feat(uci): add preregistered AGI evidence protocol

.claude.json

@@ -9,7 +9,7 @@
   "provenance": {
     "generator": "scripts/build-cognitive-manifest.py",
     "repository": "Aegis-Omega/AEGIS-OMEGA",
-    "source_ref": "feat/uci-6-collective-memory-admission-v1",
+    "source_ref": "feat/uci-7-agi-evidence-protocol-v1",
     "parent_state_hash": "e9f0ec153b0b320a1e791092f73209442ec43a982e503203d9c101ec40949cba",
     "signature_mode": "GITHUB_OIDC_ATTESTATION"
   },
@@ -488,5 +488,5 @@
       "on_success": "broadcast-attested-verified-event-stream"
     }
   },
-  "state_hash": "06ebe0307234544bf68399a40f320843806979e1909213dd2b7aa4bbdff5e0c8"
+  "state_hash": "f935e25dffad6b9cb2f89cb8eac843a65aef9536bc6a65cd0183b7d360759655"
 }

.github/workflows/uci-7-agi-evidence-protocol.yml

@@ -0,0 +1,205 @@
+name: UCI-7 AGI Evidence Protocol
+
+on:
+  pull_request:
+    branches:
+      - feat/uci-6-collective-memory-admission-v1
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  uci7-agi-evidence:
+    name: uci7 / agi-evidence-protocol
+    runs-on: ubuntu-24.04
+    timeout-minutes: 20
+    env:
+      CANDIDATE_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
+      EXPECTED_PARENT_SHA: 156062855a91b77133d8999ce34883432435b167
+      PR_BASE_SHA: ${{ github.event.pull_request.base.sha || '' }}
+    steps:
+      - name: Checkout exact candidate
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ env.CANDIDATE_SHA }}
+          fetch-depth: 0
+
+      - name: Verify exact stacked lineage
+        shell: bash
+        run: |
+          set -euo pipefail
+          actual="$(git rev-parse HEAD)"
+          test "$actual" = "$CANDIDATE_SHA"
+          test "$(git merge-base "$EXPECTED_PARENT_SHA" HEAD)" = "$EXPECTED_PARENT_SHA"
+          if [[ -n "$PR_BASE_SHA" ]]; then
+            test "$PR_BASE_SHA" = "$EXPECTED_PARENT_SHA"
+          fi
+          echo "UCI7_CANDIDATE_SHA=$actual"
+          echo "UCI7_EXPECTED_PARENT_SHA=$EXPECTED_PARENT_SHA"
+          echo "UCI7_EXACT_LINEAGE=PASS"
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+
+      - name: Install exact test runtime
+        run: python -m pip install pytest==8.3.5 jsonschema==4.23.0
+
+      - name: Validate inherited and UCI-7 schemas
+        shell: bash
+        run: |
+          set -euo pipefail
+          python - <<'PY'
+          import json
+          from pathlib import Path
+          from jsonschema.validators import validator_for
+
+          paths = [
+              Path('schemas/transition-identity-envelope.v1.schema.json'),
+              Path('schemas/decision-receipt.v1.schema.json'),
+              Path('schemas/execution-receipt.v1.schema.json'),
+              Path('schemas/effect-receipt.v1.schema.json'),
+              Path('schemas/effect-witness.v1.schema.json'),
+              Path('schemas/complete-verification-result.v1.schema.json'),
+              Path('schemas/admission-record.v1.schema.json'),
+              Path('schemas/quarantined-evidence-memory-record.v1.schema.json'),
+              Path('schemas/memory-projection-request.v1.schema.json'),
+              Path('schemas/canonical-memory-record.v1.schema.json'),
+              Path('schemas/memory-control-request.v1.schema.json'),
+              Path('schemas/memory-control-record.v1.schema.json'),
+              Path('schemas/capability-task-spec.v1.schema.json'),
+              Path('schemas/capability-trial-result.v1.schema.json'),
+              Path('schemas/evaluation-suite.v1.schema.json'),
+              Path('schemas/agi-evidence-assessment.v1.schema.json'),
+          ]
+          for path in paths:
+              schema = json.loads(path.read_text(encoding='utf-8'))
+              validator_for(schema).check_schema(schema)
+              print(f'UCI7_SCHEMA_OK {path}')
+          print(f'UCI7_SCHEMA_COUNT={len(paths)}')
+          if len(paths) != 16:
+              raise SystemExit('UCI7_SCHEMA_CARDINALITY_MISMATCH')
+          PY
+
+      - name: Run inherited UCI-4 UCI-5 UCI-6 proofline
+        shell: bash
+        run: |
+          set -euo pipefail
+          python -m pytest \
+            sovereign-omega-v2/python/tests/test_transition_receipts_pr1.py \
+            sovereign-omega-v2/python/tests/test_transition_receipts_cli_pr1.py \
+            sovereign-omega-v2/python/tests/test_effect_adapters_pr2.py \
+            sovereign-omega-v2/python/tests/test_effect_verifier_pr3.py \
+            sovereign-omega-v2/python/tests/test_complete_verifier_pr4.py \
+            sovereign-omega-v2/python/tests/test_complete_verifier_pr4_receipt_binding.py \
+            sovereign-omega-v2/python/tests/test_uci4_effect_chain_integration.py \
+            sovereign-omega-v2/python/tests/test_uci5_atomic_admission.py \
+            sovereign-omega-v2/python/tests/test_uci6_collective_memory.py \
+            sovereign-omega-v2/python/tests/test_uci6_memory_prestate.py \
+            sovereign-omega-v2/python/tests/test_uci6_memory_schemas.py \
+            sovereign-omega-v2/python/tests/test_uci6_internal_base_guard.py \
+            -q | tee "$RUNNER_TEMP/uci7-inherited-pytest.log"
+          grep -Eq '131 passed' "$RUNNER_TEMP/uci7-inherited-pytest.log"
+          echo "UCI7_INHERITED_PROOFLINE_131=PASS"
+
+      - name: Validate inherited UCI-5 CI contract guards
+        shell: bash
+        run: |
+          set -euo pipefail
+          python -m pytest sovereign-omega-v2/python/tests/test_uci5_ci_contract.py -q | tee "$RUNNER_TEMP/uci7-inherited-uci5-ci.log"
+          grep -Eq '3 passed' "$RUNNER_TEMP/uci7-inherited-uci5-ci.log"
+          echo "UCI7_INHERITED_UCI5_CI_GUARDS_3=PASS"
+
+      - name: Run preregistered UCI-7 falsifiers
+        shell: bash
+        run: |
+          set -euo pipefail
+          python -m pytest \
+            sovereign-omega-v2/python/tests/test_uci7_agi_evidence_protocol.py \
+            sovereign-omega-v2/python/tests/test_uci7_agi_evidence_schemas.py \
+            -q | tee "$RUNNER_TEMP/uci7-pytest.log"
+          grep -Eq '13 passed' "$RUNNER_TEMP/uci7-pytest.log"
+          echo "UCI7_LOCAL_PROTOCOL_13=PASS"
+
+      - name: Validate UCI-7 CI contract guards
+        shell: bash
+        run: |
+          set -euo pipefail
+          python -m pytest sovereign-omega-v2/python/tests/test_uci7_ci_contract.py -q | tee "$RUNNER_TEMP/uci7-ci-contract.log"
+          grep -Eq '4 passed' "$RUNNER_TEMP/uci7-ci-contract.log"
+          echo "UCI7_CI_CONTRACT_GUARDS_4=PASS"
+
+      - name: Validate checker-result provenance boundary
+        shell: bash
+        run: |
+          set -euo pipefail
+          python -m pytest sovereign-omega-v2/python/tests/test_uci7_checker_provenance.py -q | tee "$RUNNER_TEMP/uci7-checker-provenance.log"
+          grep -Eq '2 passed' "$RUNNER_TEMP/uci7-checker-provenance.log"
+          echo "UCI7_CHECKER_PROVENANCE_GUARDS_2=PASS"
+
+      - name: Validate constituent baseline attribution
+        shell: bash
+        run: |
+          set -euo pipefail
+          python -m pytest sovereign-omega-v2/python/tests/test_uci7_baseline_attribution.py -q | tee "$RUNNER_TEMP/uci7-baseline-attribution.log"
+          grep -Eq '5 passed' "$RUNNER_TEMP/uci7-baseline-attribution.log"
+          echo "UCI7_BASELINE_ATTRIBUTION_GUARDS_5=PASS"
+
+      - name: Write exact-head witness
+        if: always()
+        shell: bash
+        run: |
+          set -euo pipefail
+          {
+            echo "schema_version=AEGIS_UCI7_CI_WITNESS_V1"
+            echo "candidate_sha=$(git rev-parse HEAD)"
+            echo "expected_parent_sha=$EXPECTED_PARENT_SHA"
+            echo "authority=EVIDENCE_ONLY"
+            echo "agi_claim=NOT_ESTABLISHED"
+            echo "unit_tests_are_agi_proof=FALSE"
+            echo "external_replication=NOT_ESTABLISHED"
+            echo "public_benchmark_campaign=NOT_RUN"
+            echo "required_axes=6"
+            echo "hidden_answer_exposure=FAIL_CLOSED"
+            echo "safety_violation=HYPOTHESIS_REJECTED"
+            echo "closed_serialization=REQUIRED_AND_TESTED"
+            echo "schema_cardinality=16_REQUIRED_AND_TESTED"
+            echo "inherited_proofline_cardinality=131_REQUIRED_AND_TESTED"
+            echo "inherited_uci5_ci_guards=3_REQUIRED_AND_TESTED"
+            echo "uci7_local_protocol_cardinality=13_REQUIRED_AND_TESTED"
+            echo "uci7_ci_guards=4_REQUIRED_AND_TESTED"
+            echo "checker_result_provenance=PROCESS_LOCAL_ISSUANCE_REQUIRED_AND_TESTED"
+            echo "evaluated_system_runtime_binding=REQUIRED_AND_TESTED"
+            echo "collective_baseline_attribution=5_REQUIRED_AND_TESTED"
+            echo "baseline_delta_semantics=PAIRED_DESCRIPTIVE_NOT_SIGNIFICANCE"
+            echo "checker_code_cryptographic_attestation=NOT_ESTABLISHED"
+            echo "cross_process_trial_result_attestation=NOT_ESTABLISHED"
+          } > "$RUNNER_TEMP/uci7-witness.txt"
+          for path in \
+            "$RUNNER_TEMP/uci7-inherited-pytest.log" \
+            "$RUNNER_TEMP/uci7-inherited-uci5-ci.log" \
+            "$RUNNER_TEMP/uci7-pytest.log" \
+            "$RUNNER_TEMP/uci7-ci-contract.log" \
+            "$RUNNER_TEMP/uci7-checker-provenance.log" \
+            "$RUNNER_TEMP/uci7-baseline-attribution.log"; do
+            if [[ -f "$path" ]]; then
+              cat "$path" >> "$RUNNER_TEMP/uci7-witness.txt"
+            fi
+          done
+
+      - name: Upload UCI-7 witness
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: uci7-agi-evidence-witness-${{ env.CANDIDATE_SHA }}
+          path: |
+            ${{ runner.temp }}/uci7-witness.txt
+            ${{ runner.temp }}/uci7-inherited-pytest.log
+            ${{ runner.temp }}/uci7-inherited-uci5-ci.log
+            ${{ runner.temp }}/uci7-pytest.log
+            ${{ runner.temp }}/uci7-ci-contract.log
+            ${{ runner.temp }}/uci7-checker-provenance.log
+            ${{ runner.temp }}/uci7-baseline-attribution.log
+          if-no-files-found: error

docs/superpowers/specs/2026-08-20-uci7-agi-evidence-protocol-v1-design.md

@@ -0,0 +1,129 @@
+# UCI-7 — AGI Evidence Protocol v1
+
+Date: 2026-08-20
+Parent: `#278@156062855a91b77133d8999ce34883432435b167`
+Status: PREREGISTERED DESIGN / NO AGI CLAIM
+
+## Purpose
+
+UCI-7 turns `AGI = NOT_ESTABLISHED` into a falsifiable evaluation program. It does **not** define a magic scalar whose passage proves AGI. It defines a preregistered, multi-axis evidence contract whose outputs can strengthen, weaken, or reject an AGI hypothesis while preserving AEGIS authority/evidence separation.
+
+## External calibration anchors
+
+The protocol is aligned to public evaluation ideas rather than invented only for AEGIS:
+
+- ARC-AGI-2: novel abstraction/reasoning on private evaluation tasks and explicit efficiency constraints.
+- GAIA: general-assistant tasks requiring reasoning, multimodality, browsing and tool use.
+- METR task-completion time horizon: reliability as a function of human-equivalent task difficulty/duration.
+
+These are reference anchors, not imported claims that AEGIS has passed them.
+
+## Required evidence axes
+
+Every admitted evaluation suite MUST contain all required axes. No weighted average may compensate for a missing or failed required axis.
+
+1. `NOVEL_ABSTRACTION_TRANSFER` — solve held-out tasks whose rule is not supplied directly.
+2. `CROSS_DOMAIN_GENERALITY` — competence across preregistered unrelated domains.
+3. `TOOL_AND_ENVIRONMENT_AGENCY` — multi-step tool/environment work with deterministic end-state checking.
+4. `LONG_HORIZON_RELIABILITY` — task difficulty/horizon measured against a human reference, with repeated trials.
+5. `SAFE_ADAPTATION` — use admitted memory/feedback to improve later performance without violating UCI-4/UCI-5/UCI-6 invariants.
+6. `METACOGNITIVE_CALIBRATION` — predicted correctness vs actual correctness, inheriting the historical Hallucination Delta idea without treating historical Kaggle outputs as current proof.
+
+## Anti-gaming / contamination contract
+
+A suite is invalid for AGI evidence unless all of the following are preregistered before execution:
+
+- immutable suite identifier and manifest digest;
+- task IDs, axis membership, scoring/checker identities, budgets and trial counts;
+- hidden-case or answer commitments whose plaintext answers are unavailable to the evaluated agent;
+- fixed thresholds and minimum task counts;
+- provider/model/runtime/configuration identity;
+- strongest constituent baseline protocol under the same task and budget manifest;
+- deterministic checker outputs; self-reported or LLM-judge-only correctness is forbidden;
+- explicit contamination declaration per task family;
+- no threshold, task-weight or exclusion change after result observation.
+
+## Evidence objects
+
+### `CapabilityTaskSpecV1`
+
+Binds one task to:
+
+- `task_id`
+- `axis`
+- `domain`
+- `hidden_case_commitment`
+- `checker_commitment`
+- `budget_commitment`
+- `human_reference_commitment`
+- `trial_count`
+- `contamination_class`
+
+### `CapabilityTrialResultV1`
+
+Binds actual execution evidence to a task spec:
+
+- task-spec root;
+- trial index;
+- deterministic checker verdict/score;
+- output digest;
+- execution/effect/admission roots when available;
+- actual resource expenditure;
+- model/provider/runtime identity.
+
+The result MUST NOT accept caller-declared correctness as authority.
+
+### `CapabilityAxisAssessmentV1`
+
+Aggregates only preregistered task results belonging to one axis. Missing required trials fail closed.
+
+### `AGIEvidenceAssessmentV1`
+
+Produces one of:
+
+- `NOT_EVALUATED`
+- `INSUFFICIENT_EVIDENCE`
+- `PARTIAL_EVIDENCE`
+- `PREREGISTERED_THRESHOLD_MET`
+- `HYPOTHESIS_REJECTED`
+
+`AGI_PROVEN` is intentionally not a runtime status in v1. A preregistered threshold crossing is evidence, not an ontological theorem. Stronger public claims require independent replication and external benchmark receipts.
+
+## Conjunctive gate
+
+Let required axes be `A`. For preregistered axis thresholds `theta_a`:
+
+`ThresholdMet => forall a in A: Complete(a) AND Score(a) >= theta_a`
+
+Additionally:
+
+- no hidden-answer exposure may be detected;
+- all checker roots must match the preregistered manifest;
+- all task/trial cardinalities must match exactly;
+- system and constituent baseline manifests must be budget-comparable;
+- metacognitive calibration must be computed from predictions made before checker revelation;
+- safety invariant violations force `HYPOTHESIS_REJECTED` for that run regardless of capability score.
+
+## Collective-intelligence attribution
+
+AEGIS-level capability and AGI evidence are distinct from provider capability. UCI-7 therefore records the strongest constituent baseline under the same manifest and computes system-vs-baseline deltas per axis. A positive delta is evidence for collective-system contribution; it is not required to call a constituent model intelligent, but it is required for any claim that AEGIS coordination itself adds capability.
+
+## Explicit non-claims
+
+```text
+AGI = NOT_ESTABLISHED
+ARC_AGI_2_PASS = NOT_ESTABLISHED
+GAIA_PASS = NOT_ESTABLISHED
+METR_TIME_HORIZON_RESULT = NOT_ESTABLISHED
+INDEPENDENT_EXTERNAL_REPLICATION = NOT_ESTABLISHED
+CURRENT_UCI7_STATUS = PROTOCOL_ONLY_UNTIL_TESTED
+```
+
+## Implementation sequence
+
+1. RED: preregister falsifiers for missing-axis compensation, post-hoc threshold mutation, fabricated correctness, hidden-answer exposure, task-cardinality drift, baseline budget mismatch, unsafe-run override and pre-checker calibration binding.
+2. GREEN: implement pure deterministic protocol objects/evaluator only; no model calls.
+3. Serialization: closed Draft 2020-12 schemas with required `*_kind` discriminators and separate hash domains.
+4. Native CI: exact frozen-parent binding to `#278@156062855...`, schema checks, exact test cardinality and evidence artifact.
+5. Evaluation adapters: ARC/GAIA/METR-compatible result ingestion only after protocol core is green.
+6. Actual capability campaign: run external/public + held-out suites; no AGI status promotion from unit tests of the evaluator itself.