Sam Gammon (sgammon) · Sep 30, 2026 · Sep 25, 2026 · Sep 25, 2026 · Sep 25, 2026 · Sep 25, 2026
diff --git a/.gitignore b/.gitignore
@@ -9,3 +9,4 @@ artifacts/
 expectations/jdk-jtreg.reference.*.json
 !expectations/jdk-jtreg.reference.linux-x64.json
 diagnostics/
+__pycache__/
diff --git a/.gitmodules b/.gitmodules
@@ -7,7 +7,7 @@
 [submodule "suites/cpython"]
 	path = suites/cpython
 	url = https://github.com/python/cpython.git
-	branch = 3.12
+	branch = 3.13
 	shallow = true
 [submodule "suites/openjdk"]
 	path = suites/openjdk

diff --git a/BRIEFING-cpython-core.md b/BRIEFING-cpython-core.md
@@ -37,6 +37,8 @@ bun run setup                       # once: deps + submodules
 bun run testsuite --elide nightly --suite cpython-core --include 'test_json' --log
 # one module under investigation:
 bun run testsuite --elide nightly --suite cpython-core --include 'test_builtin' --log
+# by name, any module/case whose id contains 'time' (case-insensitive glob; 'test_ast.*literal_eval*' picks cases):
+bun run testsuite --elide nightly --log --filter 'cpython-core:*time*'
 # full suite (~90s):
 bun run testsuite --elide nightly --suite cpython-core
 # after fixes land in an Elide nightly, re-baseline:
@@ -96,8 +98,9 @@ Start with `test_builtin` (17), `test_descr`/`test_dict` (12), then
 
 - Runtime fixes happen in the Elide repo (GraalPy-based Python engine); this
   repo pins nightlies and measures.
-- 103 skips are already classified in `expectations/cpython-core.toml` and
-  excluded from the denominator; keep skip entries commented.
+- 103 skips are already classified in `expectations/cpython-core.toml`; keep
+  skip entries commented. Skips still count against the overall pass rate, but
+  not against the "vs expectations" rate.
 - The pass rate is already 95.9% — the value here is *classification honesty*
   (separating defects from platform limits) more than raw number movement.
 - After a fix lands: full run → check `changes.md` → `--ratchet` → commit.
diff --git a/BRIEFING-javac-jtreg.md b/BRIEFING-javac-jtreg.md
@@ -37,6 +37,8 @@ bun run setup                       # once: deps + submodules + sparse checkout
 bun run testsuite --elide nightly --suite javac-jtreg --include 'tools/javac/IllDefinedOrderOfInit.java' --log
 # one failing test with full output:
 bun run testsuite --elide nightly --suite javac-jtreg --include 'tools/javac/T6411379.java' --log --show-failure-output
+# by name, any test whose path contains 'lambda' (case-insensitive glob, see docs/running.md 'Targeted runs'):
+bun run testsuite --elide nightly --log --filter 'javac-jtreg:*lambda*'
 # full suite (~12 min):
 bun run testsuite --elide nightly --suite javac-jtreg
 # after fixes land in an Elide nightly, re-baseline:

diff --git a/BRIEFING-node-api.md b/BRIEFING-node-api.md
@@ -38,6 +38,8 @@ bun run setup                       # once: deps + submodules + sparse checkouts
 bun run testsuite --elide nightly --suite node-api --include 'test/parallel/test-path.js' --log
 # a target group:
 bun run testsuite --elide nightly --suite node-api --include 'test/parallel/test-buffer*.js' --log
+# by name, any file whose path contains 'timers' (case-insensitive glob, see docs/running.md 'Targeted runs'):
+bun run testsuite --elide nightly --log --filter 'node-api:*timers*'
 # full suite (~6 min):
 bun run testsuite --elide nightly --suite node-api
 # after fixes land in an Elide nightly, re-baseline:

diff --git a/BRIEFING-test262.md b/BRIEFING-test262.md
@@ -33,6 +33,8 @@ Counts are test executions (a file can run in both default and strict mode).
 bun run setup                       # once: deps + submodules
 # quick slice (seconds):
 bun run testsuite --elide nightly --suite test262 --include 'test/built-ins/Error/prototype/stack/**/*.js' --log
+# by name, any file whose path contains 'getTime' (case-insensitive glob, see docs/running.md 'Targeted runs'):
+bun run testsuite --elide nightly --log --filter 'test262:*gettime*'
 # full suite (~20 min):
 bun run testsuite --elide nightly --suite test262
 # after fixes land in an Elide nightly, re-baseline:
@@ -95,8 +97,10 @@ Signature view of the top clusters:
 
 - Fixes happen in the Elide runtime (GraalJS-based); this repo pins nightlies
   and measures. `--elide nightly` picks up a fresh build.
-- 10,587 skips are intentional (see `expectations/test262.toml`) and excluded
-  from the pass-rate denominator.
+- 10,587 skips are intentional (see `expectations/test262.toml`). They still
+  count against the overall pass rate (the headline is over every selected
+  test), but they are expected skips, so they do not count against the
+  "vs expectations" rate.
 - Some failing features are proposal-stage (immutable ArrayBuffer, source-
   phase imports). If Elide won't ship them yet, classify as `[skip]` with a
   comment rather than leaving them as permanent failures.

diff --git a/BRIEFING-wpt-wintertc.md b/BRIEFING-wpt-wintertc.md
@@ -36,6 +36,8 @@ bun run setup                       # once: deps + submodules + sparse checkouts
 bun run testsuite --elide nightly --suite wpt-wintertc --include 'url/urlsearchparams-constructor.any.js' --log
 # a target area:
 bun run testsuite --elide nightly --suite wpt-wintertc --include 'encoding/textdecoder-fatal-single-byte.any.js' --log
+# by name, any file whose path contains 'encodeInto' (case-insensitive glob, see docs/running.md 'Targeted runs'):
+bun run testsuite --elide nightly --log --filter 'wpt-wintertc:*encodeinto*'
 # full suite (~12s of harness runtime inside Docker):
 bun run testsuite --elide nightly --suite wpt-wintertc --threads 8
 # after fixes land in an Elide nightly, re-baseline:
@@ -106,8 +108,9 @@ machine).
 5. **Park / skip candidates**: `fetch/fetch-later` (136 — browser-only,
    needs window/iframe/BFCache), `.window.js` tests requiring `document`/
    `XMLHttpRequest` (~90). Consider adding these to `[skip]` in
-   `expectations/wpt-wintertc.toml` with a comment, so the pass-rate
-   denominator reflects reachable tests.
+   `expectations/wpt-wintertc.toml` with a comment. Note this only cleans up
+   the "vs expectations" rate; the overall pass rate keeps skipped tests in
+   its denominator on purpose.
 
 Items 1–3 alone should move the suite from 75.3% to roughly **88–90%**.
 

diff --git a/README.md b/README.md
@@ -1,242 +1,66 @@
 # conformance testing
 
 This repository runs upstream conformance test suites against pinned builds of
-[Elide](https://elide.dev) and publishes per-version Markdown/JSON reports.
-**Test262** (TC39's JavaScript conformance suite) is the first suite; the
-harness is workload-neutral so more suites — and, later, benchmarks — slot in.
+[Elide](https://elide.dev) and publishes per-version reports. Suites:
 
-Next suites are added in this order:
-
-1. `wpt-wintertc` - a sparse Web Platform Tests subset for WinterTC / ECMA-429 JavaScript-facing APIs, including fetch.
-2. `cpython-core` - CPython 3.12 top-level `Lib/test/test_*` modules and packages.
-3. `javac-jtreg` - full OpenJDK langtools `tools/javac` coverage through jtreg, compiling with Elide and initially running generated programs with a regular JDK.
-4. `node-api` - sparse Node.js core JavaScript API tests, starting with `node:path` and `process`.
-
-Tracking: [Compliance Testing meta (WHIPLASH#1172)](https://github.com/elide-dev/WHIPLASH/issues/1172),
-[Test262 (WHIPLASH#1173)](https://github.com/elide-dev/WHIPLASH/issues/1173).
-
-> [!CAUTION]
-> Full enabled slices may be RED until expectations are ratcheted and failures
-> are classified.
-
-## Bali runtime compatibility
-
-This repository also supports Bali as a Docker-based Linux AMD64 target. Elide remains the default.
-
-```sh
-bun run testsuite --target bali --suite jdk-jtreg --bali-home /path/to/linux-amd64/bali
-```
+| Suite | What it covers |
+|---|---|
+| `test262` | TC39's JavaScript conformance suite |
+| `wpt-wintertc` | a sparse Web Platform Tests subset for WinterTC / ECMA-429 APIs, including fetch |
+| `cpython-core` | CPython 3.12 `Lib/test/test_*` modules and packages |
+| `javac-jtreg` | OpenJDK langtools `tools/javac` through jtreg, compiling with Elide |
+| `node-api` | sparse Node.js core JavaScript API tests |
 
-See [Bali setup and expectations](BALI.md) and [Bali compatibility history](reports/bali/index.md).
-The runtime suite (`test/jdk`) is separate from Elide's compiler suite (`javac-jtreg`).
-Runner coverage, verified passes, reference issues, and unsupported files remain separate.
+Tracking: [Compliance Testing meta (elide-dev/runtime#1172)](https://github.com/elide-dev/runtime/issues/1172),
+[Test262 (elide-dev/runtime#1173)](https://github.com/elide-dev/runtime/issues/1173).
 
 ## Compatibility
 
 <!-- compat-summary:start -->
 
 _Generated by `bun run testsuite --update-summaries` from the latest committed reports._
 
-Last updated: 2026-08-31T23:46:11.051Z
+Last updated: 2026-09-30T17:31:58.679Z
 
 ![Latest compatibility pass rates](reports/pass-rate.svg)
 
-| Suite | Version | Digest | Pass rate | Regressions |
-|---|---|---|---:|---:|
-| cpython-core | `1.4.99+8b374d579` | `4094eb2fb1ff` | 97.0% | 0 |
-| javac-jtreg | `1.4.99+8b374d579` | `4094eb2fb1ff` | 100.0% | 0 |
-| node-api | `1.4.99+8b374d579` | `4094eb2fb1ff` | 70.1% | 0 |
-| test262 | `1.4.99+8b374d579` | `4094eb2fb1ff` | 99.9% | 0 |
-| wpt-wintertc | `1.4.99+8b374d579` | `4094eb2fb1ff` | 89.9% | 0 |
+| Suite | Version | Digest | Pass rate | vs expectations | Regressions | New passes |
+|---|---|---|---:|---:|---:|---:|
+| cpython-core | `1.6.0+b8edacc63` | `7aa30129531a` | 90.8% | 100.0% | 0 | 0 |
+| javac-jtreg | `1.6.0+b8edacc63` | `7aa30129531a` | 93.0% | 100.0% | 0 | 0 |
+| node-api | `1.6.0+b8edacc63` | `7aa30129531a` | 81.8% | 100.0% | 0 | 0 |
+| test262 | `1.6.0+b8edacc63` | `7aa30129531a` | 93.0% | 100.0% | 0 | 0 |
+| wpt-wintertc | `1.6.0+b8edacc63` | `7aa30129531a` | 95.5% | 100.0% | 0 | 0 |
+
+_Pass rate_ counts every test in the selection, including skipped/suppressed ones.
+_vs expectations_ is the share of tests at or above the checked-in baseline (only regressions count against it).
 
 <!-- compat-summary:end -->
 
 ## How it works
 
 A Bun/TypeScript harness drives each suite through its best native runner and
-normalizes the results. Test262 runs via the stock `test262-harness` + `eshost`
-using a **custom eshost `elide` host** (installed through a Bun patch). Each run
-executes inside a combined Docker image (a pinned Elide + Node + Bun), so every
-`elide run` is a fast local subprocess. Results are compared against a
-checked-in **expectations baseline** — a run is green iff there are no
-regressions.
-
-## Prerequisites
+normalizes the results. Each run executes inside a Docker image with a pinned
+Elide, so every `elide run` is a fast local subprocess. Results are compared
+against a checked-in expectations baseline: new failures are regressions, tests
+that newly pass advance the floor, and a `--ratchet` run locks the gains in.
 
-- Docker (the runner pulls/builds the Elide image)
-- Bun
-- Suite submodules and Bun dependencies prepared with one command:
-  `bun run setup`
+Skips never inflate the numbers above: a skipped test still counts as a non-pass
+in _Pass rate_. Every skip has a written reason in `expectations/<suite>.toml`;
+see [why tests are skipped](docs/reports.md#why-tests-are-skipped).
 
-`bun run setup` installs root and harness dependencies, initializes the upstream
-suite submodules, and applies the sparse checkouts used by WPT and OpenJDK. To
-only prepare submodules, run `bun run setup:suites`. To check that already
-prepared submodules are populated correctly, run `bun run setup:suites --check`.
-
-## Running
+## Quick start
 
 ```bash
-# Full suite against the current Elide nightly, with a live ✅/❌ test log
+bun run setup                                        # deps + suite submodules
 bun run testsuite --elide nightly --suite test262 --log
-
-# A quick slice (finishes in seconds) — scope to any glob
-bun run testsuite --elide nightly --log --include 'test/language/types/boolean/**/*.js'
-
-# Current broad-suite smoke commands. These may be RED until expectations are
-# ratcheted, but they should emit upstream case/subtest results instead of
-# runner setup failures.
-bun run testsuite --elide nightly --suite wpt-wintertc --include 'url/urlsearchparams-constructor.any.js' --log
-bun run testsuite --elide nightly --suite cpython-core --include 'test_json' --log
-bun run testsuite --elide nightly --suite javac-jtreg --include 'tools/javac/IllDefinedOrderOfInit.java' --log
-bun run testsuite --elide nightly --suite node-api --include 'test/parallel/test-path.js' --log
-
-# Broader slices after the smoke path is stable.
-bun run testsuite --elide nightly --suite wpt-wintertc --threads 8 --log
-bun run testsuite --elide nightly --suite cpython-core --threads 8 --log
-bun run testsuite --elide nightly --suite javac-jtreg --threads 4 --log
-
-# Run every registered suite after building the harness image once.
 bun run testsuite --elide nightly --all-suites --log
-
-# Prepare selected suite submodules first, then run. Useful in fresh clones or CI.
-bun run testsuite:ready --elide nightly --all-suites --log
-
-# Run a subset with a comma-separated suite list.
-bun run testsuite --elide nightly --suite wpt-wintertc,cpython-core --log
-
-# Run multiple selected suites at once. --threads is per suite; --suite-workers
-# controls how many suite containers run concurrently.
-bun run testsuite --elide nightly --all-suites --suite-workers 2 --threads 4 --log
-
-# Update reports plus the generated README compatibility summary.
-bun run testsuite --elide nightly --all-suites --log --update-summaries
-
-# Pin a specific build: image tag, digest, or a local Elide install directory
-bun run testsuite --elide ghcr.io/elide-dev/elide@sha256:…
-bun run testsuite --elide /path/to/elide-install      # dir containing bin/elide + lib/
-```
-
-`./bin/run ...` remains as a direct executable alias for the same Bun/TypeScript
-launcher.
-
-By default, the launcher builds a total concurrency budget of
-`available CPUs * 2`. It then runs up to that many suite workers, capped by the
-number of selected suites, and divides the remaining budget into per-suite
-threads. `--threads N` overrides the per-suite adapter thread count: Test262
-forwards it to `test262-harness`, `wpt-wintertc` runs multiple WPT files at once,
-`cpython-core` splits selected modules across worker processes, and
-`javac-jtreg` forwards it to jtreg's native `-concurrency:N`. `--suite-workers N`
-overrides how many suite containers run concurrently. `CONCURRENCY_MULTIPLIER`
-or `--concurrency-multiplier N` can change the default multiplier from `2`.
-
-`--log` streams one normalized mark per completed test to stderr (`✅` pass ·
-`❌` fail · `🛑` error · `⊘` skip); the summary line goes to stdout. Add
-`--verbose` to also mirror raw runner stdout/stderr, which is useful for
-debugging harness behavior. Failure messages are persisted in reports either
-way; console printing can be controlled with `--failure-output show|hide` or the
-aliases `--show-failure-output` / `--hide-failure-output`. The exit code is
-non-zero on any regression.
-
-## Reports
-
-Committed under `reports/<elide-version>/<short-digest>/<workload>/`:
-
-| file | purpose |
-|---|---|
-| `<workload>.md` | human rollup: pass-rate, regressions, new passes |
-| `impact.md` / `impact.json` | failures clustered by root-cause signature, ranked by blast radius |
-| `changes.md` / `changes.json` | diff vs the previous run (fixed / regressed / added / removed) |
-| `summary.json` | counts + regression/new-pass ids (machine) |
-| `results.json.gz` | every test's status (machine, for cross-version diffs) |
-| `pass-rate.svg` | static pass/fail/error/skip chart for the run |
-
-`reports/index.md` (+ `index.json` + `pass-rate.svg`) is the top-level matrix
-of the latest run per suite. Machine-readable index entries point at the
-workload-scoped report directory. Published via GitHub Pages.
-
-Passing `--update-summaries` also refreshes the generated compatibility block in
-this README from the latest report index. This is intended for mainline bot runs:
-run all suites, commit the changed `reports/`, `expectations/*.ratchet.toml` if
-ratcheting, and README summary content, then push.
-
-## Expectations & the ratchet
-
-`expectations/test262.toml` is the hand-curated baseline:
-
-```toml
-[skip]   # excluded from the run, with a reason
-"test/intl402/**" = "Intl 402 not supported yet"
-[fail]   # expected failures (link an issue)
-"test/built-ins/RegExp/property-escapes/**" = "partial (…)"
 ```
 
-A run is **green iff actual matches expected**: new failures are **regressions**
-(red, fail CI); tests that newly pass are surfaced as **new passes** (advance the
-baseline). To accept the current failure set as the baseline (so only *new*
-breakage fails CI):
-
-```bash
-bun run testsuite --elide nightly --ratchet
-```
-
-This regenerates the machine-owned `expectations/<workload>.ratchet.toml`
-(exact test ids) from the current failures. `compare` treats a test as
-expected-fail if it matches a `[fail]` glob **or** is in the ratchet set;
-`[skip]` always wins.
-
-## Analysis CLI
-
-A derived SQLite database (`.harness/results.sqlite`, gitignored) is built by
-ingesting the committed `results.json.gz` files, and powers ad-hoc queries:
-
-```bash
-bun run harness/src/cli.ts db build              # (re)build the DB from committed runs
-bun run harness/src/cli.ts impact <workload> [semver] [digest]  # impact-ordered failures for a run
-bun run harness/src/cli.ts diff <workload> [A] [B]              # version diff (default: two most recent)
-bun run harness/src/cli.ts query "SELECT …"      # read-only SQL over the results
-```
-
-The committed JSON files are the source of truth and the contract for a future
-static web UI; the SQLite DB is a disposable local index.
-
-## Layout
-
-```
-suites/test262/        Test262 (git submodule)
-manifests/             curated suite slices for WPT, CPython, and jtreg
-suites/drivers/        small bridge/wrapper programs used by external suites
-expectations/          workload baselines + machine ratchets
-reports/               committed per-version reports + top-level index
-harness/               the Bun/TypeScript harness (src/, fixtures/, patches/)
-docker/                harness images (image-ref + local install dir)
-bin/run.ts             Bun/TypeScript launcher: resolve --elide → docker build → docker run
-bin/run                executable alias for bin/run.ts
-.devcontainer/         Codespaces dev environment
-.github/workflows/     nightly + on-demand compliance runs, Pages publish
-docs/superpowers/      design specs and implementation plans
-```
-
-## Development
-
-```bash
-bun install            # installs root TypeScript/Bun types for bin/run.ts
-bun run typecheck      # type-check the host launcher
-cd harness
-bun install            # applies the eshost `elide` host patch
-bun test               # unit tests
-bun run typecheck
-```
-
-Or open the repo in a GitHub Codespace / VS Code devcontainer, which provisions
-Bun, Node, Docker-in-Docker, and Elide automatically.
-
-
-
-
-
-
-
-
-
-
+## Further reading
 
+- [Running the suites](docs/running.md): prerequisites, every launcher flag, targeted runs with `--filter`, concurrency.
+- [Reports, expectations and analysis](docs/reports.md): the report layout, why tests are skipped, the ratchet, and the SQLite analysis CLI.
+- [Contributing](docs/contributing.md): local development, tests, and the repository layout.
+- Per-suite briefings: [test262](BRIEFING-test262.md), [wpt-wintertc](BRIEFING-wpt-wintertc.md), [cpython-core](BRIEFING-cpython-core.md), [javac-jtreg](BRIEFING-javac-jtreg.md), [node-api](BRIEFING-node-api.md).
+- Design specs and plans live under [docs/superpowers](docs/superpowers/).
diff --git a/bin/bali-docker.test.ts b/bin/bali-docker.test.ts
@@ -84,6 +84,10 @@ test("Bali plan validates suites from the registry and accepts --ratchet anywher
   expect(scopedArgs.slice(scopedArgs.indexOf("--include"))).toEqual(["--include", "java/lang/**,java/util/**", "--threads", "8"]);
   expect(() => baliPlan(["--bali-home", "dist", "--threads", "zero"], "/repo")).toThrow("--threads");
   expect(baliPlan(["--bali-home", "dist"], "/repo")).not.toHaveProperty("include");
+  const filtered = baliPlan(["--bali-home", "dist", "--filter", "*Date*"], "/repo");
+  expect(filtered.filter).toBe("*Date*");
+  const filteredArgs = harnessArgs(filtered, "d", { registry: "/r", repoRoot: "/w", baliHome: "/b", suiteRoot: "/s", reports: "/p", expectations: "/e" });
+  expect(filteredArgs.slice(filteredArgs.indexOf("--filter"))).toEqual(["--filter", "*Date*"]);
 });
 
 test("Docker launcher rejects macOS artifacts and a host reference JDK, and propagates build errors", async () => {