perturbation-sim: substrate calibration (study as ground truth) + calibrated SoA member spec

.claude/board/TECH_DEBT.md

@@ -147,6 +147,38 @@ timing-race hypothesis (read the actual `cargo llvm-cov` log with a scoped token
 Cross-ref: `.github/workflows/rust-test.yml` (test job mold step vs coverage job);
 `bindspace-singleton-to-mailbox-soa-v1` (the migration this is NOT).
 
+**2026-06-16 addendum — the `test` job now hits the SAME cliff; fix extended to
+it (branch `claude/ci-test-job-debuginfo0`).** The cliff this entry called out as
+*"a 2/50 intermittent"* on the coverage job has now surfaced on the **plain
+`test`** job: `ld terminated with signal 7 [Bus error]` + an LLVM crash dump at
+the `cargo test --no-run` link step of `test_sql_query` / `intervene_counterfactual`.
+Root-caused to a **link-footprint growth, not a logic break** (a layout break would
+fail an assertion, not SIGBUS at link): **PR #507** (`0c6ef02c`, +4055/−1048 across
+`causal-edge` edge.rs/layout.rs — the ce64-v2 layout — and `cognitive-shader-driver`
+mailbox_soa.rs/driver.rs/planner_bridge.rs — MailboxSoaOwner + SurrealMailboxView,
+D-PG-6) grew the object-file set linked by the lance-graph integration tests enough
+to tip the previously-marginal `test`-job link over the same disk/RSS ceiling. It
+surfaced on the first full-workspace CI run *after* #507 (the two PRs between, #509
+and the perturbation-sim #511, are root-`exclude`d so their CI never linked the
+post-#507 tree — which is why this is "the first failing PR" yet not its fault).
+**This is a FENCE, not a root reduction:** it does not shrink #507's legitimate
+codegen; it removes the dead `debuginfo=1` weight (CI never opens a debugger) to
+buy headroom — exactly the b56bb2cd lever, now applied to the `test` job. **Fix:**
+job-level `RUSTFLAGS: "-C debuginfo=0 -C target-cpu=x86-64-v3"` on `test` (parity
+with `test-with-coverage`; mold already installed). Side effect: the `test` job
+gets its own Swatinem cache key (first run repopulates). **Confirm** on the next
+green `test` run. **Residual debt if it recurs after this:** the footprint is on a
+secular upward trend (every cognitive-layer PR adds codegen) — the durable fix is a
+bigger runner or splitting the integration-test link set, not repeatedly shaving
+flags. Separately, #507 left `intervene_counterfactual.rs:133/165` calling the
+**deprecated** `CausalEdge64::inference_type()` (the consumer-migration commit
+`8131c480` lives on the unmerged `claude/continue-ndarray-x0Oaw`) — that WARNS, does
+not fail (v1 default routes through the canonical mapping per I-LEGACY-API-FEATURE-
+GATED); tracked here as a separate latent item, not fixed on this CI branch.
+Cross-ref: `.github/workflows/rust-test.yml` (now both jobs at `debuginfo=0`); PR
+#507 (`0c6ef02c`); `claude/continue-ndarray-x0Oaw` (the pending ce64-v2 consumer
+migration).
+
 ### TD-UNBUNDLE-FROM-1 — `unbundle_from` is NOT the inverse of `bundle_into` (2026-06-07)
 
 **Open.** `crates/lance-graph-planner/src/cache/kv_bundle.rs` — `unbundle_from`

.github/workflows/rust-test.yml

@@ -29,6 +29,22 @@ jobs:
   test:
     runs-on: ubuntu-24.04
     timeout-minutes: 30
+    env:
+      # Override the workflow-level debuginfo=1 for this job too (parity with
+      # test-with-coverage, TD-CI-COVERAGE-MOLD-1). The `test` job links the
+      # full lance+datafusion integration-test set at the SAME disk/RSS cliff
+      # the coverage job hit — and #507 (+4055 lines across causal-edge +
+      # cognitive-shader-driver: ce64-v2 layout + MailboxSoaOwner/
+      # SurrealMailboxView) grew that link footprint enough to tip the
+      # previously-marginal link into a hard `ld` SIGBUS (signal 7 = object
+      # file truncated when the runner partition fills mid-link). debuginfo=1
+      # carried no value here (CI never opens a debugger); dropping it cut the
+      # coverage job's per-binary link from ~930 MB to ~252 MB (-73%, measured
+      # in b56bb2cd) and relieves BOTH ceilings (mold/GNU-ld RSS + disk). mold
+      # is already installed below. Note: a job-level RUSTFLAGS gives this job
+      # its own Swatinem cache key — the first run after this change
+      # repopulates the test cache.
+      RUSTFLAGS: "-C debuginfo=0 -C target-cpu=x86-64-v3"
     defaults:
       run:
         working-directory: lance-graph

crates/perturbation-sim/CLAM_CHAODA_FRAMING.md

@@ -0,0 +1,75 @@
+# Framing the resilience study as CLAM (hierarchy) + CHAODA (anomaly ensemble)
+
+*The resilience study is not a bespoke method — it is a CLAM cluster tree with a
+CHAODA-style anomaly ensemble, on the electrical-distance manifold. Both already
+exist in `ndarray::hpc::clam` (cited below); this doc maps the correspondence and
+states honestly what is grounded vs what is a conceptual mapping not yet wired.*
+
+> Operator prompt (2026-06-16): "you could even try to frame it — CLAM
+> (resilience) / CHAODA". Companion to `PAPER.md`, `COUNTRY_STUDY.md`,
+> `src/columns.rs`.
+
+## The correspondence
+
+| resilience-study object | CLAM / CHAODA construct | ndarray reference |
+|---|---|---|
+| recursive Cheeger/HHTL bisection into basins | **CLAM cluster tree** (`ClamTree::build`) | `hpc/clam.rs` `ClamTree` |
+| a basin (compartment) | a **`Cluster`** | `hpc/clam.rs:106` `Cluster { radius, cardinality, lfd }` |
+| basin algebraic connectivity λ₂ / mean R | cluster **radius** / spread | `Cluster::radius` |
+| basin node count | cluster **cardinality** | `Cluster::cardinality` |
+| how fragmented/space-filling a basin is | **local fractal dimension** `Lfd` | `hpc/clam.rs:81` `Lfd::compute(count_r, count_half_r)` |
+| fail-first / exposure ranking | **CHAODA anomaly score** | `hpc/clam.rs:1517` `ClamTree::anomaly_scores() -> Vec<AnomalyScore>` |
+| "this compartment can't wait" flag | CHAODA **flag threshold** (≥ 0.75) | `hpc/clam.rs` anomaly-flag test |
+
+So the study's machinery is CLAM's machinery on a different metric: instead of a
+Hamming/embedding distance, the manifold is the **electrical distance** (effective
+resistance `R_ij = (e_i−e_j)ᵀ L⁺ (e_i−e_j)`, the self-inverse `L⁺` reference). The
+HHTL tiers ARE the CLAM tree depth; the weakest compartment IS the cluster CHAODA
+would score as the outlier.
+
+## Why the three axes ARE a CHAODA ensemble (the load-bearing match)
+
+CHAODA's thesis: **no single graph-anomaly method wins; ensemble several *diverse*
+detectors** (relative cardinality, parent/child cardinality ratio, graph
+neighbourhood, stationary distribution, …) and the gain comes from their
+*non-redundancy*. The resilience study's three axes are exactly such an ensemble:
+
+- **topology** (λ₂ / Kirchhoff) — the connectivity detector,
+- **buffer** (inertia storage) — the transient detector,
+- **policy** (feed-in / dispatch) — the operational detector,
+
+ensembled into the **exposure** score. And the study's measured **low / negative
+Cronbach α** (the axes are distinct facets, `Spearman ≈ 0` between them) is not a
+defect — it is *precisely CHAODA's design goal*: low inter-detector correlation is
+what makes the ensemble add information rather than restate it. The discriminant
+finding and the CHAODA non-redundancy principle are the same statement.
+
+This also re-frames the §4.11 confound cleanly: the modifier `Weyl × (1/Fiedler)`
+failed as an independent axis because `1/λ₂` is the dominant Kirchhoff term — i.e.
+it was a **redundant detector**, the CHAODA anti-pattern. The buffer axis is the
+*orthogonal* detector the ensemble actually needed.
+
+## Honest scope
+
+- **Grounded [G]:** `CLAM`, `Cluster{radius,cardinality,lfd}`, `Lfd`, and
+  `ClamTree::anomaly_scores` all exist in `ndarray::hpc::clam` (cited). The
+  structural correspondence is exact, not metaphor.
+- **Conceptual [H]:** `perturbation-sim` is zero-dep and is **NOT wired** to
+  `ndarray::hpc::clam`. The mapping above is read off the APIs, not run. No code
+  here calls `ClamTree` or `anomaly_scores`.
+- **The falsifiable probe** that would promote [H]→[G]: build a `ClamTree` over the
+  contingency factor vectors (or the per-basin `(λ₂, Kf, buffer)` rows), run
+  `anomaly_scores`, and correlate the CHAODA ranking against the study's exposure
+  ranking (ICC/Spearman, Jirak rate). If they agree, the study *is* CHAODA on the
+  electrical manifold; if not, the framing is rhyme and gets retracted. This is the
+  gated bridge (crosses perturbation-sim's zero-dep boundary into `ndarray`,
+  behind a feature flag) — analogous to the calibration harness, not yet built.
+
+## Tie-in to the calibrated columns (`src/columns.rs`)
+
+CLAM gives two more value members for free, hung off the same HHTL-OGAR key:
+`radius` (basin spread) and `lfd` (local fractal dimension). They are helix-residue
+value members like the rest — orthogonal to topology by the key/value split — and
+the CHAODA `anomaly_score` is the *read* over the column set, the same way
+`exposure` is. The substrate that carries the study is therefore literally a CLAM
+tree of HHTL-keyed helix value members with a CHAODA read.

crates/perturbation-sim/Cargo.toml

@@ -92,3 +92,11 @@ path = "examples/explore.rs"
 [[example]]
 name = "scorecard"
 path = "examples/scorecard.rs"
+
+[[example]]
+name = "calibrate"
+path = "examples/calibrate.rs"
+
+[[example]]
+name = "hhtl_grid"
+path = "examples/hhtl_grid.rs"