Invoice + Project Quote

Quote #QT-NEE-2026-057

The tags under your articles, written by something that reads the article

Priority: High. This is the precursor — the tag vocabulary, the publish-time engine and the separate tag-cleanup proposal (QT-NEE-2026-059) all depend on this being done first, and something is writing these tags on every article you publish today. Measured across all 21,600 tagged articles: 69% take four-fifths or more of their tags straight from their own headline and 45% take every one. Four items — find and stop what is writing them, build a tag vocabulary from your own editorial tagging and real search demand, put an engine on publish that uses it, and correct the tag defects a reader can see today.

Michael,

Here is what sits on one of your articles right now. The headline is "Space Bounce Attraction Launches at Jungle Island," published 14 July 2026. The tags printed underneath it are: attraction · bounce · island · jungle · launches · space.

That is the headline with the small words removed and one tag made from each word left over. Nothing read the article. Nothing decided those were its subjects. A machine split a sentence.

It is not one article. Every published article carrying at least one tag was checked — 21,600 of them — and each tag tested against the words of its own headline:

• 14,991 articles, 69%, take four-fifths or more of their tags from their own headline. • 9,719 articles, 45%, take every single tag from the headline. • 4,176 articles, 19%, carry tags with no headline overlap at all. That is your editors' real work, and it is mixed in with the rest.

It switched on in 2019 and it has not switched off. Articles published in 2018 average 9% headline-derived tags. 2019 jumps to 80%. Articles published in 2026 still run at 69%, including the July one above.

ItemAmount
Item 1

What is writing the tags, found and stoppedshow detailhide detail

$895.00

The mechanism is still running. It wrote the tags on an article published in July 2026, and articles published this year still take 69% of their tags from their own headline. Nothing below is durable while it runs — merge or delete tags today and the next published headline starts rebuilding the pile. Its source has already been looked for in the obvious places and is not there. No active plugin and no theme file on the live site writes post tags. The single match inside the plugin folder is the WordPress function name sitting in a code comment, not a call to it. So it is most likely an import route, a syndication feed, or an outside posting tool, which makes this genuine investigation rather than a known edit — and it is why this item is priced on its own rather than folded into the build.

  • Write-path forensics — every route on your site able to write a tag enumerated and tested one by one$325
  • Controlled reproduction on staging — a test article published and its tags captured as they appear$220
  • The mechanism stopped on the live site, with a documented way to put it back$250
  • Verification across 14 days of live publishing that no new headline-derived tags appear$100
  • Every route on your site able to write a tag listed and tested — scheduled tasks, the publishing interface, the import routes, the feed handlers, and the active and inactive plugin and theme files — proof: the list with the result of each test beside it, including the routes ruled out
  • The exact mechanism named, with the evidence that identifies it — proof: a tag produced by it on demand on staging, captured as it is written
  • A test article published on staging and the tags it receives recorded before and after the change — proof: the two tag sets side by side
  • The mechanism stopped on the live site in a form that can be reversed — proof: the change itself and the reversal instruction, both written down
  • Articles published in the 14 days after the change tested against their own headlines by the same method used to measure the 21,600 — proof: the table, every article listed, none omitted
  • Your editors’ existing tags left untouched by this item — proof: the tag assignment count before and after, unchanged

There is a real possibility this is not site code at all, in which case stopping it means changing how articles arrive rather than editing a file. If that is where it lands, the finding and the recommended change are reported and this item closes there rather than expanding into work you did not agree to.

Item 2

The tag vocabulary, built from your own editorial tagging and real search demandshow detailhide detail

$1,450.00

This is the part that makes the tagging search-analysed rather than machine-guessed, and the raw material for it is already inside your site. 5,398 of your tags have never once been taken from the headline of an article carrying them. A person applied those. 710 of them appear on three or more articles and 402 on five or more — Must See, Local Attraction, Social Events, named columnists, named FIU schools. That is a record of what your newsroom actually covers, and it is the seed. Two further sources go against it. Your 179 categories, which unlike your tags are indexed and do earn clicks, so the line between a category and a tag gets drawn deliberately instead of by accident. And your own Search Console query data, so a term earns its place because people search for it, not because it reads like a topic. What comes out is a list you can read and argue with, with the evidence printed beside every term.

  • The 5,398 human-applied tags extracted and ranked by real use across your published articles$350
  • Measured search demand matched to each candidate term from your own Search Console data$475
  • Category overlap resolved so no tag is built to compete with a category page that already ranks$325
  • The vocabulary written up with a definition and the evidence per term, and revised once after your review$300
  • The 5,398 tags never taken from a headline extracted, each with the number of published articles it appears on — proof: the full ranked list, not a sample
  • Each candidate term matched against your own Search Console query data so inclusion rests on measured demand rather than judgement — proof: the demand figure printed beside every term, including the ones it disqualifies
  • Every candidate checked against all 179 of your categories — proof: the overlap list and the decision recorded on each one
  • A finished vocabulary with a one-line definition per term and the evidence that put it there — proof: the document
  • One revision after your review, with your removals and additions applied and shown as changes rather than returned as a fresh list — proof: the before and after

A controlled vocabulary is the whole difference between this and what you have now. Without one, an automatic tagger invents a fresh term for every article and in a few years you own a different 24,000-tag pile. Everything in the next item is held to this list.

Item 3

The tagging engine, reading the article and running on publishshow detailhide detail

$2,350.00

On publish, the article’s headline, body and category are read and a small number of tags are chosen from the vocabulary above — chosen on what the article is about, not on which words happen to appear in its title. A date is not a subject. A verb is not a subject. Jungle Island is. Two constraints are built in rather than hoped for. Nothing outside the approved vocabulary reaches an article without a person approving it, which is what stops the pile rebuilding. And a cap of eight tags per article, adjustable by you — your articles carry six on average today — so the tag row under a story stays a short list a reader might actually follow. Your editors keep the last word. Tags can be changed or removed in the editor exactly as now, and anything an editor sets by hand is never overwritten on a later save. Before it goes live it is measured rather than asserted. Articles your own editors tagged by hand go through it with their tags hidden, and its choices are put next to what the human chose — agreements and disagreements both.

  • The classification service — article read, subjects extracted, vocabulary terms matched and ranked$850
  • Site integration so it runs on publish and writes the tags, with editor changes always taking precedence$650
  • The guardrails — vocabulary limit, per-article tag cap, and a review queue for terms the engine wants to add$425
  • Accuracy measured against articles your editors tagged by hand, reported before it goes live$275
  • 90 days of monitoring after go-live, with drift reported and corrected$150
  • Tags chosen from the article’s body and category rather than its headline — proof: the July 2026 Jungle Island article run through it, its six headline words and the new tags side by side
  • No term outside the approved vocabulary reaches an article without a person approving it — proof: an attempt to apply an unapproved term, shown refused
  • A cap of eight tags per article, adjustable by you — proof: the setting, and a batch of newly published articles showing none exceeds it
  • A tag an editor sets or removes by hand is never overwritten on a later save — proof: the test, performed and recorded
  • Accuracy measured against articles your editors tagged by hand with their tags hidden from the engine — proof: the comparison table, agreements and disagreements both listed, nothing dropped
  • A review queue where terms the engine proposes wait for a person — proof: the queue, with a proposed term sitting in it
  • Ninety days of monitoring after go-live, each month’s tagging sampled and tested by the same headline-overlap method used to measure the 21,600 — proof: the three monthly checks

This replaces the old mechanism rather than sitting beside it, which is why the previous item comes first in the order below.

Item 4

The tag defects a reader can see today, correctedshow detailhide detail

$495.00

Three things in the existing taxonomy are wrong on the page rather than merely untidy. 305 tags store an HTML code in place of the character it stands for, so a tag renders literally as “College of Arts Sciences & Education” underneath the article. It also splits your content in two: 245 FIU articles sit under the broken spelling and 190 under the clean one, with not one article in common. 17 tags misspell a real person’s name. Gloria Burns appears as “Glorial Burns” on 65 articles. Linda Rodriguez Bernfeld appears as “Linda Rodriguez Berenfeld” on 58. Ron Magill appears as “Ron Margil” and again as “Ron Margill.” Every one was confirmed against the author credited on the articles carrying it rather than guessed, and several are your own columnists. Then the quiet ones: 323 tags attached to no published article at all, 23 orphaned database rows behind them, and two different tags both displaying as “Cutler Bay.”

  • The 305 tags storing an HTML code corrected and merged into their clean spellings$185
  • The 17 misspelled proper names corrected against the author credited on each article$165
  • Empty tags, orphaned database rows and the duplicate Cutler Bay resolved$145
  • The FIU college pair merged so the 245 articles and the 190 sit under one tag — proof: the article count on the merged tag, before and after
  • Each of the 305 tags storing an HTML code corrected — proof: the live tag page rendering the character instead of the code
  • Each of the 17 misspelled names corrected against the author credited on the articles carrying it — proof: the name, the articles, and the author credit that confirms it, one row per correction
  • 323 tags carrying no published article removed — proof: the count before and after
  • 23 orphaned database rows cleared — proof: the same query run before and after
  • The two tags both displaying “Cutler Bay” resolved into one — proof: the tag list showing a single entry
  • Nothing merged or removed that carries genuine editorial tagging — proof: the 5,398 human-applied tags checked against the change list, with the check shown

This is the only part of the existing 24,328 tags this quote touches — roughly 45 operations, every one of them a visible defect. The remainder is deliberately left alone, for the reason set out above.

Total$5,190.00

Approve & pay

Tick the items you're approving. You can approve any one of them on its own, or all of them together — you are only charged for what you tick.

Total$5,190.00
WHAT THAT HAS BUILT UPshowhide

24,328 tags. 5,271 of them, 21.7%, are near-duplicates, junk or misfiled — and those 5,271 carry 63% of every tag assignment on the site. 1,259 singular-and-plural pairs like Host and Hosts. 362 tags that are bare numbers or fragments of a date — 12, mar, $15. 15,206 tags used on exactly one article, because a word unique to one headline makes a tag unique to one article.

WHAT TAGS RETURN TODAY, STATED PLAINLYshowhide

You should have this number before you decide rather than after. Across the full 90 days to 21 September 2026, all 24,328 of your tag archive pages together earned 46 clicks from Google, against 9,170 across the site — half of one per cent.

That is not a failure waiting to be fixed. Your tag pages are set to noindex and are deliberately kept out of your sitemap, so search engines are being told not to rank them. On today's settings, tags return essentially nothing in search.

So this is not sold to you as a search-traffic project, because as things stand it would not be one. What it fixes is what a reader sees under every article, and what your editors' genuine tagging is currently buried under. Whether tags should also be opened to search — which would make the vocabulary in item 2 worth considerably more than it is worth today — is a decision that would follow this work rather than precede it.

One more piece of measurement is being brought in before anyone treats this as urgent: the site's analytics engagement data, to see whether readers actually follow tag links, independently of the search figure above. That figure comes from Search Console and measures Google only. The prices here hold either way. The measurement decides the timing, not the number.

WHY THE FIRST ITEM COMES FIRSTshowhide

The mechanism is still writing tags today. Merging or deleting tags while it runs rebuilds the problem from the next published headline onward. Finding and stopping it is therefore item 1 rather than a step inside the build, and it is worth doing on its own even if everything after it waits.

WHAT IS DELIBERATELY NOT IN THIS QUOTEshowhide

Re-tagging the 21,600 existing articles is not priced here and is not recommended today. Those tags are noindexed, they earn 46 clicks between them, and once the mechanism is stopped they stop multiplying. The roughly 24,000 that remain are inert. If tags are later opened to search, re-tagging the back catalogue becomes worth pricing properly at that point. Until then it would be a large bill set against a number already close to zero.

The four items below are in the order they have to happen in.

Four items, one project, invoiced in three stages. Each stage is invoiced when it is delivered and is payable on that invoice. Nothing is billed ahead of the work it covers.

Stage 1 — $1,390. What is writing the tags, found and stopped, and the visible tag defects corrected. Stage 2 — $1,450. The vocabulary, delivered and revised after your review. Stage 3 — $2,350. The engine live on publish, with its measured accuracy reported.

The order is not arbitrary. The mechanism is stopped before anything is cleaned, and the vocabulary exists before the engine that is held to it.

The model running cost of the tagging engine is carried by NeedAWebNow for the first twelve months after go-live. It is not a charge on this quote and it is not billed separately.

Your editors' existing tags are preserved throughout. The only terms this quote removes are the 323 that carry no published article and the misspellings and duplicate spellings named in item 4, each of which is merged into its correct term rather than deleted.

Payment methods: Stripe (card) OR Zelle → EDILBERTO MORILLO · 786-426-4902.

This quote holds at these prices through Friday, October 23, 2026.