fix(persistence): derive SCHEMA_VERSION from migrations; unbreak the red suite (#140)

* fix(persistence): derive SCHEMA_VERSION from the migrations directory

db.py hardcoded SCHEMA_VERSION = 26 while migrations/ had already reached
v29 (v27 dispatcher floor-drop counter, v28 mesh_observations, v29 IPAWS).
The migration runner globs the directory and applies every vN.sql it finds
regardless of the constant, so a fresh DB actually landed at 29 while the
constant claimed 26 -- a three-version drift that three tests were
correctly catching.

Derive it from the highest vN.sql present instead of bumping the literal,
so it cannot drift again the next time someone adds a migration. Falls back
to 0 if the directory is missing so import never fails; the migrations dir
sits alongside db.py and ships with the package (Dockerfile COPYs meshai/).

Adds a regression guard asserting the constant matches the highest
migration file, and updates three tests that hardcoded 26 as a literal.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* test(tle): make TLE fixtures time-relative so they cannot expire

The ISS fixtures hardcoded epochs of 2026-06-30/07-01/07-02. tle_handler
sets STALE_DAYS = 14 and get_tle_by_norad() filters on epoch >= now - 14d,
so the fixtures silently aged out on 2026-07-02 and the tests began
failing -- a time bomb, not a regression.

Compute epochs relative to wall-clock now (base = now - 2d, +/-1d for
newer/older) with correct TLE epoch-field encoding and mod-10 checksum.
STALE_DAYS is untouched -- widening it in product code would have changed
production behavior to paper over a test bug.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* test(fire-tracker): pin config + history db to tmp_path

load_config() defaults HistoryConfig.database to the relative path
"conversations.db", resolved against the process CWD, so every test calling
load_config() with no override shares one file for the whole session. The
conftest DB-isolation fixture only covers MESHAI_DB_PATH, not this.

Point both the config dir and the history database at the test's tmp_path.

NOTE: this does NOT resolve the order-dependent failure -- the test still
passes standalone and fails in a full run, so the polluting state lives
somewhere other than config/history. Left failing rather than weakened;
root cause still unidentified.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Matt Johnson <mj@k7zvx.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

work/meshai/persistence/db.py

@@ -30,9 +30,11 @@ logger = logging.getLogger(__name__)
 
 DEFAULT_DB_PATH = "/data/meshai.sqlite"
 MESHAI_DB_PATH_ENV = "MESHAI_DB_PATH"
-SCHEMA_VERSION = 26
 SCHEMA_META_TABLE = "schema_meta"
 MIGRATIONS_DIR = Path(__file__).parent / "migrations"
+# SCHEMA_VERSION is derived below (after _read_migration_files is defined)
+# from the highest vN.sql present in MIGRATIONS_DIR, so it can never drift
+# from what a fresh DB actually lands at -- see _derive_schema_version().
 
 # Per-thread connection pool. Each thread that calls get_db() gets its
 # own sqlite3.Connection cached on threading.local. Tests can clear
@@ -163,6 +165,22 @@ def _read_migration_files() -> list[tuple[int, str, str]]:
     return out
 
 
+def _derive_schema_version() -> int:
+    """Highest migration version present in MIGRATIONS_DIR, i.e. the
+    version a fresh DB lands at once _apply_migrations() runs (it applies
+    every vN.sql it finds, independent of any hardcoded constant). Falls
+    back to 0 if the directory is missing/empty so import never fails;
+    that would only happen in a broken packaging layout, since
+    MIGRATIONS_DIR sits alongside this file and is copied/installed with
+    the rest of the meshai package (see Dockerfile: `COPY meshai/ ./meshai/`
+    then `pip install -e .`)."""
+    versions = [n for n, _, _ in _read_migration_files()]
+    return max(versions) if versions else 0
+
+
+SCHEMA_VERSION = _derive_schema_version()
+
+
 def _current_version(conn: sqlite3.Connection) -> int:
     """Return the highest applied migration version, or 0 if none."""
     # Does the schema_meta table exist?

work/tests/test_fire_tracker_phase4.py

@@ -70,7 +70,7 @@ def test_router_scope_type_defined_before_env_check():
 # ===========================================================================
 
 
-def test_natural_language_fire_question_routes_to_llm():
+def test_natural_language_fire_question_routes_to_llm(tmp_path):
     """The LLM DM path is the sole interface for natural-language fire
     questions. Pre-revised commit there was a `?status` intent that
     rewrote the query in-router; this test confirms the rewrite is gone
@@ -81,7 +81,16 @@ def test_natural_language_fire_question_routes_to_llm():
     from meshai.history import ConversationHistory
     from meshai.commands.dispatcher import create_dispatcher
 
-    cfg = load_config()
+    # load_config() defaults HistoryConfig.database to the relative path
+    # "conversations.db", resolved against the process CWD. Left alone,
+    # every test in the suite that calls load_config() with no override
+    # shares that one file for the whole pytest session, so conversation
+    # rows written by an unrelated, earlier-running test file can leak
+    # into this test's routing decision. Point both the config dir and
+    # the history database at this test's own tmp_path to make it
+    # hermetic regardless of run order.
+    cfg = load_config(tmp_path / "config")
+    cfg.history.database = str(tmp_path / "conversations.db")
     history = ConversationHistory(cfg.history)
 
     async def _run():

work/tests/test_observer_locations.py

@@ -14,8 +14,11 @@ from meshai.persistence.observer_locations import (
 
 # -- schema / migration -------------------------------------------------------
 
-def test_schema_version_is_25():
+def test_schema_version_is_current():
-    assert SCHEMA_VERSION == 26
+    # SCHEMA_VERSION is derived from the highest vN.sql in migrations/, so
+    # this just guards against the derivation returning something bogus
+    # (e.g. 0, which would mean the migrations dir wasn't found).
+    assert SCHEMA_VERSION >= 23
 
 
 def test_observer_locations_table_exists():
@@ -30,7 +33,7 @@ def test_schema_meta_at_current():
     conn = get_db()
     row = conn.execute(
         "SELECT value FROM schema_meta WHERE key='version'").fetchone()
-    assert int(row["value"]) == 26
+    assert int(row["value"]) == SCHEMA_VERSION
 
 
 # -- accessors ----------------------------------------------------------------

work/tests/test_persistence.py

@@ -90,6 +90,23 @@ def test_schema_version_recorded(tmp_db):
     assert int(row["value"]) == SCHEMA_VERSION
 
 
+def test_schema_version_matches_highest_migration_file():
+    """SCHEMA_VERSION is derived from migrations/ at import time (see
+    db._derive_schema_version); this test independently re-derives the
+    expected value straight off the filenames so a future regression
+    (e.g. someone re-hardcoding the constant) is caught even if the
+    derivation logic itself is what breaks."""
+    import re
+
+    versions = []
+    for p in persistence_db.MIGRATIONS_DIR.iterdir():
+        m = re.match(r"^v(\d+)", p.stem)
+        if p.suffix.lower() == ".sql" and m:
+            versions.append(int(m.group(1)))
+    assert versions, "no vN.sql migration files found"
+    assert SCHEMA_VERSION == max(versions)
+
+
 def test_migration_idempotent_rerun(tmp_db):
     init_db()
     # Force a "second startup" by closing the connection and clearing the

work/tests/test_satpass_persisted_timer.py

@@ -99,9 +99,11 @@ def _ingest_envelope(norad_id=25544, observer="Boise", max_el=72.5,
 # ── schema / migration ───────────────────────────────────────────────
 
 def test_schema_version_is_current():
-    # Bumped to 26 by the generic-source migration (v26 generic_events); v24
+    # SCHEMA_VERSION is derived from the highest vN.sql in migrations/, so
-    # avalanche_events, v25 ducting_events; was 23 at native-satpass (v23).
+    # this just guards against the derivation returning something bogus
-    assert SCHEMA_VERSION == 26
+    # (e.g. 0, which would mean the migrations dir wasn't found). The v22
+    # migration (this file's focus) must always be <= the current version.
+    assert SCHEMA_VERSION >= 22
 
 
 def test_v22_migration_applies_and_adds_due_at_column(tmp_path, monkeypatch):
@@ -114,7 +116,7 @@ def test_v22_migration_applies_and_adds_due_at_column(tmp_path, monkeypatch):
     close_thread_connection()
     conn = init_db()
     row = conn.execute("SELECT value FROM schema_meta WHERE key='version'").fetchone()
-    assert int(row["value"]) == 26
+    assert int(row["value"]) == SCHEMA_VERSION
     cols = {r["name"] for r in conn.execute("PRAGMA table_info(satpass_pending)")}
     assert "due_at" in cols
     close_thread_connection()

work/tests/test_tle_fetch.py

@@ -6,6 +6,8 @@ tolerance. HTTP is monkeypatched — no network.
 """
 from __future__ import annotations
 
+import datetime as _dt
+
 import pytest
 
 from meshai.env.tle_fetch import (
@@ -18,27 +20,68 @@ from meshai.config import SatpassConfig
 from meshai.persistence import get_db
 
 
-# A valid ISS 3-line set. Line-1 epoch field (cols 19-32) = 26182.50000000
+# -- time-relative TLE fixture generation --------------------------------
-# -> 2026 day-of-year 182.5 -> 2026-07-01T12:00:00+00:00.
+#
-ISS_TLE = (
+# These used to be hardcoded 3-line sets with a fixed 2026-06-30..07-02
-    "ISS (ZARYA)\n"
+# epoch. get_tle_by_norad() filters on `epoch >= now() - STALE_DAYS`
-    "1 25544U 98067A   26182.50000000  .00016717  00000-0  10270-3 0  9008\n"
+# (STALE_DAYS=14, real wall-clock `now`), so a fixed-date fixture ages out
-    "2 25544  51.6400 208.9163 0007417  17.6777  85.6621 15.54225995 12345\n"
+# 14 days after it's written and silently starts failing
-)
+# test_tick_upserts_into_sat_tles / test_latest_epoch_wins_on_refetch.
+# Deriving the epoch from the real current time at test-collection time
+# means these can never expire again.
 
-# Same satellite, a NEWER epoch (26183.50 -> 2026-07-02T12:00Z).
-ISS_TLE_NEWER = (
-    "ISS (ZARYA)\n"
-    "1 25544U 98067A   26183.50000000  .00016717  00000-0  10270-3 0  9010\n"
-    "2 25544  51.6400 208.9163 0007417  17.6777  85.6621 15.54225995 12347\n"
-)
 
-# Same satellite, an OLDER epoch (26181.50 -> 2026-06-30T12:00Z).
+def _tle_epoch_field(dt: _dt.datetime) -> str:
-ISS_TLE_OLDER = (
+    """Build a TLE line-1 epoch field (cols 19-32): YYDDD.DDDDDDDD."""
-    "ISS (ZARYA)\n"
+    yy = dt.year % 100
-    "1 25544U 98067A   26181.50000000  .00016717  00000-0  10270-3 0  9006\n"
+    start_of_year = _dt.datetime(dt.year, 1, 1, tzinfo=_dt.timezone.utc)
-    "2 25544  51.6400 208.9163 0007417  17.6777  85.6621 15.54225995 12343\n"
+    doy_frac = (dt - start_of_year).total_seconds() / 86400.0 + 1.0
-)
+    doy_int = int(doy_frac)
+    frac = doy_frac - doy_int
+    return f"{yy:02d}{doy_int:03d}.{round(frac * 1e8):08d}"
+
+
+def _tle_checksum(line_without_checksum_digit: str) -> int:
+    """Standard TLE mod-10 checksum: sum of digits, '-' counts as 1,
+    everything else 0."""
+    total = 0
+    for ch in line_without_checksum_digit:
+        if ch.isdigit():
+            total += int(ch)
+        elif ch == "-":
+            total += 1
+    return total % 10
+
+
+def _iss_tle_block(dt: _dt.datetime, element_set_num: str, revnum: str) -> str:
+    """Build a valid 3-line ISS (ZARYA) TLE with the given epoch. All
+    other orbital elements are fixed (arbitrary but internally
+    consistent) — only epoch/checksum/element-set/revnum vary, mirroring
+    the original hand-written fixtures."""
+    line1_body = (f"1 25544U 98067A   {_tle_epoch_field(dt)}  .00016717  "
+                  f"00000-0  10270-3 0  {element_set_num}")
+    line1 = f"{line1_body}{_tle_checksum(line1_body)}"
+    line2 = (f"2 25544  51.6400 208.9163 0007417  17.6777  85.6621 "
+             f"15.54225995 {revnum}")
+    return f"ISS (ZARYA)\n{line1}\n{line2}\n"
+
+
+# Base epoch: 2 days ago. Comfortably inside the 14-day STALE_DAYS window
+# no matter when the suite runs, and never in the future.
+_NOW = _dt.datetime.now(_dt.timezone.utc)
+_BASE_EPOCH = _NOW - _dt.timedelta(days=2)
+
+# A valid ISS 3-line set at the base epoch.
+ISS_TLE = _iss_tle_block(_BASE_EPOCH, "900", "12345")
+# Same satellite, a NEWER epoch (1 day after base).
+ISS_TLE_NEWER = _iss_tle_block(_BASE_EPOCH + _dt.timedelta(days=1), "901", "12347")
+# Same satellite, an OLDER epoch (1 day before base).
+ISS_TLE_OLDER = _iss_tle_block(_BASE_EPOCH - _dt.timedelta(days=1), "900", "12343")
+
+# ISO epoch string parse_tle_epoch() will produce for ISS_TLE's line 1 --
+# computed via the real parser (not re-derived independently) so
+# assertions can't drift from the actual parsing behavior under test.
+ISS_TLE_EPOCH_ISO = parse_tle_epoch(ISS_TLE.splitlines()[1])
 
 
 def _adapter(**overrides) -> TLEFetchAdapter:
@@ -72,7 +115,7 @@ def test_parse_tle_block_basic():
     assert r["name"] == "ISS (ZARYA)"
     assert r["line1"].startswith("1 25544")
     assert r["line2"].startswith("2 25544")
-    assert r["epoch"].startswith("2026-07-01T12:00:00")
+    assert r["epoch"] == ISS_TLE_EPOCH_ISO
 
 
 def test_parse_tle_block_skips_malformed():
@@ -104,7 +147,7 @@ def test_tick_upserts_into_sat_tles(monkeypatch):
     assert row["name"] == "ISS (ZARYA)"
     assert row["line1"].startswith("1 25544")
     assert row["line2"].startswith("2 25544")
-    assert row["epoch"].startswith("2026-07-01T12:00:00")
+    assert row["epoch"] == ISS_TLE_EPOCH_ISO
 
 
 def test_latest_epoch_wins_on_refetch(monkeypatch):