Update implementation doc

docs/implementation.md

@@ -34,20 +34,23 @@ Then, for each user:
 - if a user was previously active and is now deleted, it gets the timestamp of
   their most recent message; this requires a Slack API user token to be set in
   the environment as `$USER_TOKEN`
-- if the user is fresh, it checks if they have posted, and if so, updates them
-  to active (unless the user has been deleted without ever posting, making them
-  a no-show)
+  - if there is no message to be found, the current time is used instead as the
+    departure date; assuming the action is run frequently, this isn't too far
+    off in most cases
+- if the user is fresh, the timestamp of their first message (or the current
+  time, if they haven't posted yet) is set as their join date and the user is
+  updated to active
 - applies corrections for first and last timestamp where available
 - prints the updated and corrected vales to `tenures_new.tsv`
 
 The intermediate `users.json` is now deleted, and a file containing the unified
 diff between `tenures_new.tsv` and `tenures.tsv` is generated. This also
 generates diff subdirectories for year and month if needed; if the diff file is
-empty, it is deleted again, and we exit. `tenures_new.tsv` is now renamed to
-`tenures.tsv`.
+empty, it is deleted again. `tenures_new.tsv` is now renamed to `tenures.tsv`.
 
-Finally, the main README is generated containing the link to the latest diff;
-if there is no corrections file, the lines mentioning it are filtered.
+Finally, the main README is generated containing the link to the latest diff
+and an update stats table; if there is no corrections file, the lines
+mentioning it are filtered.
 
 The Markdown tables are generated from `tenures.tsv` by removing the header
 row, sorting by the appropriate fields, and adding an employee number before
@@ -65,8 +68,9 @@ Slack][1].
 
 ### Known bugs
 
-- If a fresh user writes their first message shortly before being removed, they
-  are considered a no-show, even though they wrote messages
+- A user who joined and left between two runs and never posted anything will
+  show up as `alum` with identical join and departure date; probably a good
+  `delete` candidate for the corrections file
 - Double quotes in `tenures.tsv` break the GitHub pretty-printing, so they are
   replaced with single quotes
 
@@ -81,6 +85,8 @@ requires consistently using the same timezone everywhere), and counts how many
 first and last timestamps have occurred per month, representing people joining
 and leaving in that month.
 
+Start and end dates in the future (from the corrections file) are ignored.
+
 After going through all users, the script iterates over the months and prints
 the totals, as well as a running total.
 
@@ -117,7 +123,8 @@ required.
 In `action.yml`, three run steps tie everything together.
 
 The first step runs `slacktenure` using the required `name` input as the
-parameter. If nothing changed, it exits.
+parameter. If nothing other than the stats table and the durations changed, it
+exits.
 
 Else, the updated `tenures.tsv` file and the diff are committed and pushed.
 
@@ -165,4 +172,5 @@ validity, looking for
 - Valid timestamp format
 
 `corrtool` is called in the first run step in `action.yml` if `corrections.csv`
-exists.
+exists. Since it exits with non-zero status on error, the action stops if the
+corrections file is malformatted.