diagnostic_ir_catalogue_integrity.py`.
+- Runtime-grounding smoke artifacts are canonical at repo-local
+  `smokes/runtime-grounding-v5/`; release traces/verdicts carry package/hash/run provenance.
+- Package naming is split deliberately: `package.ps1` emits a checked local `.skill.zip` RC archive,
+  while host uploads that require `.skill` may use the same payload renamed to `daee-epistemics.skill`.
+
 ## Completed Metaaudit Repair Catalogue (2026-05-02)
 
 The `C:\ai\current\metaaudit.md` report found the skill partially compliant: the

atomics/skill/references/diagnostics/diagnostic-ir.md

@@ -337,9 +337,10 @@ bounded move → state re-read: what cleared → what remains live
 → next eligible burden → decision → next bounded move (if eligible)
 ```
 
-Literal `Recursion decision:` and `next_eligible_pass:` fields must not appear in default
-output. They belong to `:dsl`, internal/development audit, pass-review, or diagnostic trace. In default mode,
-RECURSE appears as a prose transition plus one bounded next pass, not as a state label.
+Literal `Recursion decision:`, `next_eligible_pass:`, `Governance:`, or visible
+STOP / HOLD / RECURSE / PARTIAL governance labels must not appear in default output. They
+belong to `:dsl`, internal/development audit, pass-review, or diagnostic trace. In default
+mode, continuation appears as a prose transition plus one bounded next pass, not as a state label.
 
 Numbered essay headings do not substitute for this progression. "Move 1 / Move 2 /
 Move 3" is not a post-render gate unless the response also shows, in ordinary prose,
@@ -449,7 +450,7 @@ Why already present:
 Released module(s):
 Bounded move:
 state re-read:
-Governance: STOP | HOLD | RECURSE | PARTIAL
+Release status: prose closure/hold/partial/continuation status; no literal STOP/HOLD/RECURSE/PARTIAL label
 ```
 
 This is a DSL/audit shape, not a normal-answer ledger. Ordinary default answers may compress it,

atomics/skill/references/diagnostics/diagnostic-ir.schema.json

@@ -181,6 +181,11 @@
       "type": "string",
       "enum": ["strong", "provisional", "low"]
     },
+    "decisive_missing_differentiator": {
+      "type": "string",
+      "minLength": 1,
+      "description": "Conditional-mandatory when confidence is not strong or read_status is not dominant; names the signal that would refine, falsify, or collapse the remaining ambiguity."
+    },
     "alignment_state": {
       "type": "string",
       "enum": [
@@ -424,6 +429,36 @@
       },
       "then": false
     },
+    {
+      "if": {
+        "properties": {
+          "confidence": {
+            "not": {
+              "const": "strong"
+            }
+          }
+        },
+        "required": ["confidence"]
+      },
+      "then": {
+        "required": ["decisive_missing_differentiator"]
+      }
+    },
+    {
+      "if": {
+        "properties": {
+          "read_status": {
+            "not": {
+              "const": "dominant"
+            }
+          }
+        },
+        "required": ["read_status"]
+      },
+      "then": {
+        "required": ["decisive_missing_differentiator"]
+      }
+    },
     {
       "if": {
         "properties": {

atomics/skill/references/diagnostics/framework-pipeline.yaml

@@ -744,7 +744,7 @@ support_claims:
           - "Layer A"
           - "Layer B"
           - "state re-read"
-          - "If Governance = RECURSE, continue with Burden-Cycle N+1"
+          - "If the release status says another bounded pass is licensed, continue with Burden-Cycle N+1"
 
   restoration_trace_shape:
     owners:

atomics/skill/references/diagnostics/recursive-state-transitions.md

@@ -398,10 +398,11 @@ re-read burden-cycle shape in `diagnostic-render-contract.md`, then repeats only
 recursive sufficiency.
 
 Default mode must not print literal state fields such as `Recursion decision:`,
-`next_eligible_pass:`, `post_render_gate:`, or STOP / HOLD / RECURSE / PARTIAL as a visible
-governance label. Literal state labels and post-render fields are for `:dsl`, internal/development audit,
-pass-review, or diagnostic trace. Default RECURSE is valid only when the answer visibly
-performs the transition and the one bounded next pass; naming the decision is not execution.
+`next_eligible_pass:`, `post_render_gate:`, `Governance:`, or STOP / HOLD / RECURSE / PARTIAL
+as a visible governance label. Literal state labels and post-render fields are for `:dsl`,
+internal/development audit, pass-review, or diagnostic trace. Default continuation is valid
+only when the answer visibly performs the transition and the one bounded next pass; naming the
+decision is not execution.
 
 ### Minimum Visible Transition Spine
 
@@ -551,7 +552,7 @@ Why already present:
 Released module(s):
 Bounded move:
 state re-read:
-Governance: STOP / HOLD / RECURSE / PARTIAL
+Release status: prose closure/hold/partial/continuation status; no literal STOP/HOLD/RECURSE/PARTIAL label
 ```
 
 ## Grounded Noetic Re-Read Shape

atomics/skill/references/rubrics/diagnostic-render-contract.md

@@ -183,8 +183,8 @@ Layer A block is the compact diagnostic frame for default mode; it is not raw Di
 - Cleared:                      [what this live burden cleared]
 - Remaining input-anchored burdens: [enumerated from original input, not a topic list]
 - Held routes rechecked:        [result after this pass]
-- Next bounded pass:            [if RECURSE]
-- Governance: RECURSE | HOLD | PARTIAL | STOP
+- Next bounded pass:            [prose reason if another bounded pass is licensed]
+- Release status:               [prose closure/hold/partial/continuation status; no literal STOP/HOLD/RECURSE/PARTIAL label]
 
 ### Restorative Response
 [Required once in default output after state/noetic re-read. Bounded to what the released operation(s) actually landed. Do not promote it into a new burden-cycle. Do not release held downstream burdens.]
@@ -193,8 +193,9 @@ Layer A block is the compact diagnostic frame for default mode; it is not raw Di
 [Required once at the very end after Restorative Response. Synthesize what cleared, what remains held, and the final governed takeaway. Do not substitute for state/noetic re-read.]
 ```
 
-If Governance = RECURSE, continue with Burden-Cycle N+1. If STOP / HOLD / PARTIAL, close with a
-brief prose reason — no literal governance label in default mode.
+If the release status says another bounded pass is licensed, continue with Burden-Cycle N+1.
+If closure, hold, or partial traversal is correct, state the reason in prose — no literal
+STOP / HOLD / RECURSE / PARTIAL governance label in default mode.
 
 **Layer A required default fields:** read status, confidence, claim_level,
 pattern_profile, reason-category, concealment, deformation, DO-orient, live noetic
@@ -341,9 +342,10 @@ required field set; everything else remains internal. It must surface enough gov
 diagnostic fact for the current pass to be auditable as compiler traversal rather than essay.
 
 **Compact state re-read:** The state re-read block above is compact. It must enumerate
-remaining input-anchored burdens, not merely state the governance decision. "Governance: STOP"
-alone is not a valid state re-read — it must show that no input-anchored eligible burden
-remains, or name the burden and the reason for HOLD / PARTIAL.
+remaining input-anchored burdens, not merely state a governance decision. A line such as
+"Governance: STOP" is forbidden in default output and is not a valid state re-read. The
+default must show that no input-anchored eligible burden remains, or name the burden and
+the prose reason for a hold or partial close.
 
 **Single-Pass Layer A/B Cosplay:** A response that prints Layer A + Layer B + state re-read
 exactly once and then stops — without proving no eligible input-anchored live burden remains,
@@ -869,7 +871,7 @@ Trace: M9-predication-mode + diagnostic-render-contract.
 - Cleared: composition / dependence pressure dissolved through the lexical and category split.
 - Remaining input-anchored burdens: none in this prompt.
 - Held routes rechecked: full attribute exposition and source-comparison remain held.
-- Governance: STOP
+- Release status: closed for this input; no same-input eligible burden remains after the category correction.
 
 ### Restorative Response
 The restored order is that real predication does not become dependency merely by being
@@ -932,7 +934,7 @@ Trace: FPD + M1 + output-release.
   as subordinate supports for that tribunal.
 - Remaining input-anchored burdens: none newly licensed in this prompt.
 - Held routes rechecked: broad punishment doctrine and pastoral expansion remain held.
-- Governance: STOP unless a later prompt supplies a burden not already handled as a submove.
+- Release status: closed for this input unless a later prompt supplies a burden not already handled as a submove.
 
 #### Restorative Response
 The restored order is that mercy, guidance, accountability, and worship-worthiness cannot be