research: five executed dependence-aware certificate experiments and integrated paper

.github/workflows/dependence-certificates.yml

@@ -0,0 +1,65 @@
+name: Dependence-aware graph certificates
+on:
+  pull_request:
+  push:
+    branches: [main]
+permissions:
+  contents: read
+concurrency:
+  group: certificates-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+jobs:
+  replay:
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    env:
+      PYTHONDONTWRITEBYTECODE: '1'
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262
+        with:
+          ref: ${{ github.event.pull_request.head.sha || github.sha }}
+          persist-credentials: false
+          fetch-depth: 0
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065
+        with:
+          python-version: '3.12'
+      - name: Verify pre-execution protocol ancestry and unchanged protocol bytes
+        shell: bash
+        run: |
+          set -euo pipefail
+          protocol=bc70b76621daca198b3b97eb4f698b17d4c40a0b
+          git merge-base --is-ancestor "$protocol" HEAD
+          git show "$protocol:graph_synthesis/certificates/PROTOCOL.md" | cmp - graph_synthesis/certificates/PROTOCOL.md
+      - run: python -m pip install -r graph_synthesis/certificates/requirements.txt
+      - name: Verify results, manuscript and original evidence
+        shell: bash
+        run: |
+          set -euo pipefail
+          python -B -m unittest discover -s graph_synthesis/certificates/tests -v
+          python -B -m graph_synthesis.certificates.run --check
+          python -B -m graph_synthesis.verify --output "$RUNNER_TEMP/certificates-original.json"
+          python - <<'PY'
+          import hashlib,json
+          from pathlib import Path
+          sha=lambda p:hashlib.sha256(Path(p).read_bytes()).hexdigest()
+          for path,want in json.loads(Path('graph_synthesis/certificates/artifact-manifest.json').read_text())['sha256'].items():
+              assert sha(path)==want,path
+          build=json.loads(Path('manuscript/paper-current.build.json').read_text())
+          assert sha('manuscript/paper-current.md')==build['source_sha256']
+          assert sha('graph_synthesis/render_current_paper.py')==build['renderer_sha256']
+          assert sha('manuscript/paper-current.pdf')==build['pdf_sha256']
+          print('Certificate artifacts and current PDF hashes verified')
+          PY
+          python -B -m graph_synthesis.certificates.report --figures --update-paper
+          git diff --exit-code -- graph_synthesis/certificates manuscript/paper-current.md CURRENT_RESULTS.md
+          python -B -m graph_synthesis.render_current_paper --output "$RUNNER_TEMP/certificates-paper.pdf"
+          git diff --exit-code -- manuscript/paper-current.html
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
+        with:
+          name: certificates-replayed-paper
+          path: |
+            ${{ runner.temp }}/certificates-paper.pdf
+            ${{ runner.temp }}/certificates-paper.build.json
+            ${{ runner.temp }}/certificates-original.json
+          if-no-files-found: error
+          retention-days: 14

.github/workflows/jev-multicall.yml

@@ -55,7 +55,7 @@ jobs:
       - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065
         with:
           python-version: '3.12'
-      - run: python -m pip install -r graph_synthesis/requirements-figures.txt
+      - run: python -m pip install -r graph_synthesis/certificates/requirements.txt
       - name: Regenerate report, graphs and complete manuscript
         run: |
           python -B -m graph_synthesis.report_multicall --directory experiments/jev-multicall-20260918
@@ -69,7 +69,8 @@ jobs:
           python -B -m graph_synthesis.frontier.report --update-paper
           python -B -m graph_synthesis.novel_mechanisms.report --update-paper
           python -B -m graph_synthesis.novel_mechanisms.regret --report
-          git diff --exit-code -- experiments/jev-multicall-20260918/REPORT.md manuscript/paper-current.md CURRENT_RESULTS.md graph_synthesis/risk_control graph_synthesis/reliability graph_synthesis/structural graph_synthesis/source_structural graph_synthesis/assumption_aware graph_synthesis/uncertainty graph_synthesis/frontier
+          python -B -m graph_synthesis.certificates.report --update-paper
+          git diff --exit-code -- experiments/jev-multicall-20260918/REPORT.md manuscript/paper-current.md CURRENT_RESULTS.md graph_synthesis/risk_control graph_synthesis/reliability graph_synthesis/structural graph_synthesis/source_structural graph_synthesis/assumption_aware graph_synthesis/uncertainty graph_synthesis/frontier graph_synthesis/certificates
           python -B -m graph_synthesis.render_current_paper --output "$RUNNER_TEMP/multicall-paper.pdf"
           git diff --exit-code -- manuscript/paper-current.html
       - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02

CURRENT_RESULTS.md

@@ -82,6 +82,14 @@ Original H4 is concurrent replication of the separately merged bipartite study,
 
 <!-- FRONTIER_RESEARCH_END -->
 
+<!-- CERTIFICATES_RESEARCH_START -->
+
+## Five dependence-aware certificate experiments after PR #17
+
+[Executed report](graph_synthesis/certificates/RESULTS.md), [frozen protocol](graph_synthesis/certificates/PROTOCOL.md), [five figures](graph_synthesis/certificates/figures/) and [complete updated paper](manuscript/paper-current.pdf). H1 not met; H2 met; H3 met; H4 met; H5 met. Source-random-effects forecasting is saved-response reuse; lineage bounds, flow certificates, repair queries and delta updates are controlled algorithms. Concurrent overlap with H2-H4 is documented in the merge-time audit. No fresh Jev calls, worldwide novelty claim, new semantic-accuracy claim or production-policy change. Negative and assumption-breaking controls are retained.
+
+<!-- CERTIFICATES_RESEARCH_END -->
+
 Read the [updated full paper](manuscript/paper-current.md), [PDF](manuscript/paper-current.pdf), or [HTML](manuscript/paper-current.html).
 
 The fresh full Jev run does **not** reproduce the original interval-based entity improvement: its selected-minus-baseline macro-F1 interval includes zero. Entity accuracy still rises descriptively from 405/413 to 408/413; relation accuracy rises from 286/339 to 290/339, also without a resolved macro-F1 effect. This repeats previously inspected data; it is not independent validation.

graph_synthesis/certificates/MERGE_AUDIT.md

@@ -0,0 +1,19 @@
+# PR #20 merge-time overlap and preservation audit
+
+Reviewed main: `c726b66c3a07a115eb928b3f2629de972c728083`. Original experiment baseline and protocol remain `a62a3257645d8e35cd4e45be53bfa9511d27724b` and `bc70b76621daca198b3b97eb4f698b17d4c40a0b`. No primary endpoint, decision rule or comparator was retuned on resumption. The five original executed outcomes are retained, including H1 failure.
+
+The repository's [current-results inventory](../../CURRENT_RESULTS.md) now includes several parallel studies completed while PR #20 remained open. Therefore this merge does **not** establish five distinct never-before-attempted mechanisms relative to current main, nor worldwide priority. It preserves five baseline-scoped experiments and additional controlled replication evidence.
+
+| Certificate experiment | Relationship to intervening work | Merge-time interpretation |
+|---|---|---|
+| H1 source beta-binomial random effects | The [source-structural study](../source_structural/RESULTS.md) also studies direct source risk, but this experiment freezes an exchangeable count model and same-marginal comparator. | Different fitted forecast; H1 target still fails. |
+| H2 arbitrary-dependence DNF LP | The [assumption-aware](../assumption_aware/RESULTS.md), [uncertainty](../uncertainty/RESULTS.md) and [novel-mechanisms](../novel_mechanisms/RESULTS.md) studies also investigate marginal/dependence-safe probabilities. | Overlapping probability-bound research with explicit conjunction constraints and saved LP witnesses; not an independent new mathematical idea. |
+| H3 bipartite min-cut certificates | [Source-structural](../source_structural/RESULTS.md), [novel-mechanisms](../novel_mechanisms/RESULTS.md) and [frontier](../frontier/RESULTS.md) studies also execute bipartite optimization. | Concurrent replication/implementation extension; do not count repeatedly as distinct novelty. |
+| H4 repair-invariant queries | [Assumption-aware](../assumption_aware/RESULTS.md) and [uncertainty](../uncertainty/RESULTS.md) studies also expose alternative repairs. | Overlapping repair semantics; this suite additionally records exact and 5%-tolerant conjunctive witnesses. |
+| H5 component-local delta index | The connected-tree/path updates in [assumption-aware](../assumption_aware/RESULTS.md) and [frontier](../frontier/RESULTS.md) target different locality regimes. | Explicit mutable component index with insert/delete/split operations; connected control still has zero counted saving. |
+
+Main's generic research-marker renderer, all intervening study source/results, the original 161-file inventory, and the DSPy optimizer are preserved. Only the certificate subtree, its CI integration and additive current manuscript/index are changed. The complete paper appends the certificate section after the frontier section. Preservation is checked by removing only the new certificate block and requiring byte-identical pre-existing Markdown.
+
+Primary-source metadata was checked on September 18, 2026: [Kaski, Mannila and Mohapatra](https://arxiv.org/abs/2605.03556); [Faour, Fuchs and Kuhn](https://arxiv.org/abs/2111.10577); [Pardal et al.](https://arxiv.org/abs/2402.09265); [Staworko and Chomicki](https://arxiv.org/abs/0809.1551); [SciPy HiGHS documentation](https://docs.scipy.org/doc/scipy-1.17.0/reference/optimize.linprog-highs.html). Bibliographic correction: the frozen protocol's shortened Faour/Kuhn author attribution omits Marc Fuchs; the complete author list is recorded here without rewriting the frozen protocol. These sources document antecedents, not validation of this implementation.
+
+No new Jev calls, independent source-disjoint semantic test, matched KARMA run, real reviewer experiment or production-policy change is claimed. This is an author-side engineering integration audit, not independent external scientific peer review.