Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@trapdata/ml/models/classification.py`:
- Around line 440-441: The description strings for the classifier classes
TuringThailandSpeciesClassifier, TuringMadagascarSpeciesClassifier, and
TuringJapanSpeciesClassifier are inconsistent with the dates encoded in their
weights_path filenames; update each class's description field to match the date
in its weights_path (e.g., use 2024-11-21 for the Thailand model, 2024-07-01 for
the Madagascar model, and 2024-11-22 for the Japan model), and confirm with the
Turing team before committing to ensure the corrected human-readable dates are
accurate.

---

Outside diff comments:
In `@trapdata/api/api.py`:
- Around line 46-50: The startup currently calls initialize_service_info()
inside lifespan() without error handling so any HTTPError from
get_or_download_file()/get_labels() during classifier instantiation will abort
FastAPI startup; wrap the initialization in a try/except in lifespan() (catch
Exception or requests.HTTPError) and on failure log the full error (use
logger.exception or include the exception message) and set
app.state.service_info to a safe default (e.g., empty dict or partial result) so
the service can start; alternatively, move try/except into
initialize_service_info() around each classifier construction so failing
classifiers are skipped but others are initialized.

---

Nitpick comments:
In `@trapdata/ml/models/classification.py`:
- Around line 441-442: The new classifier assignments use single-line string
literals for weights_path and labels_path which is inconsistent with the rest of
the file; update the four new classifiers to use the same parenthesized
multi-line implicit string concatenation style as the other classifiers by
replacing the single-line weights_path and labels_path values with
parenthesized, split strings (keeping the exact URLs intact) for the variables
named weights_path and labels_path in the relevant classifier blocks in
classification.py.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@trapdata/ml/models/classification.py`:
- Around line 451-476: Four classifiers (TuringThailandSpeciesClassifier,
TuringMadagascarSpeciesClassifier, TuringJapanSpeciesClassifier,
TuringSingaporeSpeciesClassifier) reference weights_path/labels_path URLs that
return 404 and will fail during the base-class download step; either update each
class's weights_path and labels_path to valid, accessible URLs (ensure the
verification script returns HTTP 200 for both files) or remove/disable their
registration in the API (trapdata/api/api.py) until the assets are
uploaded—locate the classes above and replace the broken URLs with the correct
object-arbutus links or comment out/unregister these class definitions so the
API does not advertise non-functional models.

---

Duplicate comments:
In `@trapdata/ml/models/classification.py`:
- Line 453: The description strings for some trained models are incorrect;
update the relevant description assignments (the description variable in the
country-specific model definitions) to match their filenames: for Thailand
change "Trained on 11th November 2024 by Turing team using Resnet50 model." to
"Trained on 21st November 2024 by Turing team using Resnet50 model.", for
Madagascar change "Trained on 11th November 2024 by Turing team using Resnet50
model." to "Trained on 1st July 2024 by Turing team using Resnet50 model.", and
for Japan change "Trained on 19th November 2024 by Turing team using Resnet50
model." to "Trained on 22nd November 2024 by Turing team using Resnet50 model.";
ensure these exact strings replace the current description assignments so the
dates now match their filenames.

---

Nitpick comments:
In `@trapdata/ml/models/classification.py`:
- Around line 454-476: The weights_path and labels_path strings in
TuringThailandSpeciesClassifier, TuringMadagascarSpeciesClassifier,
TuringJapanSpeciesClassifier, and TuringSingaporeSpeciesClassifier are
single-line URLs while the rest of the file uses parenthesized multi-line string
concatenation for long URLs; update the weights_path and labels_path assignments
in those classes (refer to the class symbols TuringThailandSpeciesClassifier,
TuringMadagascarSpeciesClassifier, TuringJapanSpeciesClassifier,
TuringSingaporeSpeciesClassifier and the attributes weights_path and
labels_path) to use the same parenthesized multi-line string concatenation style
used elsewhere for consistency and readability.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@trapdata/api/api.py`:
- Line 16: Remove the unused "noqa: E501" suppression on the import line
starting with "from .models.classification import (" (RUF100) and convert the
explanatory text that follows it ("disabled: Singapore - no category map,
Thailand - map/weights mismatch") into a regular inline or above-line Python
comment so the rationale is preserved but the dead noqa directive is eliminated.

In `@trapdata/api/models/classification.py`:
- Around line 249-250: Add a disable comment above the
MothClassifierTuringThailand class like the one used for
MothClassifierTuringSingapore to make its disabled state explicit and prevent
surprise errors; specifically note that this classifier is disabled because the
category map is not available and that attempting to instantiate it will raise a
RuntimeError in model.load_state_dict() due to a class-count mismatch (3822 vs.
3800). Place the comment immediately before the MothClassifierTuringThailand
class definition to mirror the existing pattern used for
MothClassifierTuringSingapore.

In `@trapdata/ml/models/classification.py`:
- Around line 451-455: Add a 3-line "# NOTE:" comment block above the
TuringThailandSpeciesClassifier declaration (like the one above
TuringSingaporeSpeciesClassifier) explaining that this classifier is
intentionally disabled because the labels file contains 3822 classes while the
Resnet50 model checkpoint has 3800 output neurons which causes
model.load_state_dict() to raise a size-mismatch RuntimeError; instruct
maintainers to not instantiate this class until the weights or labels are
corrected (or the checkpoint is replaced) and reference the class name
TuringThailandSpeciesClassifier and the failing call model.load_state_dict for
clarity.

---

Duplicate comments:
In `@trapdata/ml/models/classification.py`:
- Around line 453-467: Update the description strings to match the dates
embedded in each class's weights filename: change
TuringThailandSpeciesClassifier.description to use "21st November 2024" to match
the file date 2024-11-21, change TuringMadagascarSpeciesClassifier.description
to use "1st July 2024" to match 2024-07-01, and change
TuringJapanSpeciesClassifier.description to use "22nd November 2024" to match
2024-11-22; ensure the same wording/format ("Trained on <ordinal> <Month> <Year>
by Turing team using Resnet50 model.") is preserved for each class
(TuringThailandSpeciesClassifier, TuringMadagascarSpeciesClassifier,
TuringJapanSpeciesClassifier).