Sprint 2 Retrospective — Billing Completion, Auth & Pymble Testing¶

Sprint: Sprint 2 — Billing Completion, Auth Foundation & Pymble Testing Dates: 6 Apr – 17 Apr 2026 Participants: Ola, Samawal, Monther, Areej, Sarah Facilitator: Sarah Shalfoun

Per-Person Feedback¶

Name	What Worked	What Didn't Work	Suggestions
Ola	Sprint felt organised; user stories were clear; AI assistance accelerated delivery.	Rebase-and-merge workflow caused friction; AI context was lost between sessions.	Pre-handle recurring issues (e.g. database migrations) before they hit `develop`.
Samawal	Clear user stories and clear sprint scope; strong team spirit.	AI hallucinations during implementation; AI budget limits hit too quickly.	Check existing PRs for migration conflicts before opening a new PR; fix the build pipeline; use Opus for planning and Sonnet for execution.
Monther	Temporary planning files and reviewing updated files together worked well.	Opus 4.7 context-window issues.	—
Areej	Documentation quality across the sprint was good.	User stories were clear for developers but not for QA.	Separate business impact from technical detail in future stories; spend more time pairing with developers next sprint.
Sarah	Strong team spirit and collaboration; quality-focused mindset; productive sprint planning discussions.	Story and PR conventions were not consistently followed.	—

ℹ️ Empty cells above are intentional. Every concern raised in the table is consolidated and addressed in the Issues and Decisions sections below.

Issues¶

1. Rebase, Merge & Migration Conflicts¶

Several merges in Sprint 2 surfaced hidden conflicts, most painfully around database migrations that share a single sequence file.

Pre-handle recurring issues — migrations specifically:

Always check the current develop branch and any open PRs for migration changes before opening a new PR.
Treat the migration sequence file as a shared, manually-coordinated artifact.
Fix the build pipeline so it flags this class of hidden conflict (and any other recurring issue we identify) at PR time, not after merge.

2. AI Context Loss & Hallucinations¶

Both Ola, Samawal and Monther reported losing AI context mid-task and occasional hallucinations, made worse by Opus 4.7 context-window limits.

Interim mitigation (until the corporate AI workflow is ready): keep handling these as we currently do — short scoped sessions, temporary planning files, and reviewing updated files together as Monther described.

3. AI Budget Limits Hitting Too Quickly¶

Samawal hit Anthropic budget caps faster than expected during Sprint 2.

Interim mitigation: use Opus for planning and Sonnet for execution until the corporate AI workflow lands. This preserves reasoning quality where it matters and lowers per-task cost on the bulk of execution work.

4. Stories Clear for Devs, Not for QA¶

Areej raised that current user stories are written from a developer's perspective and don't give QA a clean view of what to test.

Interim mitigation: in the next sprint, story authors will explicitly separate business impact from technical detail, and Areej will spend dedicated time with developers during refinement.

5. Inconsistent Adherence to Story & PR Conventions¶

Story format and PR conventions (see branch conventions) were not consistently followed in Sprint 2.

Interim mitigation: call this out in Sprint 3 kickoff and enforce at PR review.

Decisions¶

The following decisions were agreed by the team and will be implemented in Sprint 3.

Build Pipeline¶

Fix outstanding pipeline issues so it runs reliably on every PR.
Add a migration hidden-conflict check to the pipeline (detects parallel changes to the migration sequence file / overlapping migration numbers across open PRs).
Establish a regular cadence to review recurring issues that surface in retros and add them to the pipeline as automated checks.
Re-add the pipeline to branch protection rules for both develop and main, so PRs cannot be merged without it passing.

PR Workflow¶

Pre-PR conflict check is mandatory: before opening a PR, the author must check for conflicts against develop and against any open PRs — particularly for migrations.

AI Usage (Interim, Until Corporate AI Workflow Ships)¶

Use Opus for planning, Sonnet for execution — keep planning quality high while reducing budget burn.
Continue current mitigations for context loss and hallucinations (short scoped sessions, temporary planning files, reviewed-files walkthroughs).

Story Quality¶

Separate business impact from technical detail in future user stories so QA can read them independently.
QA pairs with developers during refinement in the next sprint.

Carryovers Closed¶

Story	Note
#38 PDF Invoice Generation	Resolved in Sprint 2; closed in this retrospective. Sprint 3 hardens it with CI coverage in story #82.
#80 Tenant Management	Largely complete; ~2 SP carried into Sprint 3 (Ola).

Action Items (Owners & Tracking)¶

#	Action	Owner	Target
1	Fix build pipeline issues and re-attach to `develop` + `main` branch protection	Head of Technology	Sprint 3, week 1
2	Add migration hidden-conflict check to the pipeline	Build pipeline owner	Sprint 3
3	Standing review of recurring retro issues → pipeline checks	Head of Technology	Recurring (every retro)
4	Communicate "check `develop` + open PRs before opening a new PR" rule and add to branch conventions	Sarah	Before Sprint 3 kickoff
5	Adopt Opus-for-planning / Sonnet-for-execution split	All developers	Immediate, Sprint 3
6	Story template update: separate business impact from technical detail	Sarah / Areej	Before Sprint 3 refinement
7	QA paired with developers during Sprint 3 refinement sessions	Areej	Sprint 3