ESA, NOAA, and LOC stock adapters generate unstable clip IDs

Summary

The ESA, NOAA, and Library of Congress stock-source adapters use Python's built-in hash() to generate source_id values from media URLs. Python randomizes string hashes between processes, so the same media URL receives different source_id and clip_id values across separate OpenMontage runs. This can cause cache misses, duplicate corpus records, repeated downloads, and unreliable deduplication.

Operating system

Ubuntu 24.04 (all)

Pipeline

No response

Runtime / renderer

No response

Steps to reproduce

1. Open a terminal in the OpenMontage repository.

2. Confirm that the affected adapters use Python's built-in hash():

   git grep -n 'hash(' -- tools/video/stock_sources/esa.py tools/video/stock_sources/noaa.py tools/video/stock_sources/loc.py

3. Create an ESA Candidate using the original adapter logic in a Python process with hash seed 1:

   PYTHONHASHSEED=1 python3 -c 'from tools.video.stock_sources.base import Candidate; url="https://www.esa.int/ESA_Multimedia/Videos/2026/08/Sophie_Adenot_s_first_spacewalk"; candidate=Candidate(source="esa", source_id=f"esa_{hash(url) & 0xFFFFFFFF:08x}", source_url=url, download_url=url, kind="video"); print(candidate.clip_id)'

4. Repeat the same operation in a separate Python process with hash seed 2:

   PYTHONHASHSEED=2 python3 -c 'from tools.video.stock_sources.base import Candidate; url="https://www.esa.int/ESA_Multimedia/Videos/2026/08/Sophie_Adenot_s_first_spacewalk"; candidate=Candidate(source="esa", source_id=f"esa_{hash(url) & 0xFFFFFFFF:08x}", source_url=url, download_url=url, kind="video"); print(candidate.clip_id)'

5. Compare the two clip_id values. They are different even though the source and media URL are identical.

On the reproduced environment, the output was:

PYTHONHASHSEED=1 -> esa_esa_7c5d721c PYTHONHASHSEED=2 -> esa_esa_d14f73b6

The exact hexadecimal values can vary by Python build, but changing PYTHONHASHSEED changes the ID. NOAA and Library of Congress use the same URL-hashing pattern and therefore have the same problem.

Expected behavior

• The same media URL should always produce the same source_id and clip_id, regardless of which Python process handles it or which hash seed is used.
• URL-backed IDs should use a deterministic algorithm such as SHA-256.

Actual behavior

The same URL produces different IDs in separate Python processes:

PYTHONHASHSEED=1 -> esa_esa_7c5d721c PYTHONHASHSEED=2 -> esa_esa_d14f73b6

Because Candidate.clip_id is used by the corpus and shared clip cache, these changing IDs can result in duplicate records, repeated downloads, cache misses, and failed deduplication.

Relevant logs or error output

$ PYTHONHASHSEED=1 python3 -c 'from tools.video.stock_sources.base import Candidate; url="https://www.esa.int/ESA_Multimedia/Videos/2026/08/Sophie_Adenot_s_first_spacewalk"; candidate=Candidate(source="esa", source_id=f"esa_{hash(url) & 0xFFFFFFFF:08x}", source_url=url, download_url=url, kind="video"); print(candidate.clip_id)'
esa_esa_7c5d721c

$ PYTHONHASHSEED=2 python3 -c 'from tools.video.stock_sources.base import Candidate; url="https://www.esa.int/ESA_Multimedia/Videos/2026/08/Sophie_Adenot_s_first_spacewalk"; candidate=Candidate(source="esa", source_id=f"esa_{hash(url) & 0xFFFFFFFF:08x}", source_url=url, download_url=url, kind="video"); print(candidate.clip_id)'
esa_esa_d14f73b6