Adds 274-lyrics-index.py and its output 274-lyrics-index.csv: every sung
line resolved to its roster entry, with coordinates and timing.
98 lyric lines -> 99 rows. Columns are start/stop as mm:ss.xx and as
seconds, the line as sung, the roster's full name for it, kind, km,
lat/lon, and km_from/km_to for district headers. stop is the next
timestamp in the file, spacer lines included, so every line has one.
Four aliases cover the gap between sung and registered names: the three
power stations are sung without their operator prefix, and
モトツ橋にモトツシェルター names two facilities, so it emits two rows sharing
one timing and one coordinate. 東に続く is typed as closing, no object.
District headers resolve from the run of objects they introduce, not from
274-districts.json. Those bands are sampled along the road and so cannot
describe a district the road does not enter: 宮下町 has no band at all,
本町東 exists only as 本町東一丁目, and 千栄's band sits at 107.25-107.75
while the station it introduces is at km 104.52, 666 m off the
carriageway inside 千栄.
Object names in the roster are unique, so no other disambiguation is
needed. The script exits non-zero and names any line it cannot resolve.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two places where the sung order did not follow the carriageway.
楓駐車帯 moves up one line, between 千紫橋 and 万紅橋. It sits at km 57.08,
between their 56.60 and 57.25 — it was sung after 万紅橋.
穂別長和 splits into two headers around the 稲里 block. The road runs
長和 → 稲里 → 長和, but the two 長和 stretches had been welded into one
block sung first, so the 稲里 block then started 10.84 km behind where the
previous line left off. Now:
長和 62.36 - 62.94 オマナイ橋, 桂木橋
稲里 64.44 - 71.86 大和橋 … 穂別ダム神社
長和 72.02 - 75.28 長和トンネル … 稲里トンネル
This is the same repeated-header pattern the song already uses for
由仁町川端 / 栗山町滝下 at the opening, where the road crosses back over
the 夕張川.
Applied to both 274.md and 274.lrc; the two are now identical, 98 lines
each. Lyric order is monotonic by kilometre end to end.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Eleven files were cluttering wip/ alongside an unrelated draft. They now
sit together in their own directory.
Checked against the pipeline in novoyuuparosk-auto-wiki before moving:
- discovery is source_dir.rglob("*.md"), so nested directories are found
- the skip test is `"wip" in rel.parts`, which still excludes wip/274/
while the song is in progress
- relative links resolve as (path.parent / rel_path), i.e. against the
file's own directory, so the three docs keep linking to each other
- the lrc field resolves as abs_path.parent / lrc_rel, so 274.lrc must
and does sit beside 274.md
Both scripts locate their inputs from __file__, so they needed no change;
verified after the move that --check is clean and the page rebuilds all
132 entries.
Non-Markdown files are never read by the pipeline at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
115 timestamped lines, 14.54 s to 8:12, monotonic throughout.
Resolver check against 274-lookup.csv: of 97 lines carrying text, 75
match an object name exactly and 17 match a full address, leaving 5.
Four of those need aliases rather than corrections — the three power
stations appear without their operator prefix (川端発電所, 滝の上発電所,
北電日高発電所 against 北海道企業局 川端発電所, 北海道企業局 滝の上発電所,
北海道電力 日高発電所) and one is the merged モトツ橋にモトツシェルター.
The fifth, 東に続く, is closing text and not an object.
So the lyric-to-row index needs a four-entry alias table and nothing else.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
50.52 紅葉山駐車帯 41 m off 夕張市紅葉山
57.08 楓駐車帯 39 m off 夕張市楓
Both signposted and both absent from OpenStreetMap, which carries no
parking or rest-area features anywhere on this stretch. One of them is a
bay the author had previously recorded as unnamed. Responsible body is
北海道開発局 札幌開発建設部 岩見沢道路事務所 by jurisdiction — R274 is
指定区間 全線 and every adjacent structure registers to that office — but
no document names either bay, since parking has no published register.
Author's trimming removed 38 rows: the 丁目 grid, all five bus stops,
川端駅, and some civic furniture. That leaves 132 rows with 132 distinct
object names, so the planned lyric-to-row index is unambiguous with no
special case except モトツ橋/シェルター, which already share coordinates.
274-romaji.py learns ちゅうしゃたい and ていしゃたい so parking bays split
as Momijiyama chuushatai rather than one run-on token; --check is clean.
CSV and GeoJSON re-propagated from the Markdown.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
The file called 274-dev.md was never dev notes — it was the roster. It
now sits alongside 274-lookup.csv and 274-lookup.geojson under the same
stem, since all three carry the same table.
274-dev.md -> 274-lookup.md the roster, source of truth
274-method.md -> 274-dev.md how it was built
References updated: DOC in 274-romaji.py, the roster path in
274-transect-build.py, and the docstrings in both. 274.md now links to
the dev notes and the reference tables separately; each of the two docs
links to the other and back to the song page.
Titles follow: "274 - reference tables" and "274 - how the roster was
built".
Verified after renaming: 274-romaji.py --check reports no changes and
274-transect-build.py rebuilds 168 entries, so neither script lost its
input.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Row clicks called draw(), which rebuilt the whole tbody — so any attempt
to drag-select text in the roster was wiped on mousedown. Selection was
never blocked, just destroyed. Row clicks now call select(), which swaps
the highlight class and repaints the SVG marks in place.
Object names sit in a bare button with a dotted underline and cursor:copy;
clicking copies the name and flashes "copied" beside it. Propagation is
stopped so copying and row-selecting stay independent.
Clipboard uses navigator.clipboard with a hidden-textarea fallback, since
the vendored copy opens over file:// where the async API is often absent.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Records where each number came from, which are derived rather than
measured, and which were corrected by hand against reality.
Covers the eight sources and what each does and does not give; the
kilometre datum and why it is not official KP; the two methods used to
position structures (sequence alignment on length, watercourse crossings
as anchors) and where each is unreliable; the 250 m reverse-geocode sweep
that produces the district bands.
Two sections matter most for anyone picking this up later. "Human in the
loop" lists every correction that came from signs, satellite imagery or
knowledge held by no dataset — the 十三里 plate, the ホルカクルキ Ainu
gloss, モトツ橋 as a sheltered span, the 登川 community shift, and the ten
entries found by hand that OpenStreetMap lacks entirely. "Errors found
and fixed" records the pipeline's own failure modes, including the
duplicate-name coordinate bug and the 穂高トンネル 1991 m / 1991 collision.
Named 274-method.md for now; intended to become 274-dev.md once the
current roster file is renamed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Field sweep additions, all absent from OpenStreetMap and found by the
author on Google Maps:
40.16 滝下八幡神社 栗山町字滝下
43.07 滝下発電所 栗山町字滝下 (北海道企業局)
57.44 登川神社 夕張市楓
66.78 豊進神社 むかわ町穂別稲里
71.86 穂別ダム神社 むかわ町穂別稲里
79.67 福山神社 むかわ町穂別福山
93.58 名石駐車場 占冠村字中央
102.36 日高神社 日高町宮下町
104.52 日高発電所 日高町字千栄 (北海道電力)
106.22 右左府発電所 日高町字日高
Plus 滝ノ上信号場 and 楓信号場, which OSM tags railway=signal /
railway=junction with no place tag, so both fell through every filter.
Corrections: モトツ橋 moved 86.16 → 87.12. It does not cross パンケモトツ沢川;
it is the sheltered span between モトツトンネル and 福山トンネル, roofed by
モトツシェルター — the identical 62 m lengths are the same deck. Author
also fixed 下福山橋 → 77.05 and オロロップ橋 → 77.55, matching route01's order.
Readings: kana is the source of truth. Romaji regenerated from it in
wāpuro style, no macrons. Adds 274-romaji.py, which does that
deterministically and is idempotent against the current table.
Artifact now shows Romaji in the table and kana under each strip mark.
CSV and GeoJSON re-propagated. 168 entries, 118 with readings.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
穂別ダム神社 at 71.86 km, 18 m off the road and 166 m before the dam.
Absent from OpenStreetMap entirely — nothing tagged place_of_worship
within 900 m of it — so it came from the author's Google Maps sighting,
positioned from the pin and geocoded to 勇払郡むかわ町穂別稲里.
Readings: author's pass takes kana coverage from 42 to 83 of 173.
Cluster summary regenerated: 穂別 now carries five objects (キャンプ場,
ダム, ダム神社, 大橋, 川).
Page rebuilt, artifact republished, CSV and GeoJSON re-propagated.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
滝ノ上信号場 (45.21 km) and 楓信号場 (57.76 km) were both absent. OSM tags
them railway=signal / railway=junction with no place tag, so they fell
through both the railway filter and the place layer that had caught
滝ノ下信号場 and 十三里信号場.
滝ノ上 was a passenger station until its demotion to 信号場 in 2024.
Strip marks now carry the full object name; the km line underneath shows
the extracted Name when it differs. Labelling by the stripped root alone
was ambiguous where several objects share a name.
CSV and GeoJSON re-propagated from the Markdown. 172 entries.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Keeps the interactive map available offline and under version control
while the song is in progress.
- 274-transect.html — self-contained; carries its data in a script block,
fonts from Google Fonts, no other external requests. Opens by double
clicking. Same page as the published artifact.
- 274-route.json, 274-districts.json — the route line and the 23 大字
bands from the 250 m reverse-geocode sweep.
- 274-transect-build.py — re-reads the roster out of 274-dev.md and
rewrites the page's data block in place, leaving markup and script
alone. Run it after editing the table.
Everything under wip/ so the auto-wiki pipeline leaves it alone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Roster now carries, for every entry: kana and romaji readings, the
administrative address with its reading, and the 大字 it sits in.
- Kana/Romaji columns prefilled from OSM name:ja-Hira / name:ja-Latn
where they exist (42 of 170); the rest are blank rather than guessed,
pending the plaque readings.
- Address and Address kana for all 170, from 国土地理院 reverse geocoding
joined to GSI muni.js and 日本郵便 郵便番号データ. Confirms 千栄=ちさか
and 穂別長和=ほべつおさわ.
- District transect: the route sampled every 250 m gives 23 大字 bands
with entry and exit kilometres. R274 does not pass through 字ニニウ at
all — it runs 穂別福山 → 字中央 → 字日高, and both municipal boundaries
fall inside tunnels (穂高 and 日高).
- Fixes a real bug: duplicate-named features shared one set of
coordinates, so both セイコーマート pointed at the same spot 50 km
apart. 12 coordinate cells were wrong.
- 沙流川 added at 105.67 km, where the road joins its valley at 右左府,
rather than at its nearest point past the range end.
- Kind labels normalised to English throughout.
CSV and GeoJSON regenerated from the Markdown, which is now the source
of truth. GeoJSON also carries the district bands as geometryless
features.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Replaces the three separate tables (structures / watercourses / PoI) with
one table in road order, and adds a Name column: the proper noun with
generic tails stripped (橋, 川, トンネル, 覆道, 郵便局, 発電所, 町, 丁目),
Ainu size qualifiers stripped (パンケ, ペンケ, ポン), and variants aliased
(十三哩→十三里, 滝ノ上→滝の上, 浪ノ沢→浪の沢, 能見→熊見, クルキ→久留喜,
三川 compounds folded together).
Objects are kept in full alongside, with kilometre, offset, kind, WGS84
coordinates and the register attributes.
188 entries, 119 distinct names, 15 names carried by three or more objects.
Adds machine-readable copies for later visualisation work:
274-lookup.csv and 274-lookup.geojson (route line plus one point per
entry, all properties carried through).
Name extraction is mechanical and meant to be corrected by hand.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Author's calls: 名称不明 has no name to use, and ベリベツ川 is mostly farm
waterway now and sits west of 川端橋, outside the range.
デチン沢川 kept pending verification — 沙流川水系, 河川コード 8101070000
(the 「その他」 group), 42.87965 / 142.43823.
Notes that the list is a shortlist to pick from, not a set of things each
of which earns a line.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
The previous cross-check fetched only the 「その他」 group and ten main
stems, so 549 of the 561 named rivers in the three systems were never
looked at. Named rivers each carry their own 河川コード and are served
separately. Full sweep now done.
Corrects the claim that ホルカクルキ川 has no W05 counterpart: it does,
河川コード 8101030169, and its position agrees with OSM to the metre
(58.26 km, 9 m off). The 北海道 name plate on the bank confirms the
spelling and gives the Ainu original ホロカ・クルヒ (horka・kurihi) —
'to turn back' and 'its shadow'; horka marks a stream running counter to
its parent river. route01's ホロカクルキ was a slip.
31 of 34 names now confirmed by both sources, all but one agreeing to
within 110 m. Remaining: 名称不明 and デチン沢川 (W05 only), ベリベツ川
(OSM only). サヌシュッペ川 parallels the road and is tangent at two
reaches; it was never a disagreement.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Adds the official river dataset (国土数値情報 河川データセット, via the
Geoshape repository at 国立情報学研究所) as a second source alongside
OpenStreetMap, covering the 石狩川, 鵡川 and 沙流川 systems.
Main stems sit under their own 河川コード rather than the 「その他」 group,
so 夕張川, 鵡川, 穂別川, サヌシュッペ川 and both ウシャップ rivers had to be
fetched separately.
30 of 34 names appear in both sources; 29 of those agree on position to
within 110 m, which validates the projection method. Adds a source column
and notes the four that need care: 名称不明 and デチン沢川 (official only),
ホルカクルキ川 (no W05 counterpart at all), and サヌシュッペ川, which
parallels the road so the two datasets are tangent 5 km apart.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Brings in the auto-wiki pipeline update: wiki.siblings frontmatter dropped
repo-wide in favour of ordinary relative Markdown links, plus the README
section documenting it.
No conflicts. 274.md and 274-dev.md were already authored under the new
convention and are unchanged by this.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Follows the pulse / pulse_dev split. 274.md keeps the frontmatter, the
intro and the lyrics workspace; everything below moves to 274-dev.md,
which will document how the PoI candidates were assembled and hold the
filtered charts.
Uses the new relative-Markdown-link convention rather than a siblings
macro, in both directions. Dev page carries publish: false to match
274.md while both are under wip/.
Contents moved verbatim; the dev page's own prose is still to be written.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
Brings the source files in line with the renderer already deployed on
novoyuuparosk-auto-wiki master, which no longer substitutes ALL_CAPS macro
placeholders. Without this, pulse.md would publish the literal text PULSE_DEV.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Drops the `wiki.siblings` frontmatter block from all eight files that carried
one — six of them declared it empty — and replaces the two live `ALL_CAPS`
placeholders with ordinary relative Markdown links:
see PULSE_DEV. -> see [Dev gibberish](pulse_dev.md).
Both now resolve to real wikilinks at publish time, and both read as links in
Gitea's blob view instead of as a naked token in the middle of a sentence.
Requires the matching renderer change in novoyuuparosk-auto-wiki; the macro
substitution it depended on no longer exists.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Replace the 由仁-十勝清水 reference section with the narrowed
川端橋 - 右左府 scope.
- structures table gains a kilometre key, derived two ways: sequence
alignment of the 北海道開発局 register against OSM way geometry by
length, and watercourse crossings as anchors where a bridge is named
after the stream it crosses (草木橋, ポン久留喜橋, オロロップ橋,
モトツ橋, 福山大橋, 新川橋)
- watercourses deduplicated 176 -> 32; the inflation was single rivers
split across many OSM ways
- PoI and watercourse tables sorted on the same datum so all three
interleave by distance
- working language switched to English, Japanese kept for proper nouns
- gloss fixed: 川端 as 'riverside', not 'end of a river'
Distances are measured from a datum near 上野幌 and are not official
kiloposts (official KP is roughly this plus 14 km). Full-route data and
the source registers live in the KB at ~/infra/kb/inbox/surreal/.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S6MpcBisZYhX45oDGRRJoz
This repo and novoyuuparosk-auto-wiki both moved out of the `mikkeli`
account, so the reusable-workflow reference follows. Same owner on both
sides means no cross-owner read gate on the private callee.
Point to the songs pipeline schema in the novoyuuparosk-auto-wiki repo
next door, and document the side-by-side `columns` fenced-block shorthand
(=== separators, Markdown emphasis) with kairo/kimito_en.md as an example.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Convert 'You and me and a gentle journey' side-by-side lyrics to the auto-wiki columns shorthand (```columns fence, columns split by ===, **bold** titles). Rendered wiki output is unchanged; the source is now Markdown-native rather than raw HTML.
Swap the <b>…</b> title tags for **…** now that the columns shorthand
converts Markdown emphasis to wikitext. Rendered output is unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the hand-written {=html} flex/poem block with the new ```columns
fenced shorthand (JP and EN split by ===). The pipeline's post-Pandoc
transform expands it to the same wikitext, so the rendered page is
unchanged — just far cleaner to author.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Workflow logic now lives in novoyuuparosk-auto-wiki/publish-songs.yml
(reusable, workflow_call). Gains incremental publishing: only changed
song files are republished instead of --all on every push.
mikkeli
wan