RandomCodeSpace · aksOps · Jul 30, 2026 · Jul 29, 2026 · Jul 29, 2026 · Jul 29, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -13,6 +13,24 @@ Decisions and their rationale — including what was rejected — live in
 
 ### Changed
 
+- **JavaScript and TypeScript now emit dotted member reads as
+  `FieldAccess`.** The outermost static selector is emitted once: namespace
+  imports can reach exported definitions, lexical locals remain
+  `LocalBinding`, and expression/receiver roots retain the existing honest
+  `NeedsExpressionType` / `NeedsReceiverType` reasons. Call and construction
+  targets keep their more precise kinds, plain assignment targets are writes
+  rather than reads, and computed `a[b]` access still emits no invented name.
+  Type-position members also keep their existing `TypeUse` rows instead of
+  gaining a duplicate field-access row.
+
+  The Fastify denominator share is measured at **14.2% -> 14.3%**
+  (4,435 / 31,136 -> 5,395 / 37,700); D adds 960 denominator occurrences and
+  6,564 total occurrences. The card's >=40% figure is a D+J program forecast:
+  D's complete 6,565-site syntax allocation has a 29.2% arithmetic ceiling
+  even if every site were incorrectly placed in the denominator. Stream J owns
+  the separate 21,542-local-binding resolution work that the forecast includes.
+  This change preserves the honest `LocalBinding` partition instead of gaming
+  the percentage with false or duplicate rows.
 - Java now retries typed applicability only for a legacy
   `AmbiguousOverload` whose complete file-local argument vector is known.
   Legacy resolved, external, local, and other unresolved rows keep their

diff --git a/README.md b/README.md
@@ -217,7 +217,7 @@ reasons are where that shows. The buckets that need a type environment —
 `NeedsExpressionType`, `NeedsReceiverType`, `NeedsTypeInference`,
 `AmbiguousOverload`, `UnindexedSupertype`, `DynamicDispatch` — are the majority
 of what tier 1 leaves unlinked on nine of the ten real corpora, from 51.9% on
-`flask` to 100.0% on both Go corpora. The tenth is `vue-core` at 42.1%, where
+`flask` to 100.0% on both Go corpora. The tenth is `vue-core` at 46.6%, where
 the largest single bucket is instead the 14,618 names a declared test runner
 injects into the global scope.
 
@@ -255,23 +255,20 @@ rate without the share invites reading the first number as the second.
 | Java | `commons-lang` `598dfc1` | 38,222 | 63,385 | 15,162 | 12,274 | **75.7%** | 39.1% |
 | Java | `gson` `3ff35d6` | 13,306 | 16,737 | 6,706 | 5,684 | **70.1%** | 44.8% |
 | Java | `probes` `e4dc880` | 13 | 7 | 1 | 1 | **92.9%** † | 63.6% |
-| JavaScript | `fastify` `94bcbcc` | 2,795 | 5,159 | 21,542 | 1,640 | **63.0%** | 14.2% |
-| JavaScript | `express` `dbac741` | 2,267 | 702 | 3,039 | 5,552 | **29.0%** | 67.6% |
-| JavaScript | `probes` `e4dc880` | 6 | 0 | 1 | 2 | **75.0%** † | 88.9% |
-| TypeScript | `vue-core` `fa2885d` | 26,297 | 3,694 | 9,564 | 27,945 | **48.5%** | 80.4% |
-| TypeScript | `zod` `1fb56a5` | 17,080 | 1,952 | 8,143 | 19,784 | **46.3%** | 78.5% |
-| TypeScript | `probes` `e4dc880` | 12 | 0 | 1 | 3 | **80.0%** † | 93.8% |
+| JavaScript | `fastify` `00708f3fae0cf8134e718fc8ec35e436d4f481b7` | 2,865 | 5,348 | 26,957 | 2,530 | **53.1%** | 14.3% |
+| JavaScript | `express` `00708f3fae0cf8134e718fc8ec35e436d4f481b7` | 2,272 | 748 | 3,642 | 5,910 | **27.8%** | 65.1% |
+| JavaScript | `probes` `00708f3fae0cf8134e718fc8ec35e436d4f481b7` | 6 | 0 | 1 | 2 | **75.0%** † | 88.9% |
+| TypeScript | `vue-core` `00708f3fae0cf8134e718fc8ec35e436d4f481b7` | 29,763 | 3,817 | 20,551 | 30,557 | **49.3%** | 71.2% |
+| TypeScript | `zod` `00708f3fae0cf8134e718fc8ec35e436d4f481b7` | 17,530 | 2,059 | 14,456 | 21,773 | **44.6%** | 70.4% |
+| TypeScript | `probes` `00708f3fae0cf8134e718fc8ec35e436d4f481b7` | 12 | 0 | 1 | 3 | **80.0%** † | 93.8% |
 | Python | `django` `af67523` | 19,103 | 13,326 | 8,405 | 6,185 | **75.5%** | 53.8% |
 | Python | `flask` `22d9247` | 1,185 | 2,317 | 2,146 | 877 | **57.5%** | 31.6% |
 | Python | `probes` `e4dc880` | 5 | 0 | 2 | 1 | **83.3%** † | 75.0% |
 
-Ten real corpora and five synthetic probe pins. Since the last published table
-four rows are new, three rates moved, one row changed without its rate moving,
-and every other cell is byte-identical. A rate moves for more than one reason
-and only one of those reasons is "it got better", so each movement below is
-attributed per row in [`CHANGELOG.md`](CHANGELOG.md) — by a whole-row join
-against a binary built from the previous commit — rather than inferred from
-these totals.
+Ten real corpora and five synthetic probe pins. The table is a direct rendering
+of the committed baselines. Its six ECMA rows include the static member-read
+capability; their movement is attributed by a whole-row join in the Stream D
+decision record rather than inferred from these totals.
 
 1. **Both Go rates fell, and that is coverage.** The extractor now emits the
    two member sites it never had: a selector *read* (`pkg.Name`, `t.field`)
@@ -297,16 +294,19 @@ these totals.
    the `local_binding` column in [`baselines/`](baselines), which is gated for
    drift exactly so this cannot be done quietly.
 
-`express` is the row that changed without its rate moving: `XMLHttpRequest` was
+`express` previously changed without its rate moving: `XMLHttpRequest` was
 missing from the host global list while `WebSocket`, `AbortController` and
-`fetch` were all in it, so `external` went 701 → 702 and the rate reads 28.99%
-on both sides. The same change reclassified the 1,728 express and 13,833
-vue-core references that a declared test runner injects — `it`, `describe`,
-`expect` — out of `NoMatchingDefinition`, a bucket whose contract is that it
-means *arthron's* bug. They became `Unresolved(UnknownPackage)` rather than
-`External` deliberately: `Unresolved` keeps them inside both of the rate's
-terms, where `External` would take them out of both and raise the rate without
-linking anything. No gated number moved.
+`fetch` were all in it, so `external` went 701 → 702 and the rate read 28.99%
+on both sides. Its current baseline is 2,272 resolved, 748 external, 3,642
+local-binding and 5,910 unresolved: a 27.8% rate after static member reads
+entered the denominator. The same earlier change reclassified the 1,728
+express and 13,833 vue-core references that a declared test runner injects —
+`it`, `describe`, `expect` — out of `NoMatchingDefinition`, a bucket whose
+contract is that it means *arthron's* bug. They became
+`Unresolved(UnknownPackage)` rather than `External` deliberately:
+`Unresolved` keeps them inside both of the rate's terms, where `External` would
+take them out of both and raise the rate without linking anything. No gated
+number moved in that earlier change.
 
 What the third change costs is *edges*, and the counts are in
 [`CHANGELOG.md`](CHANGELOG.md): 5,374 commons-lang and 3,189 gson occurrences
@@ -395,7 +395,7 @@ in either.
 
 **The rate's denominator is published beside it, every time.** Excluding those
 two is correct and it also makes the denominator smaller than the surface the
-scan read — on `fastify` the rate covers 14.2% of the references JavaScript
+scan read — on `fastify` the rate covers 14.3% of the references JavaScript
 emitted. A high rate over a small share is a real measurement and a partial
 one, and the only way to read it as the first without the second is for the
 second not to be there. So `arthron scan` prints it under every language line,

diff --git a/baselines/javascript-express.toml b/baselines/javascript-express.toml
@@ -10,9 +10,9 @@
 # drift in either fails the gate and must be re-based deliberately.
 format = 1
 corpus = "corpus/javascript/express"
-commit = "dbac741"
+commit = "00708f3fae0cf8134e718fc8ec35e436d4f481b7"
 language = "javascript"
-resolved = 2267
-external = 702
-local_binding = 3039
-unresolved = 5552
+resolved = 2272
+external = 748
+local_binding = 3642
+unresolved = 5910
diff --git a/baselines/javascript-fastify.toml b/baselines/javascript-fastify.toml
@@ -10,9 +10,9 @@
 # drift in either fails the gate and must be re-based deliberately.
 format = 1
 corpus = "corpus/javascript/fastify"
-commit = "94bcbcc"
+commit = "00708f3fae0cf8134e718fc8ec35e436d4f481b7"
 language = "javascript"
-resolved = 2795
-external = 5159
-local_binding = 21542
-unresolved = 1640
+resolved = 2865
+external = 5348
+local_binding = 26957
+unresolved = 2530
diff --git a/baselines/javascript-probes.toml b/baselines/javascript-probes.toml
@@ -10,7 +10,7 @@
 # drift in either fails the gate and must be re-based deliberately.
 format = 1
 corpus = "corpus/javascript/probes"
-commit = "e4dc880"
+commit = "00708f3fae0cf8134e718fc8ec35e436d4f481b7"
 language = "javascript"
 resolved = 6
 external = 0

diff --git a/baselines/typescript-probes.toml b/baselines/typescript-probes.toml
@@ -10,7 +10,7 @@
 # drift in either fails the gate and must be re-based deliberately.
 format = 1
 corpus = "corpus/typescript/probes"
-commit = "e4dc880"
+commit = "00708f3fae0cf8134e718fc8ec35e436d4f481b7"
 language = "typescript"
 resolved = 12
 external = 0

diff --git a/baselines/typescript-vue-core.toml b/baselines/typescript-vue-core.toml
@@ -10,9 +10,9 @@
 # drift in either fails the gate and must be re-based deliberately.
 format = 1
 corpus = "corpus/typescript/vue-core"
-commit = "fa2885d"
+commit = "00708f3fae0cf8134e718fc8ec35e436d4f481b7"
 language = "typescript"
-resolved = 26297
-external = 3694
-local_binding = 9564
-unresolved = 27945
+resolved = 29763
+external = 3817
+local_binding = 20551
+unresolved = 30557
diff --git a/baselines/typescript-zod.toml b/baselines/typescript-zod.toml
@@ -10,9 +10,9 @@
 # drift in either fails the gate and must be re-based deliberately.
 format = 1
 corpus = "corpus/typescript/zod"
-commit = "1fb56a5"
+commit = "00708f3fae0cf8134e718fc8ec35e436d4f481b7"
 language = "typescript"
-resolved = 17080
-external = 1952
-local_binding = 8143
-unresolved = 19784
+resolved = 17530
+external = 2059
+local_binding = 14456
+unresolved = 21773
diff --git a/docs/decisions.md b/docs/decisions.md
@@ -371,6 +371,133 @@ store language semantics.* The resolver remains the only authority; the store
 owns persistence and tallying of the verdict.
 
 ---
+## 2026-07-29 — static ECMA member reads enter the graph once
+
+The JavaScript and TypeScript extractors emitted calls, imports, type uses and
+heritage, but omitted a dotted property when the property was read as a value.
+That was an invisible denominator, not an honest miss: `imported.member` could
+name an exported repository definition and there was no row to resolve.
+
+Measured with the pinned ast-grep 0.44.1 parser and the product's own file
+walker over all four ECMA corpora, the syntax cut is **33,731** outermost
+non-call/non-construction member reads after excluding plain assignment
+targets:
+
+| corpus | import binding | `this` / `super` | name root | complex expression | total |
+|---|---:|---:|---:|---:|---:|
+| fastify | 171 | 221 | 5,631 | 542 | 6,565 |
+| express | 0 | 253 | 716 | 43 | 1,012 |
+| vue-core | 3,045 | 741 | 11,918 | 1,498 | 17,202 |
+| zod | 570 | 212 | 6,766 | 1,404 | 8,952 |
+| **total** | **3,786** | **1,427** | **25,031** | **3,487** | **33,731** |
+
+`name root` is syntactic, not a claim that every such name is local:
+23,359 are bound by a lexical environment that is not a node, 544 are module
+bindings that are nodes, and 1,128 are unbound or global names. Keeping those
+classes separate is what prevents the large local term from hiding module
+edges.
+
+The syntax count is not the new-row count. **106 TypeScript sites already have
+the more precise `TypeUse` kind** (13 in vue-core, 93 in zod). Emitting them
+again as `FieldAccess` produced a second, false `NeedsExpressionType` beside a
+resolved import-type row in the fixture suite. They are excluded, not lost.
+The resulting extraction delta is **33,625 `FieldAccess` occurrences**:
+6,565 fastify, 1,012 express, 17,189 vue-core and 8,859 zod. Store
+deduplication means these occurrence counts are not predictions of committed
+row movement; the corpus join and re-base own that measurement.
+
+The Fastify whole-row join measures **2,865 resolved, 5,348 external, 26,957
+local-binding and 2,530 unresolved** occurrences: a rate denominator of
+**5,395 of 37,700 references (14.3%)**, from **4,435 of 31,136 (14.2%)**.
+The member-read change adds 960 denominator occurrences but 6,564 total
+occurrences, and its rate is consequently 63.0% -> 53.1% (2,795 / 4,435 ->
+2,865 / 5,395). This is an accounting result, not a reason to relabel local
+names: the shape audit finds 1,889 appeared lexical name-root rows marked
+`locally_bound`, which policy correctly keeps outside both terms.
+
+The card's Fastify >=40% denominator-share expectation is a repository
+program forecast, not a Stream D-only acceptance. D owns only 6,565 authorized
+Fastify syntax occurrences; even the invalidly optimistic allocation that put
+every one in `Resolved + Unresolved` could reach only
+**(4,435 + 6,565) / (31,136 + 6,565) = 11,000 / 37,701 = 29.2%**. The later
+Stream J owns resolving the existing 21,542 local bindings that the forecast
+also needs. The measured D result is 14.3%; it must not be inflated by
+reclassifying locals, emitting computed or write-only names, or duplicating
+nested, call, construction, or type-position rows.
+
+The final pre-rebase join uses the complete reference-row key
+(`file + kind + declaration space + enclosing FQN + raw target + argc +
+locally_bound`) against fresh release builds of `1ea3c87` and `00708f3`.
+Every pre-existing row holds byte-for-byte in all six ECMA corpora: zero
+vanished rows, zero changed outcomes or occurrence counts, zero duplicate
+keys. Every appeared row is the new `FieldAccess` kind. Rows and occurrences
+are separate units:
+
+| corpus | appeared rows / occurrences | in-repo resolved rows / occurrences | new `External` rows / occurrences | `LocalBinding` rows / occurrences | other unresolved rows / occurrences |
+|---|---:|---:|---:|---:|---:|
+| fastify | 2,452 / 6,564 | 41 / 70 | 85 / 189 | 1,906 / 5,415 | 420 / 890 |
+| express | 489 / 1,012 | 5 / 5 | 38 / 46 | 306 / 603 | 140 / 358 |
+| vue-core | 8,589 / 17,188 | 1,883 / 3,466 | 81 / 123 | 5,186 / 10,987 | 1,439 / 2,612 |
+| zod | 5,173 / 8,859 | 346 / 450 | 50 / 107 | 3,218 / 6,313 | 1,559 / 1,989 |
+| JavaScript probes | 0 / 0 | 0 / 0 | 0 / 0 | 0 / 0 | 0 / 0 |
+| TypeScript probes | 0 / 0 | 0 / 0 | 0 / 0 | 0 / 0 | 0 / 0 |
+
+The unresolved side introduces no taxonomy. Its full per-reason attribution,
+again as rows / occurrences, is:
+
+| corpus | per-reason appeared rows / occurrences |
+|---|---|
+| fastify | `NoMatchingDefinition` 32 / 100; `NeedsTypeInference` 38 / 60; `LocalBinding` 1,906 / 5,415; `NeedsReceiverType` 66 / 189; `NeedsExpressionType` 284 / 541 |
+| express | `UnknownPackage` 1 / 1; `NeedsTypeInference` 56 / 61; `LocalBinding` 306 / 603; `NeedsReceiverType` 42 / 253; `NeedsExpressionType` 41 / 43 |
+| vue-core | `NoMatchingDefinition` 101 / 134; `NeedsTypeInference` 250 / 619; `LocalBinding` 5,186 / 10,987; `NeedsReceiverType` 153 / 359; `NeedsExpressionType` 930 / 1,495; `UnindexedSupertype` 5 / 5 |
+| zod | `NoMatchingDefinition` 137 / 179; `NeedsTypeInference` 289 / 392; `LocalBinding` 3,218 / 6,313; `NeedsReceiverType` 6 / 12; `NeedsExpressionType` 1,125 / 1,404; `UnindexedSupertype` 2 / 2 |
+| JavaScript / TypeScript probes | none |
+
+The six baselines were written only by `arthron gate --rebase --commit
+00708f3fae0cf8134e718fc8ec35e436d4f481b7`:
+
+| corpus | before `(resolved, external, local, unresolved)` | after `(resolved, external, local, unresolved)` |
+|---|---|---|
+| fastify | 2,795 / 5,159 / 21,542 / 1,640 | 2,865 / 5,348 / 26,957 / 2,530 |
+| express | 2,267 / 702 / 3,039 / 5,552 | 2,272 / 748 / 3,642 / 5,910 |
+| JavaScript probes | 6 / 0 / 1 / 2 | 6 / 0 / 1 / 2 |
+| vue-core | 26,297 / 3,694 / 9,564 / 27,945 | 29,763 / 3,817 / 20,551 / 30,557 |
+| zod | 17,080 / 1,952 / 8,143 / 19,784 | 17,530 / 2,059 / 14,456 / 21,773 |
+| TypeScript probes | 12 / 0 / 1 / 3 | 12 / 0 / 1 / 3 |
+
+Target pins were written only by `arthron pin --write` from the same binary.
+Old pins held with zero vanished and zero moved targets: fastify
+1,045 held / 41 appeared, express 496 / 5, vue-core 10,573 / 1,883,
+zod 10,289 / 346, JavaScript probes 6 / 0, and TypeScript probes 12 / 0.
+A fresh comparison against the generated pins then held every row with zero
+appeared, vanished, or moved.
+
+**Decided: emit the outermost static `member_expression` once as
+`FieldAccess`.** It reuses the same target roots, binding verdict and resolver
+paths as call position: an imported namespace member can resolve, a lexical
+name remains `LocalBinding`, `this` / `super` retain receiver lookup, and an
+expression root remains `NeedsExpressionType`. No unresolved reason changes.
+
+**A plain assignment target (`a.b = x`) is a write without a read and is not
+emitted by this read rule.** An augmented assignment (`a.b += x`) reads the
+old value and is emitted. A member used as the callee of a call or constructor
+keeps its `Call` / `New` row and is not duplicated as a field access.
+
+*Rejected: emitting nested selectors separately.* `a.b.c` is one read target;
+an extra `a.b` row double-counts syntax that exists only as the outer
+expression's operand.
+
+*Rejected: emitting computed `a[b]` access.* The key is a runtime value. Using
+its source text as a member name would invent a target.
+
+*Rejected: emitting type-position members again.* Their existing `TypeUse` row
+is both more precise and already in the measurement.
+
+*Rejected: changing outcomes to satisfy the D-only percentage.* `LocalBinding`
+means a lexical name is not a node, and the extractor's shape tests separately
+guard its partition from imported, expression, receiver, computed, write-only
+and call/new sites. A percentage is not authority to create a wrong edge or a
+second reference.
 
 ## 2026-07-29 — the walk keeps a file's declarations and forgets its references