JSM diagnostic and worked reconciliation
Why Did a JSM Demand Score Drop? Reconcile the Change
Explain a falling Wezy demand score by separating vote inputs, organization attribution, recency and formula changes. Use a worked reconciliation, not guesses.
In short
Compare the same request at two documented observation times. Recalculate with the same formula first, then isolate a policy change. A lower score does not by itself mean customers withdrew support.
In this guide
A lower demand score is a result to explain, not evidence by itself that customers lost interest. In Wezy, current support, organization representation, the rolling recent-vote term and the active scoring policy can change independently. Diagnose the change on one request before explaining a movement in a dashboard ranking.
The useful output is a reconciliation: how much of the difference is explained by the recorded inputs, how much by a changed formula, and what remains unknown. This is different from preparing a monthly demand report.
First establish that the two observations are comparable
Capture the request key, observation time and timezone, V, O, R30, the active formula and its version. For an Advanced formula, also record W and the organization-weight policy. Note whether the value came from the current Wezy panel, a published Jira field, a historical snapshot or an exported file.
Do not compare a newly refreshed panel with a stale search result and call the difference a lost vote. The Wezy field guide separates vote confirmation, Jira field publication and indexing. Similarly, a ranking position can change because another request improved even when this request’s score stayed constant.
If the earlier observation is missing, write “prior inputs unavailable.” Current values cannot reconstruct the exact earlier attribution or history.
Reconcile Standard scoring one component at a time
The documented Standard formula is:
S = V + 3 × O + 2 × R30
Change in S = change in V + 3 × change in O + 2 × change in R30
The second line follows algebraically when the formula is unchanged. The metric definitions matter: V counts active votes, O distinct represented organizations, and R30 active votes first activated within the last 30 days.
Consider this fictional two-observation record:
| Input | Earlier | Later | Contribution to the score change |
|---|---|---|---|
| V | 10 | 11 | +1 |
| O | 3 | 2 | −3 |
| R30 | 4 | 1 | −6 |
| Standard score | 27 | 19 | −8 |
The calculations are 10 + 9 + 8 = 27 and 11 + 6 + 2 = 19. The score falls by eight even though there is one more active vote. That rules out the simplistic explanation “eight votes were lost.” It does not identify which customer changed what.
R30 can decrease as eligible active votes age out. Organization representation can change without an equal change in active votes. Withdrawal and reactivation do not restart the retained first-activation date. Investigate the recorded state instead of assuming each net difference corresponds to one identifiable action.
Separate a formula change from an input change
When the active policy changed, do not subtract the headline scores and attribute the whole difference to customers. Use a two-stage bridge:
A = score(earlier formula, earlier inputs)
B = score(earlier formula, later inputs)
C = score(later formula, later inputs)
Input effect under the earlier formula = B − A
Formula effect at the later inputs = C − B
Total observed change = C − A
For the fictional record above, A is 27 and B is 19. Suppose the later Advanced configuration is (a,b,c,d) = (1,1,2,1) and the later W is 4. C is 11 + 2 + 8 + 1 = 22.
The bridge is therefore −8 from recorded input changes under the old formula, +3 from the new formula at the later inputs, and −5 overall. Calling that a five-point decline in customer enthusiasm would hide the policy change.
This is an analytical convention, not causal attribution. Reversing the order of the bridge can allocate interactions differently. State the method and do not mix current organization weights into an old historical score. Wezy’s history documentation says historical calculations retain their original formula context.
Use the symptom to choose the next check
| Observation | First check | What not to conclude |
|---|---|---|
| V unchanged, R30 lower | First-activation age and observation boundary | That customers necessarily withdrew |
| V rises, O unchanged | Additional voters may represent an existing organization or no organization | That the new vote was ignored |
| W changes but S does not | Active formula and coefficient c | That a weight setting necessarily failed |
| All scores scale similarly | Coefficients and activation/version change | That demand multiplied across the project |
| Panel and Jira search disagree | Field publication, indexing, account and request identity | That one display proves data loss |
| Request vanishes from a ranking | Applied filters, access, pagination and status | That its vote history became zero |
A zero organization weight removes its weighted contribution, not its active votes or its presence in O. A Standard installation’s non-applicable W must not be substituted with zero to fabricate an Advanced calculation.
Make the explanation reproducible
Use this short internal record, storing only information the intended audience is allowed to see:
Request and comparison purpose:
Earlier and later observation times, including timezone:
Source of each observation:
Earlier/later V, O, R30 and applicable W:
Earlier/later formula and configuration version:
Input bridge:
Policy bridge:
Unexplained difference or unavailable evidence:
Owner and next check:
For historical periods, verify complete covered UTC days. Missing coverage is not zero demand, and an installation cannot supply a prior period it never recorded. An unexplained remainder is a reason to inspect the evidence or contact support, not to adjust the fields manually until the arithmetic matches.
Stop when the comparison cannot be reconstructed
A missing earlier formula or an unavailable required input prevents a complete bridge. Report the component changes you can substantiate and leave the remainder unexplained; do not assume the current formula was active earlier. This is especially important after an edition change or an organization-weight override.
Keep the precision actually available in the source when doing the worksheet calculation, then round only the presentation. A small difference between displayed rounded values and a separately rounded calculation is not evidence of a customer action. Conversely, rounding is not an acceptable explanation for a material discrepancy without checking the inputs. Preserve the original values and calculation method in the review record.
What to tell stakeholders
A defensible message is: “The score fell from 27 to 22. Under the former formula, the changed recorded inputs account for −8; activating the new formula adds 3 at the later inputs. Active votes rose from 10 to 11. We are checking the attribution and recency changes.” This is an illustrative explanation, not a real customer result.
Use Wezy’s scoring and history guide as the source of product behavior and the calibration worksheet before future formula changes. Teams evaluating Wezy on Atlassian Marketplace should require an explainable demand signal—not a number that appears precise but cannot be reconciled.
Published by the team behind Wezy Customer Voting. Product capabilities are linked to their official documentation; worked examples are illustrative.
Let customers vote on your JSM portal.
Install Wezy from Atlassian Marketplace, open one request for voting and see the first vote in Jira.