MAHDTech Prioritize Backlog
A backlog is a pile until it's groomed. Grooming turns the pile into an ordered queue whose top is unambiguously the next thing to do - ordered by value against cost and risk, low-value items cut, and no blocker ever sitting above the work it gates.
This is generic backlog grooming - ranking any pile of work (ideas, features, tickets) by value, whatever tracker it lives in. It does not touch the .tars/issues/ backlog that the /backlog-* skills automate; that is a separate ticket pipeline, not the pile you groom here.
Prepare the items
Pull every candidate into one list. Each item is a discrete unit of value - an outcome someone wants, not a task ("users can reset their password", not "add a reset endpoint"). Split anything too big to compare against its neighbours; merge slivers that only deliver together.
Pick a lens and score
Score every item through one lens - mixing lenses makes the ranks incomparable. Each trades effort for fidelity; pick the cheapest that actually separates your items:
- Value vs effort - plot each item on value against effort. Do the high-value / low-effort quadrant first; interrogate the low-value / high-effort one. Fastest, coarsest.
- RICE - score Reach × Impact × Confidence ÷ Effort. Adds reach and an explicit confidence discount, so a wide, well-understood item beats a narrow, speculative one of equal appeal.
- Cost of delay / WSJF - rank by cost-of-delay ÷ duration (weighted shortest job first). Reach for it when timing dominates - a deadline, a compounding cost, a closing window - because it prices what waiting costs, not just what doing is worth.
These are options, not gospel. Whatever the lens, the axis is the same: value earned against cost and risk paid.
Cut before you order
The highest-leverage move in grooming is subtraction. For each low-scoring item ask: if it never ships, who notices? If the honest answer is nobody, cut it. Cut duplicates, items overtaken by newer work, and anything whose value has decayed below its cost. A backlog that only grows is a graveyard - its tail is never reached, so keeping it costs attention for nothing.
Cluster and sequence
- Cluster items that share a component, a theme, or a user journey, so you batch related work instead of thrashing between contexts.
- Sequence by score, then reorder for dependencies: no item may sit above one that blocks it, however high it scores. Walk the list top-down and pull each blocker above the work it gates. When a low-value item is the only thing unblocking a high-value one, its real priority is borrowed from what it unblocks.
Keep it small
Cap the groomed, ordered zone - only the next chunk needs to be sharp. Everything past the line stays coarse: a rough parking lot you re-groom, not a queue you maintain. Re-groom on a cadence (each iteration, each intake) rather than letting the pile drift.
Handoffs
- Once the top items are ordered and sharp, split them into buildable slices with
/to-tickets. - When an item is too big to see the end of - you can't yet state its value or its cost - it isn't a backlog item but an investigation; chart it with
/wayfinder.