Align remaining documentation and provenance

README.md

@@ -16,7 +16,7 @@ share no code and no process; they share a directory of files, and the format of
 those files is the whole of the interface between them.
 
 ```text
-J-Quants API (Free plan, delayed)
+J-Quants API (Free plan)
      |
      v
 finance  (cron, 18:10 on weekdays)

doc/BASIC_DESIGN.md

@@ -121,8 +121,9 @@ historical ratio formula. It is handed frames; it reads nothing.
 
 ### 4.9 `finance/charts.py`
 
-`ChartRenderer` draws the two panels and writes the PNG, and `chart_prefix()`
-maps a window length to a file name prefix.
+`ChartRenderer` draws the price panel and, when `axis=2`, adds the oscillator
+panel, then writes the PNG. `chart_prefix()` maps a window length to a file name
+prefix.
 
 This is the one module that could not be ported by changing API calls. The
 previous implementation registered a subclass of a private pandas plotting class
@@ -198,10 +199,10 @@ continue, and returning both the results and the failures.
 The summary pipeline: read the stock list, load the stored indicator frames,
 aggregate, write, and optionally keep a dated copy.
 
-It is where the plan's delay enters the summaries. Staleness is measured against
-the newest date the plan publishes, not against today; compared with today, a
-source publishing weeks in arrears would drop every stock as stale and write an
-empty table every evening.
+It is where the configured publication window enters the summaries. Staleness is
+measured against the newest date permitted by that configured window, not
+against today. Otherwise a configured delay could make data that is current for
+the configured subscription look stale.
 
 ### 4.14 `finance/notification.py`
 

doc/DATA_CONTRACT.md

@@ -17,7 +17,7 @@ is a change to a published interface and needs the consumer considered first.
 ## 1. The boundary
 
 ```text
-J-Quants API (Free plan, delayed)
+J-Quants API (Free plan)
      |
      v
 finance  (batch, cron, 18:10 on weekdays)
@@ -301,10 +301,10 @@ drawn and the caption renders as boxes.
 
 ## 9. `data_source.txt`
 
-Where the generated data came from and how old it is. It exists because the
-source publishes in arrears: the newest figure in this directory is weeks
-behind the run that produced it, and a dashboard showing it without saying so
-would read as live market information.
+Where the generated data came from and how old it is. `generated` records when
+the job ran; `last_trading_day` records the newest trading day the run actually
+analysed. Keeping both lets the dashboard show data age without encoding a
+subscription's current publication delay in the contract.
 
 - Tab separated, one key and value per line, no header, UTF-8.
 - Three keys, in this order: `source`, `generated`, `last_trading_day`.
@@ -319,7 +319,7 @@ would read as live market information.
 | `last_trading_day` | The newest trading day the run analysed, `YYYY-MM-DD` |
 
 ```text
-source	J-Quants API (Free plan, delayed)
+source	J-Quants API (Free plan)
 generated	2026-07-21
 last_trading_day	2026-04-24
 ```
@@ -345,9 +345,9 @@ that no longer exists.
 Nothing replaces it. Restoring it would mean choosing a new source for
 fundamentals; no free source licensed for this use and offered for machine
 access has been adopted. Scraping one is not an option here. The link has been
-removed from the dashboard so that nothing dangles, and neither repository
-refers to the
-file.
+removed from the dashboard so that nothing dangles. No runtime code in either
+repository consumes `ref_index.csv`, and the dashboard no longer links it from
+the user interface.
 
 The four market indices the pipeline used to chart are withdrawn on the same
 grounds and by the same reasoning. See
@@ -383,9 +383,9 @@ splits, and **not** for dividends.
   the old inconsistency deliberately, and there is no reading of the data under
   which that is the more correct answer.
 
-No formula changed. `test/test_contract.py` regenerates all 42 computed columns
-of the committed fixture and compares them column by column, which is the
-evidence that the calculations are the same ones.
+No formula changed. `test/test_contract.py` regenerates the computed indicator
+columns of the committed fixture and compares them column by column, which is
+the evidence that the calculations are the same ones.
 
 **Other differences of meaning, and how each is handled:**
 
@@ -400,7 +400,7 @@ evidence that the calculations are the same ones.
 | Delisting | The provider stops returning rows; the summary drops the stock as stale |
 | Stock code | Four characters in the stock lists and in every generated file name; five, with a trailing zero, only inside the adapter |
 | Dates | `YYYY-MM-DD`, parsed to a tz-naive midnight index. No timezone is involved: these are trading dates, not timestamps |
-| Delay | The plan publishes in arrears. Recorded in `data_source.txt`, never hidden |
+| Delay | The configured publication window limits the newest date a fetch may request; `data_source.txt` records the actual `last_trading_day` rather than a copied provider delay. |
 
 **Legacy rows.** Data stored under the previous provider is not merged with
 this. Where the meaning of a series cannot be shown to be the same, mixing it is

doc/VERSIONS

@@ -40,6 +40,8 @@ v1.0.1 (Release Date: TBD)
 - Source the API key from a root-only environment file created by deploy.sh and
   checked by run.sh before the job starts, and leave it empty in CI.
 - Preserve the last valid summary and fail when no stock can be aggregated.
+- Keep data_source.txt provenance to the provider and plan, leaving data age
+  to last_trading_day instead of a mutable delay label.
 
 v1.0 (2026-08-14)
 -----------------

finance/analysis.py

@@ -112,8 +112,8 @@ def last_trading_day(self) -> date | None:
         Return the date of the newest row this run analysed.
 
         It is the date the dashboard shows as the age of the data. The
-        plan publishes with a delay, so it is not today and must not be
-        presented as though it were.
+        configured publication window may place that date before today,
+        so it must not be filled in with today.
         """
         if self.indicator_frame.empty:
             return None
@@ -317,10 +317,10 @@ def run_many(
     """
     Run several stocks, continuing past one that fails.
 
-    A nightly job covering thirty stocks must not lose twenty-nine of
-    them because one delisted code cannot be fetched. Each failure is
-    logged with its code and returned, so the caller can report a
-    partial run through its exit status rather than exiting zero.
+    A nightly list run must not lose every other stock because one code
+    cannot be fetched. Each failure is logged with its code and
+    returned, so the caller can report a partial run through its exit
+    status rather than exiting zero.
 
     Returns:
         The results that succeeded and the (code, error) pairs that did

finance/charts.py

@@ -7,9 +7,9 @@
 #  Description:
 #  Draw the PNG charts finance-dashboard displays: a candlestick price
 #  panel with moving averages, Bollinger bands and the parabolic SAR,
-#  and below it an oscillator panel carrying RSI, MFI, stochastics and
-#  the rest, captioned with the closing figures and the two model
-#  outputs.
+#  captioned with the closing figures and the two model outputs. When
+#  axis=2 an oscillator panel carrying RSI, MFI, stochastics and the
+#  rest is added below it; axis=1 draws the price panel alone.
 #
 #  This is the one module that could not be ported by changing API
 #  calls. The old implementation registered a subclass of a private

finance/cli/charts.py

@@ -68,6 +68,9 @@
 #       Build the J-Quants source from the settings and record where the
 #       data came from.
 #       Replace optparse, honour -u, and call the shared pipeline.
+#  v1.1 2026-08-24
+#       Record provider and plan provenance without embedding a mutable
+#       publication-delay claim.
 #
 ########################################################################
 
@@ -105,7 +108,7 @@
 # that a reader knows the figures are not live. The plan's delay is not
 # repeated here: it is a published property that can change, and the
 # last trading day beside this line is the fact that matters.
-SOURCE_DESCRIPTION = "J-Quants API (Free plan, delayed)"
+SOURCE_DESCRIPTION = "J-Quants API (Free plan)"
 
 logger = logging.getLogger(__name__)
 

finance/stocklist.py

@@ -5,10 +5,10 @@
 # finance/stocklist.py: Stock list parsing
 #
 #  Description:
-#  Read the comma separated listings that decide which stocks a run
+#  Read the comma separated stock lists that decide which stocks a run
 #  covers: stocks.txt, topix_core30.txt and the operator's private
-#  holdings file. All three share one format and differ only in how many
-#  columns they carry.
+#  holdings file. These stock lists share one format and differ only in
+#  how many columns they carry.
 #
 #  The first two columns, the code and the short name, are the ones
 #  every file has and the only two finance-dashboard reads out of
@@ -17,11 +17,10 @@
 #
 #  Every entry names a listing on the Tokyo exchange. Market indices
 #  used to be listed here too and were fetched from a provider this
-#  pipeline no longer uses. The J-Quants
-#  Free plan does not carry index values, so there is nothing to fetch
-#  them from and no special case left to make for them: a code in a
-#  stock list is a listing, and the data source refuses anything that
-#  cannot be one.
+#  pipeline no longer uses. No index dataset has been adopted for the
+#  current pipeline, so there is no special case left to make for market
+#  indices: a code in a stock list is a listing, and the data source
+#  refuses anything that cannot be one.
 #
 #  Author: id774 (More info: http://id774.net)
 #  Source Code: https://github.com/id774/finance

run.sh

@@ -91,9 +91,9 @@ CORE30_LIST=${CORE30_LIST:-topix_core30.txt}
 # a run wants unless the operator is deliberately shortening it.
 START_DATE=${START_DATE:-}
 
-# The chart windows. LONG_DAYS is inside the two years the Free plan
-# keeps, and above the 300 rows that select the long_ file name; a
-# larger value would draw the same chart under the same name.
+# The chart windows. LONG_DAYS is above the 300 rows that select the
+# long_ file name; a larger value would draw the same chart under the
+# same name.
 DAYS=${DAYS:-240}
 LONG_DAYS=${LONG_DAYS:-480}
 SHORT_DAYS=${SHORT_DAYS:-60}

test/test_contract.py

@@ -23,8 +23,8 @@
 #    underscores. Renaming a column breaks it; reordering does not.
 #
 #  The indicator file is checked against test/ti_N225.csv, which was
-#  generated by the implementation this one replaces. Regenerating all
-#  42 of its computed columns and finding them equal is what shows the
+#  generated by the implementation this one replaces. Regenerating its
+#  computed indicator columns and finding them equal is what shows the
 #  contract survived the rewrite.
 #
 #  Test Cases:
@@ -56,6 +56,9 @@
 #       Pin data_source.txt, which tells the dashboard how old the data
 #       it is showing is.
 #       Initial release.
+#  v1.1 2026-08-24
+#       Keep provenance text independent from data age and verify the
+#       generation and trading dates as separate facts.
 #
 ########################################################################
 
@@ -465,22 +468,21 @@ def read_as_the_dashboard_does(path) -> dict:
 
 def test_data_source_states_the_source_and_the_last_trading_day(settings, today):
     path = settings.data_file(DASHBOARD_DATA_SOURCE_FILE)
-    storage.write_data_source(path, "J-Quants API (Free plan, delayed)", today, date(2015, 3, 20))
+    storage.write_data_source(path, "J-Quants API (Free plan)", today, date(2015, 3, 20))
 
     values = read_as_the_dashboard_does(path)
     assert list(values) == list(DASHBOARD_DATA_SOURCE_KEYS)
-    assert values["source"] == "J-Quants API (Free plan, delayed)"
+    assert values["source"] == "J-Quants API (Free plan)"
     assert values["generated"] == "2015-03-23"
     assert values["last_trading_day"] == "2015-03-20"
 
 
-def test_the_last_trading_day_is_older_than_the_day_it_was_generated(settings, today):
+def test_generation_and_trading_dates_are_kept_separate(settings, today):
     """
-    The two dates differ, and the file is what lets a reader see it.
+    Keep the generation date and the last trading day as separate facts.
 
-    A plan that publishes in arrears means the newest row is weeks
-    behind the run. Recording only one date would leave the dashboard
-    presenting delayed figures as though they were today's.
+    The two values may differ, and recording both lets the dashboard show
+    the age of the data without inferring it from the day the job ran.
     """
     path = settings.data_file(DASHBOARD_DATA_SOURCE_FILE)
     storage.write_data_source(path, "source", today, date(2015, 3, 20))