The walkthrough
68.2 seconds. Captions are burned in; an
.srt and .vtt sit beside the file.
The friction, as reported
The board ranks ideas by votes, inside each forum separately.
The same request can be filed twice and decided two different ways.
Nothing on the board makes that visible.
Nobody filed this one. It was synthesised from 380 harvested ideas across 7 forums, so there is no requester to quote.
What the walkthrough shows
- Row three has 666 votes and is marked Accepted.
- The same request also sits in Invoices and Quotes.
- That copy carries 518 votes of its own.
- One is Accepted. The other is Not in pipeline.
- The panel can show the contradiction. It cannot explain it.
- Why these differ: not stated.
- Row three still reads 666. Its rank has not moved.
- Seven remittance ideas sit inside this single forum.
- 980 votes between them, and no entry shows that.
- Row seven still reads 319. It is still seventh.
- Here is a link the machine proposed from matching text.
- That match was wrong. A person decided, and it is gone.
- Same order. Same counts. Two rows now carry a marker.
The assessment
The opportunity, in a different shape.
Merging the duplicates would produce one number that overstates them both.
Showing both counts and both statuses side by side costs less and settles more.
The reasoning
This one was not filed by anyone. It was synthesised from the harvest, and the evidence for it is stronger than for several of the board's own ideas.
The obvious solution — merge the duplicates, show a combined total — is the wrong form for four specific reasons.
A merged total overstates support for any single deliverable. 1790 people did not endorse a combined bills-and-invoices dispute flag. 531 endorsed a sales-invoice flag; 1259 endorsed a bills flag whose description never mentions invoices. Publishing 1790 as one number claims an endorsement nobody gave.
Nothing decides “same idea” reliably except a person. A lexical near-duplicate scan over all 380 titles returns roughly 50 candidate pairs at a 0.34 threshold, of which about six are genuinely the same request. Relaxing the threshold to buy recall costs far more than it returns. Worse, the scores rest on judgement calls invisible in the output: treat “option” as a stopword, which is entirely defensible, and the clearest true duplicate on the board drops from 0.500 to 0.333 and is missed.
A visible combined number is a target, and votes from different forums are incomparable quantities — page one costs 310 votes in one forum and 62 in another.
The reshape is to cluster and disclose. Show related entries together, each with its own count and its own status, side by side and unchanged. Never sum. Never re-rank. The value was never the arithmetic — it was seeing that the same request was filed twice and decided two different ways.
The cost is stated rather than buried: this removes the ability to decline quietly. Once an unapprove row for invoices visibly sits beside an Accepted one for bills, 518 people can see the identical request succeeded for the other document type.