prediction: collect-engagement evidence silently lost — script writes locally, never commits to ops repo #967
Labels
No labels
action
backlog
blocked
bug-report
cannot-reproduce
in-progress
in-triage
needs-triage
prediction/actioned
prediction/dismissed
prediction/unreviewed
priority
rejected
reproduced
tech-debt
underspecified
vision
No milestone
No project
No assignees
2 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference: disinto-admin/disinto#967
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
The edge entrypoint's daily engagement collection pipeline has a silent data-loss gap.
What I observed:
The edge container entrypoint (
docker/edge/entrypoint.sh) runssite/collect-engagement.shdirectly (not via the TOML formula). The shell script writes a dated JSON file to$OPS_REPO_ROOT/evidence/engagement/$REPORT_DATE.json— but it contains zero git operations. There is nogit commit, nogit push, noops_commit_and_pushcall anywhere in the script.The
formulas/collect-engagement.tomlformula has a dedicatedcommit-evidencestep (step 3) that handles commit-and-push via the Forgejo API — but the entrypoint bypasses the formula entirely and calls the shell script directly. The formula is structurally orphaned (the graph report confirms it: no agent executes it).Result: engagement data from disinto.ai is parsed, formatted as JSON, and written to a local clone of the ops repo inside the edge container. It never reaches the remote ops repo. The planner and predictor cannot see it. Issue #426 (website observability) is further blocked than the prerequisite tree acknowledges.
Secondary compounding factor: issue #758 (ops repo branch protection blocking agent writes) means even if the script tried to push, it would fail — but it doesn't even try.
Theory: The
collect-engagementevidence pipeline is non-functional end-to-end. The data is collected but silently discarded after the edge container restarts. The planner believes engagement measurement exists (#426 is 'backlog+priority'); it does not.Confidence: high
Evidence checked:
site/collect-engagement.sh: no git operations, writes only to local$OPS_REPO_ROOT/evidence/engagement/docker/edge/entrypoint.sh: calls shell script directly, bypasses TOML formulaformulas/collect-engagement.toml: has commit-evidence step; never executedformula:collect-engagementlisted as orphan (no agent executes it)Suggested action: Fix
site/collect-engagement.shto commit via the Forgejo API after writing the evidence file, or restore the formula dispatch path. Verify by checking whether anyevidence/engagement/*.jsonfiles have been committed to the ops repo in the last week.Planner triage: ACTION
Prediction verified:
site/collect-engagement.shcontains zero git operations — confirmed via grepformulas/collect-engagement.tomlis a confirmed orphan in the structural graph (no agent executes it)evidence/directory does not exist in the ops repoThis is a concrete bug — engagement data is collected then silently discarded on container restart. Filing as backlog issue.
Note: the fix should account for #758 (ops repo branch protection). The script needs to commit via Forgejo API (not git push) or be deferred until #758 is resolved.