Four axes. Four candidates.
One locked-before-scoring verdict.
Skills wins by 10 pts (460 vs 450). Inside the 2% tie-band — tie-breaker hierarchy kicks in. Both finalists ship runnable live demos, so rung 1 doesn't break it. Plugins as packaging layer nests Skills + MCP servers + hooks, so the honest read of the matrix is: ship the combined session (this expedition's actual call), with Skills as the headline primitive and MCP as the depth chapter.
Harness scored last (300) because session 2 already covered it — series-continuity penalty was the dominant signal. The framework worked: a clear quantitative justification for "don't re-teach last month."
The matrix
| Axis | Weight | MCP protocol | Skills markdown unit | Plugins packaging | Harness subagents+hooks |
|---|---|---|---|---|---|
|
Audience fit
Topics land when matched to the group's current pain point, not the presenter's interest [2]
|
× 35% | 5+175 | 5+175 | 4+140 | 3+105 |
|
Series continuity
Third-in-a-series sessions earn attendance from arc, not novelty — bridge cleanly from AI/security [3]
|
× 25% | 4+100 | 5+125 | 4+100 | 2+50 |
|
Speaker readiness
Authentic, lived material beats researched material; pick what the presenter has actually shipped [4]
|
× 25% | 4+100 | 4+100 | 3+75 | 4+100 |
|
Demo viability
A working demo is the highest-retention element of a 60–90 min virtual session. Tie-breaker rung 1.
|
× 15% | 5+75 | 4+60 | 3+45 | 3+45 |
| Weighted total | 45090% | 460★ TOP — 92% | 36072% | 30060% | |
Why not RICE?
RICE = (Reach × Impact × Confidence) / Effort. Built by Intercom to compare product-feature ideas where reach is measured in "customers per quarter" and effort in "person-months". Both axes degenerate for a fixed-audience, fixed-duration session.
| Reach | DEGEN same N attendees regardless of topic |
| Impact | partial — but already captured by audience-fit axis |
| Confidence | partial — captured by speaker-readiness axis |
| Effort | DEGEN bounded to the 90-min slot |
Score still computes; stops discriminating. Weighted scoring is the more flexible choice when criteria are domain-specific [7] [6].
Tie-breaker hierarchy
Apply rungs in order. First rung that yields a clear winner ends the tie.
- Runnable live demo > slides-only. The 8-12 min active block is the session's retention spike.
- Internal speaker with lived experience > external expert presenting researched material [2].
- Topic the group has asked about > topic the organizer thinks they should care about.
The 30-minute process
List 3–5 candidates
Stop at five. The matrix breaks down at high candidate counts — every additional column doubles the discussion overhead.
Lock weights BEFORE seeing candidates
Agree with co-owners before the candidate list is visible. Prevents gaming scores toward a predetermined pick [3].
Score 1–5, compute, tie-break
Multiply, sum, rank. Apply the tie-breaker hierarchy if results land within ~5% of each other.
Sanity-check continuity
Does the winner connect to the prior session in one sentence? If not, re-score continuity — it was probably under-weighted.
Discipline rules
Weights must sum to 100%
Mechanical check before scoring. If they don't, fix before reading candidates — never adjust after.
Pick ≤4 axes
Eight+ collapses into multi-hour scoring sessions and dilutes weight signal below noise floor.
Never change criteria mid-evaluation
If a new axis seems vital after scoring two candidates, you've found a flaw in step 1 — restart, don't patch [3].