# postgres version
postgres (PostgreSQL) 16.15 (Homebrew)
# initdb (disposable cluster at [temporary disposable cluster])
# server up; running experiment.sql
============================================================

===== PHASE 0: schema expanded, backfill already complete =====
psql:experiment.sql:10: NOTICE:  table "entries" does not exist, skipping
-- State after backfill (expect duration_minutes = minutes everywhere):
 id | label | minutes | duration_minutes
----+-------+---------+------------------
  1 | a     |      60 |               60
  2 | b     |      30 |               30
  3 | c     |      45 |               45
(3 rows)

-- Skill trigger query (expect 0 before the legacy writer acts):
 drift_rows
------------
          0
(1 row)


===== LEGACY WRITER ACTS (knows only the old column) =====
-- W1: update an already-backfilled row (id=1, was 60/60):
-- W2: insert a new row after backfill (supplies no duration_minutes):
-- Raw state after W1 + W2:
 id | label | minutes | duration_minutes
----+-------+---------+------------------
  1 | a     |      90 |               60
  2 | b     |      30 |               30
  3 | c     |      45 |               45
  4 | w2    |     120 |
(4 rows)


===== STRATEGY A: new-column-only read =====
-- (H1/H2) A trusts duration_minutes alone. Watch id=1 and id=4:
 id | label | value_seen_by_a
----+-------+-----------------
  1 | a     |              60
  2 | b     |              30
  3 | c     |              45
  4 | w2    |
(4 rows)


===== STRATEGY B: old-column-authoritative compatibility read =====
-- (H1/H2) B treats minutes as the source of truth during the window:
 id | label | value_seen_by_b
----+-------+-----------------
  1 | a     |              90
  2 | b     |              30
  3 | c     |              45
  4 | w2    |             120
(4 rows)

-- Side by side, with a correctness verdict for A vs the authoritative value:
 id | label | authoritative | a_sees | a_is_correct
----+-------+---------------+--------+--------------
  1 | a     |            90 |     60 | f
  2 | b     |            30 |     30 | t
  3 | c     |            45 |     45 | t
  4 | w2    |           120 |        | f
(4 rows)


===== DRIFT MEASUREMENT (skill trigger) =====
-- (H3) IS DISTINCT FROM count during the window (expect 2):
 drift_rows_is_distinct
------------------------
                      2
(1 row)

-- (H6) Plain <> count for contrast (expect 1: it misses the NULL row):
 drift_rows_plain_neq
----------------------
                    1
(1 row)

-- The rows each filter flags:
 id | label | minutes | duration_minutes | flagged_by_is_distinct | flagged_by_plain_neq
----+-------+---------+------------------+------------------------+----------------------
  1 | a     |      90 |               60 | t                      | t
  2 | b     |      30 |               30 | f                      | f
  3 | c     |      45 |               45 | f                      | f
  4 | w2    |     120 |                  | t                      |
(4 rows)


===== STRATEGY C, WRONG ORDER: reconcile while writer still active =====
-- Reconcile now:
-- Trigger immediately after reconcile (momentarily 0):
 drift_after_reconcile_only
----------------------------
                          0
(1 row)

-- ...but the writer is NOT retired, so one more legacy write lands (W3 update id=2):
-- (H4) Trigger again (expect non-zero: the race re-opened drift):
 drift_reconcile_raced_writer
------------------------------
                            1
(1 row)


===== STRATEGY C, RIGHT ORDER: retire the writer, THEN reconcile =====
-- (Writer retirement is modeled by: no further minutes-only writes occur after this line.)
-- Final reconciliation:
-- (H5) Trigger after retire-then-reconcile (expect exactly 0):
 drift_after_retire_then_reconcile
-----------------------------------
                                 0
(1 row)

-- (H5) A and B now agree on every row:
 id | label | b_value | a_value | a_matches_b
----+-------+---------+---------+-------------
  1 | a     |      90 |      90 | t
  2 | b     |      75 |      75 | t
  3 | c     |      45 |      45 | t
  4 | w2    |     120 |     120 | t
(4 rows)


===== DONE =====
============================================================
# done; disposable cluster will be removed on exit