During Advisory 3 for Tropical Storm Edouard, a rare timing condition caused the technical Forecast Advisory (TCM) to arrive before the Public Advisory (TCP). The system mistakenly marked the advisory package as complete and sent it for processing before the Public Advisory had arrived.
As a result, GPT Rewrite received the technical forecast product instead of the public-facing advisory. This caused some GPT-rewritten Advisory 3 posts to contain technical terminology and units, including Z time and wind speeds in knots, rather than the normal public-facing wording.
We already had safeguards in place to prevent incomplete or incorrect advisory data from being processed, but this specific timing edge case did not occur during our multi-day testing and was not accounted for. Additionally, GPT Rewrite should have rejected the unexpected product and fallen back to the standard non-GPT format, but that secondary safeguard did not trigger as intended.
We have corrected both points of failure. The advisory pipeline now performs additional checks to confirm that all required products have arrived before an advisory package can be marked as ready for processing. GPT Rewrite now also independently verifies that it has received the correct public-facing product before generating a rewrite.
If either validation fails, the system will fall back to the standard non-GPT format rather than allowing an incorrect product to be rewritten.
Testing confirms these changes are working as expected, and we will continue monitoring upcoming advisories under live conditions.
If your Tropical Storm Edouard Advisory 3 post contains technical forecast wording, Z times, wind speeds in knots, or other information that does not match the normal public-facing format, we recommend correcting the Advisory 3 post or removing it.
We apologize that this particular edge case slipped through the cracks. Tropical Operations is still new, and while we conducted multi-day testing and had safeguards in place before release, this specific sequence exposed a gap that our testing did not reproduce. We have addressed that gap and added additional layers of validation to help prevent a similar issue from occurring again.
