docs: review Expo and LRT platform allocations

docs/gtfs-mapping-review-inventory.md

@@ -95,10 +95,10 @@ dates, trip records and literal platform evidence.
 | Jurong East EWL `EW24_A` | Source label B / C; both candidates support the direction. | Preserve the alternatives rather than infer an allocation from direction. |
 | Tanah Merah `CG_A`/`CG_B` and Changi Airport `CG2_A`/`CG2_B` | Literal D / E and A / B groups. | Review complete alternative sets and resulting occurrence semantics. |
 | Ang Mo Kio `NS16_A`/`NS16_B` | Source groups A / C and B / D. The 29 October correction adds physical C/D with `not_in_service` status following review of their depot role. | Complete physical groups still need every member to pass passenger boarding checks; see the [AMK correction and regenerated diagnostics](gtfs-amk-platform-evidence-2026-10-28.md). Choosing A/B alone would hide the source alternatives. |
-| Expo airport services | Six patterns conflict with current literal-label/service scopes, overlapping grouped endpoint blockers. | Resolve dated allocation evidence or a source correction; grouped-platform support alone does not resolve these conflicts. |
+| Expo airport services | Six patterns conflict with current literal-label/service scopes. Dated photographs support canonical A/east and B/west. | Follow the [allocation evidence review](gtfs-platform-allocation-review-2026-10-31.md); investigate the source's reverse labels without matching by suffix. |
 | CCL | Revision/code gaps, multiple candidates and conflicts with undated current platform scopes, including unresolved dates in all 119 patterns. | Review these causes separately; do not choose the latest revision or use current records as historical proof. |
-| Sengkang East | Six patterns conflict at the five intermediate stations. West patterns have supported proposals. | Verify intermediate-station labels and service allocations without applying town-centre evidence to other stations. |
-| Bukit Panjang LRT | All six patterns conflict at both Choa Chu Kang occurrences; their ordered station geometry is valid. | Verify physical allocations and preserve both endpoint occurrences. Dropping the return occurrence would leave the initial conflict. |
+| Sengkang East | Six patterns conflict at five intermediate stations; eight of ten canonical bindings trace to two observation roots. Maintainer-supplied Bakau and Rumbia boards support both roots' existing directions. | Retain current bindings and recorded inferences, with the image limits and unresolved source conflicts in the [allocation evidence review](gtfs-platform-allocation-review-2026-10-31.md). |
+| Bukit Panjang LRT | Six patterns conflict at both Choa Chu Kang occurrences. Historical signage distinguishes boarding platforms 1/2 from platform 4. | Review current per-platform service and boarding allocations using the [station evidence](gtfs-platform-allocation-review-2026-10-31.md); preserve both endpoint occurrences. |
 | Marina South Pier | Four geometric NSL patterns conflict at their terminal physical platform; two no-geometry skip paths also have that endpoint concern. | Distinguish terminal arrival and source occurrence rules before changing service scope or diagnostics. |
 | Five source-path disagreements | Three NSL patterns contain skipped/excursion occurrences; two TEL patterns omit Gardens by the Bay. | Obtain service-specific evidence or a provider correction. No interpolation, reversal, deleted occurrence or implicit exclusion is supplied. |
 | Unused `LTA` agency | The source declares it but no route references it. | Record a reviewed explicit disposition rather than infer an operator alias or automatic ignore. |

docs/gtfs-platform-allocation-review-2026-10-31.md

@@ -0,0 +1,205 @@
+# Expo and LRT platform allocation review — 31 October 2026
+
+The next 18 unresolved pattern/date cells cover Expo, both Choa Chu Kang
+occurrences and Sengkang East's five intermediate stations. Photographs support
+Expo's existing canonical directions. The LRT evidence has narrower limits:
+Choa Chu Kang's signs distinguish boarding from alighting. Maintainer-supplied
+boards support Bakau's and Rumbia's existing directions; the remaining Sengkang
+East bindings retain their recorded same-line inference.
+
+This packet records evidence and the remaining decisions. It changes no source
+rows, canonical platforms, mappings, calendars or eligibility rules.
+
+## Exact affected population
+
+Regenerated against merged [PR #444](https://github.com/foldaway/mrtdown-data/pull/444),
+commit `1788292457da1f02ddaf29b4209f4ed5c178def7`, using the pinned
+[21 October capture](https://sources.mrtdown.org/captures/schedule/9a4271536f4bee1d690b10c887fcd0063875a36f3918543d2e32e21ac4dc7e4a.json)
+and [original ZIP](https://sources.mrtdown.org/blobs/schedule/9a4271536f4bee1d690b10c887fcd0063875a36f3918543d2e32e21ac4dc7e4a.zip).
+The source SHA-256 is
+`9a4271536f4bee1d690b10c887fcd0063875a36f3918543d2e32e21ac4dc7e4a`
+(2,981,645 bytes). LTA attribution and licence are in the
+[public handoff](gtfs-public-handoff.md).
+
+| Scope | Pattern/date cells | Recurring trips | Pattern/date pairs | Trip/date pairs | Conflicting occurrence indices |
+| --- | ---: | ---: | ---: | ---: | --- |
+| Expo | 6 | 675 | 730 | 80,453 | 1, between Tanah Merah and Changi Airport |
+| Choa Chu Kang | 6 | 1,248 | 730 | 159,364 | **0 and 18**, preserving both terminal visits |
+| Sengkang East | 6 | 1,491 | 730 | 180,151 | 1–5, preserving both Sengkang town-centre visits |
+
+These disjoint sets total **18 cells, 3,414 trips and 419,968 trip/date pairs**.
+All remain unresolved. Every trip's route, calendar and complete ordered stop
+list was joined back to the ZIP. Affected occurrences have ordinary/default
+pickup and drop-off, with no source instruction prohibiting boarding.
+WD contributes 261 dates, WE 52 and PH 52 over literal 2026 calendars; the
+accepted Sunday-on-public-holiday policy is still a separate implementation.
+The [cross-line appendix](gtfs-cross-line-mapping-review-2026-10-28.md#no-path-traces-and-exact-smaller-exception-cells)
+lists all 18 cell keys and representative trips.
+
+## Expo: evidence supports retaining the canonical allocation
+
+Canonical [Expo](../data/station/XPO.json) assigns A to `EWL_CG_E` toward
+Changi Airport and B to `EWL_CG_W` toward Tanah Merah. Both are directly
+maintained records, without a same-line inference. Their last review date is
+21 July 2026, not a dated observation of every supplied service.
+
+| First-hand photograph | Visible operator signage |
+| --- | --- |
+| [ZKang123 / zhenkang, 27 June 2020, platform A](https://commons.wikimedia.org/wiki/File:CG1_Expo_Platform_A.jpg) | A beside an onward route strip to Changi Airport. |
+| [ZKang123 / zhenkang, 27 June 2020, platform B](https://commons.wikimedia.org/wiki/File:CG1_Expo_Platform_B.jpg) | B beside the route strip through Tanah Merah toward the main EWL. |
+| [LN9267, 12 August 2025](https://commons.wikimedia.org/wiki/File:Expo_Station_East_West_Line_Changi_Airport_Branch_Line_platform_12-08-2025%281%29.jpg) | B and the direction Tanah Merah on the same platform-door sign; Changi Airport appears behind the origin on the route strip. |
+
+The October source does the reverse: eastbound `CG1_A` explicitly labels B,
+and westbound `CG1_B` labels A. Its parent is `CG1` in both cases. Physical
+identity follows those literal labels; the stop-ID suffix cannot replace them.
+
+**Recommendation:** retain the canonical A/east and B/west bindings and keep all
+six source conflicts visible. These observations strengthen a provider-label
+investigation; they do not prove an unchanged allocation on every 2026 date or
+authorize rewriting the source. A dated change notice or provider clarification
+would be needed to explain the reverse assertion.
+
+## Choa Chu Kang: separate service allocation from boarding infrastructure
+
+Canonical [Choa Chu Kang](../data/station/CCK.json) assigns platforms 1/3 to
+`BPLRT_B` via Petir and 2/4 to `BPLRT_A` via Senja, without boarding restrictions.
+The feed instead uses literal platform 1 at both ends of every Service A trip,
+and literal platform 2 at both ends of every Service B trip.
+
+Two photographs by ZKang123 / zhenkang, dated 17 February 2021, narrow the question:
+
+- [Platform 1](https://commons.wikimedia.org/wiki/File:BP1_Choa_Chu_Kang_LRT_Platform_1_20210217_155346.jpg)
+  displays Service B via Petir, with passengers boarding. This supports one
+  observed B/1 departure, not an exclusive or permanent B/1 allocation.
+- [Platform 4](https://commons.wikimedia.org/wiki/File:BP1_Choa_Chu_Kang_LRT_Platform_4_20210217_155442.jpg)
+  has a notice directing passengers to board at platform 1 or 2. The temporary
+  out-of-service display in the image supplies no permanent status for 2026.
+
+**Recommendation:** obtain current station-specific confirmation of which
+services can board at **each of 1 and 2**, and separately review the restrictions
+on **each of 3 and 4**. The current service-only records do not describe the
+observed boarding distinction. Do not swap A/B, grant both services to both
+boarding platforms, or mark both outer faces from one photograph alone.
+
+The initial occurrence already conflicts. An alighting-only rule at the final
+occurrence cannot resolve that initial conflict, and the feed's repeated 1/2
+must not be silently changed to 3/4. Any subsequent correction must preserve
+the complete 19-occurrence path and assess pickup/drop-off at both ends.
+
+## Sengkang East: supplied boards support the two observation roots
+
+The canonical clockwise service runs
+`SKG → CPV → RMB → BAK → KGK → RNG → SKG`; counterclockwise reverses it.
+At every intermediate station, platform 1 lists `SKLRT_E_CW` and 2 lists
+`SKLRT_E_CCW`. Eight of those ten bindings are inferred: platform 1 traces to
+directly maintained `BAK_SKLRT_1`, and platform 2 to `RMB_SKLRT_2`.
+The October feed uses literal 2 clockwise and 1 counterclockwise at all five.
+
+The inspected [Bakau photograph, 21 July 2020](https://commons.wikimedia.org/wiki/File:SE3_Bakau_Platform_1_20200721_111814_1.jpg)
+shows station identity but no readable platform-direction pairing. The
+[Rumbia photograph, 30 May 2018](https://commons.wikimedia.org/wiki/File:SE2_Rumbia_Platform_2.jpg)
+shows platform 2; its small route diagram is insufficient here to authenticate
+the complete direction and date allocation. Both photographs are by
+ZKang123 / zhenkang. Their titles alone cannot supply the missing directions.
+
+The maintainer subsequently supplied a crop identified as Bakau's platform board.
+It shows both platform headings and opposite highlighted arms of the East Loop.
+Reading those arms against the canonical station order supports platform 1 via
+Kangkar and Ranggung (`SKLRT_E_CW`), and platform 2 via Rumbia and Compassvale
+(`SKLRT_E_CCW`). This is an interpretation of the route highlighting; the small
+station text is blurred. The crop's capture date, author and original URL were
+not supplied, and its station identity comes from the maintainer's description.
+It supports retaining Bakau's current binding.
+
+The second supplied crop, identified as Rumbia's board, supports platform 1
+via Bakau, Kangkar and Ranggung (`SKLRT_E_CW`), and platform 2 via Compassvale
+(`SKLRT_E_CCW`). Platform 1's highlighted path continues around the longer arm;
+platform 2 highlights the shorter arm through Compassvale to Sengkang. This is
+also a route-diagram interpretation, with blurred text and part of platform 2's
+board obscured. Its capture date, author and original URL were not supplied;
+station identity again comes from the maintainer.
+
+**Recommendation:** retain both stations' existing direction bindings. These
+boards support the two directly maintained observation roots; the eight inferred
+bindings retain their existing provenance, including those at Compassvale,
+Kangkar and Ranggung. The supplied images do not independently verify those
+three stations or every 2026 date. Keep all five source conflicts and both
+town-centre visits; no source label swap is authorized by this evidence review.
+
+## Maintainer input — 31 October
+
+| Scope | Supplied review input | Effect on this review |
+| --- | --- | --- |
+| Choa Chu Kang | “I believe so” in response to both services boarding at either 1/2 and 3/4 being alighting-only. | Tentative recollection of that arrangement, retained for a later canonical correction review. It does not establish an exclusive service allocation or historical validity over all 2026 dates. |
+| Bakau | “Platform 1 is East Loop,” followed by a crop identified as Bakau's platform board. | The opposite highlighted route arms support the existing 1/clockwise and 2/counterclockwise bindings, with the image limits above. |
+| Rumbia | “same, Platform 1 is East Loop,” followed by a crop identified as Rumbia's platform board. | The highlighted routes support 1 via Bakau (clockwise) and 2 via Compassvale (counterclockwise), retaining the existing platform 2 observation root and the image limits above. |
+| Expo | “I think so?” about A toward Changi Airport and B toward Tanah Merah. | Tentative agreement, alongside the independent photographs above; the source conflict remains unresolved. |
+
+These are maintainer statements, with their expressed uncertainty preserved.
+Both supplied boards answer the requested direction clarification at their
+identified stations; the other bindings remain recorded inferences.
+No canonical allocation or digest-bound mapping approval is supplied here.
+
+## Reproduction and evidence integrity
+
+All seven cited photographs were downloaded and their pixels inspected.
+Dates are author-recorded metadata; they are observations, not operator-issued
+all-year allocation statements. Each linked file page identifies its author
+and CC BY-SA 4.0 licence. Originals and inspection crops remain outside Git.
+The separately supplied Bakau board crop is identified by SHA-256
+`2bd25076506e474759caf36b5d2b4d20ef2cccbcf8fe1a4af6134dece972abab`
+(2,098,446 bytes), filename
+`codex-clipboard-8d0224f1-01f0-410e-b4eb-a4eaded682a8.png`.
+The supplied Rumbia board crop hashes to
+`2cae51cd9a544537e6442eb024fa1de8bbd14af53d543a5f3da0cd439af194ec`
+(1,740,092 bytes), filename
+`codex-clipboard-5b15a8e2-6671-4b14-9032-cd0416dcd10b.png`.
+Both crops have unspecified licences; they remain outside Git and are not republished.
+Newer Choa Chu Kang/Rumbia originals returned HTTP 429 and were not inspected;
+no conclusion relies on their captions or thumbnails.
+
+| Original | SHA-256 | Bytes |
+| --- | --- | ---: |
+| Expo A, 2020 | `ed1ea9b145ae8079606a6679810a386ee74f963d2f1286089a7fb599b0b13675` | 3,952,173 |
+| Expo B, 2020 | `e003e12157d575534f39fce34eccb68fcd1fb24117dda2ccb80b5707966115e4` | 3,914,833 |
+| Expo B, 2025 | `da38e813958b4141a297111ea3d0461776d8e16c60afdef9ec18b81a4b66d800` | 11,961,759 |
+| Choa Chu Kang 1, 2021 | `77d72d0d04bd72afcc6f4e0a5d4cda0127c856eebcc88b6c533d2820496d3801` | 3,774,631 |
+| Choa Chu Kang 4, 2021 | `fa6489a08eb3954f1ba43b615a19616fd9a8e9a584349d4604194f16853f7e95` | 3,380,155 |
+| Bakau, 2020 | `39077e92f3a4325af37a09b29d4a2d26de8d789c2bb46765d8a551c1bd825953` | 4,421,003 |
+| Rumbia 2, 2018 | `99a953f3411bacb52240ef2c357fd32f824f45de04326b810bc1e35ef6865824` | 5,499,307 |
+
+With the pinned receipt/ZIP downloaded as in the [NEL instructions](gtfs-nel-mapping-review-2026-10-28.md),
+build the CLI and regenerate using committed data at the merge above:
+
+```sh
+npm run build:cli
+TZ=UTC node scripts/draft-gtfs-scoped-review.mjs \
+  /tmp/nel-review-input-20261028/schedule.zip \
+  /tmp/nel-review-input-20261028/capture.json \
+  NSL,EWL,NEL,CCL,DTL,TEL,BPLRT,SKLRT,PGLRT \
+  /tmp/platform-allocation-review-20261031 \
+  --canonical-commit 1788292457da1f02ddaf29b4209f4ed5c178def7 \
+  --platform-groups --blockers --assignments
+```
+
+Select blocker patterns containing `CG1_A/B`, `BP1_A/B` or `SE1_A/B` in
+`stopIds`; join `dateSetId` to the full draft's `dateSets`. Join **every**
+`tripIds` member to source trips and stop-times, sorting occurrences by numeric
+`stop_sequence`. This reproduces the population above without selecting only
+convenient examples. Generated packets remain outside Git.
+
+| Regenerated identity | SHA-256 |
+| --- | --- |
+| Topology projection 1 | `02ebd83efd71e0facbe7b34ecb374ec0fbec62b581db56417bfc4ede88b3365f` |
+| Normalized full draft | `9b5219384f880c13984d74ad71d4b2a8a8dc6dba33cb6f9a4275837bb4c37aa4` |
+| Blocker bytes, 4,244,923 bytes | `920e426a56f04f25c0b51c53ea336238204d829542d3866aed9ce03214172ee2` |
+| Normalized assignments | `ad4756a9ba792ac3afeb0af55efaed2eff3387e7613413687db73d2989a351b9` |
+
+Full accounting and assignment validation pass. There are still **189 unresolved
+decisions and 990,143 unassigned trip/date pairs**. Mapping review and generation
+eligibility remain false. This evidence review recovers no departures; canonical
+corrections, occurrence semantics, holiday overrides and publication need their
+own reviewed changes.
+
+Validation: CLI build, lint, documentation links and 74 existing grouped-platform,
+grouped-review and blocker tests passed with Node 24 and `TZ=UTC`.