What this index refused
Every other page here reports what was established. This one reports what was thrown away — because an index that admits everything has decided nothing, and the decisions are the product.
Two independent gates, and an article has to clear both. The resolver decides which seat, party or person a string denotes; it matches exactly or not at all. The relevance gate decides whether the article is election coverage in the first place. Neither can do the other's job, and the first version of this index had only the first one.
Gate one — is this election coverage at all?
A story that mentions a place is not a story about that constituency. This gate scores each article on electoral vocabulary, its upstream category, its topics and the organisations named in it; below the threshold nothing is resolved from it at all.
| Upstream category | Scanned | Admitted | Rate |
|---|---|---|---|
| politik | 19,021 | 10,787 | 56.7% |
| jenayah | 13,230 | 34 | 0.3% |
| sukan | 9,704 | 13 | 0.1% |
| lain-lain | 9,062 | 10 | 0.1% |
| ekonomi | 8,719 | 94 | 1.1% |
| kesihatan | 4,356 | 23 | 0.5% |
| komuniti | 3,525 | 61 | 1.7% |
| hiburan | 3,316 | 5 | 0.2% |
| alam-sekitar | 2,976 | 10 | 0.3% |
| pendidikan | 2,952 | 29 | 1.0% |
| infrastruktur | 2,582 | 56 | 2.2% |
| agama | 1,807 | 16 | 0.9% |
| pertahanan | 961 | 0 | 0.0% |
| teknologi | 905 | 1 | 0.1% |
This is the whole argument, measured rather than asserted. Crime and sport arrive in bulk and are almost never admitted; politics is admitted more than half the time. The gate was added because a car-arson story in the township of Bandar Kinrara had been published as coverage of a state seat — the name matched correctly, and the claim was still false. Categories here are the upstream newsroom's own labels, not this index's, and cover the 83,721 articles that arrive with one.
| Newsroom feed | 83,721 scanned · 12,728 scored 3 or more · 11,150 ended with a resolved entity |
| Seat-targeted search | 9,639 scanned · 4,935 scored 2 or more · 3,633 ended with a resolved entity |
The two paths hold different evidence and are gated differently. A newsroom article arrives with nothing said about elections, so it must earn 3 points. An item fetched by searching for a named seat already carries the fact that someone looked for that seat, so it is admitted at 2 — and the front page's feed then filters harder still, so a weak match appears on its seat's own page, where it is one row among many carrying its own score, and never as the lead.
Gate two — is this a name the registry knows?
The resolver matches exactly or returns nothing. It never falls through to the closest name, so a string it does not know goes to a human review queue instead of becoming a wrong answer. These are the loudest of them.
Places — 807 strings, 388 occurrences
| String | Times | Why it resolved to nobody |
|---|---|---|
| petaling jaya | 18 | no seat with this name in the state the story was filed from — a real seat name, in a different state |
| putrajaya | 17 | no seat with this name in the state the story was filed from — a real seat name, in a different state |
| machap | 13 | the name exists in more than one state, and nothing in the story said which; no seat anywhere in the registry has this name — a name several states share |
| bukit gambir | 12 | no seat anywhere in the registry has this name; no seat with this name in the state the story was filed from — contains a seat name but is not one |
| kota bharu | 10 | no seat with this name in the state the story was filed from — a real seat name, in a different state |
| dewan rakyat | 6 | no seat anywhere in the registry has this name — not a constituency at all |
| janda baik | 6 | no seat anywhere in the registry has this name; no seat with this name in the state the story was filed from — not a constituency at all |
| felda | 5 | no seat anywhere in the registry has this name; no seat with this name in the state the story was filed from — not a constituency at all |
| istana bukit serene | 5 | no seat with this name in the state the story was filed from — not a constituency at all |
| national | 5 | no seat anywhere in the registry has this name — not a constituency at all |
| semenanjung | 5 | no seat anywhere in the registry has this name; no seat with this name in the state the story was filed from — not a constituency at all |
| ampang | 4 | no seat with this name in the state the story was filed from — a real seat name, in a different state |
| forest city | 4 | no seat anywhere in the registry has this name; no seat with this name in the state the story was filed from — not a constituency at all |
| pandan | 4 | no seat with this name in the state the story was filed from — a real seat name, in a different state |
| penjara taiping | 4 | no seat anywhere in the registry has this name; no seat with this name in the state the story was filed from — contains a seat name but is not one |
| pilahan | 4 | no seat anywhere in the registry has this name; no seat with this name in the state the story was filed from — not a constituency at all |
| pulau sebatik | 4 | no seat anywhere in the registry has this name; no seat with this name in the state the story was filed from — contains a seat name but is not one |
| sg4 | 4 | no seat anywhere in the registry has this name — not a constituency at all |
| taiping | 4 | no seat with this name in the state the story was filed from — a real seat name, in a different state |
| butterworth | 3 | no seat anywhere in the registry has this name; no seat with this name in the state the story was filed from — not a constituency at all |
A further 144 place strings have been read by a person and permanently suppressed, on top of the 285 names — states, countries, foreign capitals — the resolver refuses before it looks anything up.
People — 1,634 strings, 3,526 occurrences
| String | Times |
|---|---|
| ramlan harun | 51 |
| ibrahim | 44 |
| nor azman mohamad | 27 |
| danish hairudin | 26 |
| ahmad zaharuddin sani ahmad sabri | 25 |
| syaza shukri | 24 |
| azmi hassan | 20 |
| jalaluddin | 20 |
| saifuddin shafi muhammad | 20 |
| zaliha | 20 |
| mazlan ali | 19 |
| ab rahaman arsad | 18 |
| hisommudin bakar | 18 |
| khairul shahril idrus | 18 |
| marzuki | 18 |
| samsuri | 18 |
| salleh | 16 |
| annuar | 15 |
| marzuki mohamad | 15 |
| nga | 15 |
One reason for all of them: no candidate in the registry carries this name exactly. The registry is a candidate roll, so royalty, foreign leaders, commentators and officials are all correctly absent from it — and so are real politicians whose name the roll spells differently. Which of those a string is takes a person to decide, and until someone has, it is nobody.
Parties & organisations — 435 strings, 593 occurrences
| String | Times |
|---|---|
| 同心党 | 10 |
| parti buruh | 7 |
| parti rakyat india malaysia | 7 |
| bn johor | 6 |
| gagasan rakyat | 6 |
| labour party | 6 |
| mandiri | 5 |
| umno-bn | 5 |
| gerakan reset | 4 |
| ipr | 4 |
| parti kancil | 4 |
| ph johor | 4 |
| reform uk | 4 |
| reset | 4 |
| 统民党 | 4 |
| afd | 3 |
| bn melaka | 3 |
| democratic party | 3 |
| fatah | 3 |
| labour | 3 |
Nearly every one of these is a real organisation that is not a political party — a pollster, a statutory body, a broadcaster, a government-of-the-day label. They are refused because the registry holds parties and coalitions, and being named in political news does not make something one.
The resolver, refusing
Each of these runs through the live resolver as this page is built. Nothing below is a stored answer — change a rule and this page changes.
Ayer Keroh — matched
- my.melaka.n16 N16 Ayer Keroh — Melaka · state_inferred · confidence 0.80
An exact registry name. One seat, named, with the method that matched it.
Ayer — no match
No exact match in the registry. The resolver never falls through to fuzzy or closest matching — a name that fails to match is queued for human review, not guessed.
The demonstration. Three seats begin with this word — Ayer Keroh, Ayer Molek, Ayer Limau — and a search box would offer all three. This returns nothing, because a prefix is not a name.
Melaka — suppressed on purpose
This is the name of a state, country or region — never a constituency. The resolver suppresses it deliberately; that stoplist is what keeps big place names from drowning the review queue.
Note that this is a decision, not a miss. The suppression happens BEFORE any lookup, so a state name never reaches the registry at all — and never reaches the human review queue either.
Kuala Lumpur — suppressed on purpose
This is the name of a state, country or region — never a constituency. The resolver suppresses it deliberately; that stoplist is what keeps big place names from drowning the review queue.
The same list, and the reason it runs to hundreds of entries rather than sixteen: it holds foreign capitals and countries too, so “nepal” and “washington” are refused by this rule and not by the registry.
Paloh — no match
No exact match in the registry. The resolver never falls through to fuzzy or closest matching — a name that fails to match is queued for human review, not guessed.
Two states have a Paloh. With nothing in the story to say which, the answer is nobody rather than the likelier one.
Kinrara — matched
- my.selangor.n30 N30 Kinrara — Selangor · state_inferred · confidence 0.80
Matches, correctly — and this is the string that made the second gate necessary. It resolved from a story about a car arson in the township. Resolution was right; the CLAIM that the article was about the seat was false. Deciding that is the relevance gate's job, not this one's.
Type your own into the seat lookup — it runs the same resolver in your browser against the same registry, and refuses the same way.
No article is shown on this page, and that is deliberate. A refused string is one about which nothing was established. Printing it beside a headline would manufacture exactly the association these gates exist to prevent, with this index's authority behind it. Strings and counts only.