Weekly constraint
Every system always has at least one constraint: the one thing that limits what it can achieve. That is the idea behind Eliyahu Goldratt’s The Goal. Fix it and the next one shows up. Each Monday we name ours and what we do about it, or keep the current one until it is solved.
- This week
- Resolved
- Not resolved
This constraint has not been updated this week.
-
The next constraint is named on Mon 26 Oct
-
Teams cannot work together in one place
Chats are personal, so teams cannot think through a question together. We are making rooms a shared board where people and yggy work at the same time.
Our ask. Work in a team? Tell us how you share findings today. Send feedback
-
Building on yggy takes too many steps
From API key to first answer takes too many steps. We are shipping SDKs, a CLI with browser login and clearer docs to shorten it.
Our ask. Building on the API? Tell us where you got stuck. Send feedback
-
Extraction throughput too low
Documents waited in line longer than we want, especially large imports. We worked on how many documents we can extract at the same time.
-
Cross-project work has no home
Questions and investigations that spanned several projects had no shared place. We replaced project-bound cases with rooms that work across projects.
-
Pages slow or failing on big graphs
Large workspaces loaded slowly or failed when the graph was busy. We cut redundant requests and kept totals showing when parts of a page were slow.
-
Same number not matched across files
Serial, lot and customer numbers were not reliably recognised as the same item across files. We added default identifier rules and fixed stalled approvals.
-
Same item under several names
One real item showed up under several names and some facts were dropped, so the graph was hard to trust. We joined names into one item.
-
Extraction results failing to save
Long extractions often failed while saving to the graph and errors piled up. We made saves retry and stay signed in, and stopped jobs failing silently.
-
Records of one item do not connect
The same machine or part under different IDs in different files often stayed unconnected. We fixed matching and made unclear cases visible for review.
-
Unknown terms stall document processing
When a document used a term yggy did not know, it sat paused and its facts stayed hidden. Now documents land at once and the open term waits as a note.
-
Extraction runs fail or retry endlessly
Document processing kept failing on empty or refused AI output and retried in loops. We made failures end cleanly and kept partial results.
-
Documents lost detail when read
Tables and data files were read incompletely, with values and links dropped. We changed how structure is kept so every row and field survives.
-
Slow, fragile build and release checks
Our automated checks and release builds kept stalling and queuing behind each other. We reworked how they run so changes reach you more reliably.
-
Scanned documents took too long to read
Reading scanned and dense documents was slow and often timed out. We tuned the pipeline, let single pages fail softly and showed live progress.
-
Chat ignored project boundaries
Chat and graph views did not respect which project you were in. We scoped chat to the active project and fixed requests that got blocked.
Constraints are not promises or a roadmap. They show where we put our attention, including things that are hard for us.