/awol: one unparseable roster record aborts the entire report

[SyniRon]

Problem

/awol aborts the whole roster report when a single member record fails to parse. Two spots in commands/awol.go call HandleError + return inside the per-member loop:

• date parse failure (commands/awol.go:~169-173)
• uniform/milpac URL parse failure (commands/awol.go:~195-199)

One bad record kills the report for everyone. The error strings also name neither the offending member nor the bad value, so staff can't tell who/what tripped it.

Scope / origin

Pre-existing behavior — predates and was preserved through #159 (PR #162). Explicitly deferred out of #159 to keep that PR scoped; filing as the agreed follow-up.

Proposed behavior (needs a decision)

Skip-and-continue instead of abort: log the offending member + bad value at WARN (via utils.Warn), drop that one member from the report, and render the rest. Optionally surface a small "N records skipped" note in the embed/file output so the omission isn't silent.

Decision for triage: silently skip + log, or skip + visibly annotate the report? Partial-but-rendered beats all-or-nothing, but staff should probably know coverage was incomplete.

Acceptance criteria (draft)

• A single unparseable date or uniform URL no longer aborts the report; remaining members render.
• Each skipped record is logged at WARN naming the member (username/discord id) and the bad value.
• Decision on visible annotation captured + implemented (silent-log vs in-report note).
• Behavior covered by tests via the existing runAwol seam (inject a roster with one malformed record, assert the rest render).
• golangci-lint, go test ./... -race, coverage floors, and go build green.

Refs

• Surfaced in PR feat(awol): accountable-day count + days-AWOL severity report (#159) #162 review (/awol accountable-day rework, /awol: accountable-day count (subtract LOA) + days-AWOL severity report #159).
• Both call sites: commands/awol.go:~169-173, :~195-199.

[SyniRon]

This was generated by AI during triage.

Agent Brief

Category: bug
Summary: /awol aborts the whole roster report when a single member record fails to parse. Switch to skip-and-continue, matching the existing /afsm and /s6 pattern.

Current behavior:
In the per-member loop of runAwol, two parse failures abort the entire report instead of skipping the one bad record:

• An unparseable LastForumPostDate calls HandleError and returns.
• A uniform URL that doesn't match the milpac-ID regex calls HandleError and returns.

One bad record kills the report for everyone, and the error text names neither the offending member nor the bad value, so staff can't tell who or what tripped it.

Desired behavior:
Both of these are genuine upstream errors (the forum should always return a parseable date, and a roster profile should always carry a valid uniform URL), so treat them the way /afsm treats per-member failures rather than swallowing them silently:

• On either parse failure, capture the error to Sentry via utils.CaptureError, including the member's username and discord id and the bad value, then skip that one member and continue the loop. Do not abort.
• Count skipped records. When the count is greater than zero, append a visible note to the rendered report along the lines of ⚠️ N records skipped due to errors (reported), mirroring the (reported) wording /afsm already uses, and log a summary line at INFO.
• The rest of the roster renders normally.

This keeps the "partial but rendered beats all-or-nothing" outcome while making the omission visible to staff and routing the genuine error to Sentry, consistent with ADR 0001 (manual capture at chosen sites).

Key references in the codebase:

• /afsm already implements exactly this shape: per-member utils.CaptureError + skippedCount++ + continue, then a post-loop ⚠️ ... skipped due to errors (reported) annotation and an INFO summary. Follow that precedent for wording and structure.
• The runAwol(r, cache, now, i) seam takes an injected roster and is already used by the existing /awol tests. Use it to drive the new tests.
• utils.CaptureError(msg, err, kv...) is the Sentry path; utils.HandleError stays the user-facing path. Per ADR 0001, keep that split intact.

Acceptance criteria:

• A single unparseable forum post date no longer aborts the report; the remaining members render.
• A single unparseable uniform URL no longer aborts the report; the remaining members render.
• Each skipped record is captured to Sentry via utils.CaptureError and logged, naming the member (username and discord id) and the offending value.
• When one or more records are skipped, the rendered report shows a visible (reported)-style note with the skipped count.
• Tests via the runAwol seam inject a roster containing one malformed date record and one malformed uniform URL and assert the other members still render and the skipped note appears.
• golangci-lint run, go test ./... -race -cover, the coverage floors, and go build are all green.

Out of scope:

• Changing how parseable records are scored or sorted.
• Reworking the LOA cache-health degradation path (that's a separate, already-shipped behavior).
• Backfilling or repairing the bad source data on the forum or roster side; this issue only changes how /awol reacts to it.