Honewords In Progress

Find the word you actually mean: start broad, pick what fits, and see what sets each word apart.

The meaning map with sad at the center and reflection opened through quiet and lingering, where the branch forks to pensive "lost in thought, faintly sad" and melancholy "a mood that settles in"; the path, filters and describe-it box on the left, best matches under the map, and melancholy's word card on the right.
The word sad in the middle of a map of its shades, with one branch followed through to melancholy. Where that branch splits, pensive is marked "lost in thought, faintly sad" and melancholy "a mood that settles in". Filters and a box to describe what you mean are on the left, the closest words sit below, and melancholy's meaning and an example are on the right.
Architecture diagram: Honewords not wired yet word with no entry Meaning map and wordcard Describe it: a typedsentence matched againsteach map's cues by fixedrules Compare: two or threewords side by side, andwhat tells them apart Word lookup server: anentry by word, or a requestfor a missing one Word entries: a card forevery word, a map forevery broad word Request queue: visitorsahead of the backlog, await measured from realthroughput Offline generator, planned:a language model writeseach entry once Entry rules: three to sevenqualifiers, at most fivewords per fork, a short tellon every card
The map and its two views run today on sample data built into the page. The dotted lines are still to come: wiring the page to the server, and an offline generator whose entries reach the store only by passing the entry rules.The map works today on a small set of words built into the page. The dotted lines are not done yet: connecting the page to the server, and an AI that writes new entries ahead of time, with a separate set of checks each entry must pass before it is saved.

The Meaning Map

Each broad word branches into three to seven qualifiers, directions of meaning such as "longing" or "hopes were raised", never relation types like "stronger". Where a qualifier forks, each word's tell is its last step. Narrow by intensity, duration and tone, show the opposite and its branches, or open a word's card.

Each broad word opens into a handful of directions, such as "longing" or "hopes were raised". Where one splits into several words, a few words on the map say how each differs. Narrow by how strong, how lasting or how formal a word is, show the opposite, or open any word for its meaning and an example.

Describe It

Each map carries cues: regular expressions tied to a qualifier, and sometimes to the one beneath it. The sentence is matched against them and the hits become a suggested path. No model runs while someone is on the page, so the box recognizes phrasings it was given; it does not read.

Each map comes with phrases to watch for, each pointing at one of its directions. Your sentence is checked against them, and whatever matches becomes the suggested route. Nothing clever runs while you use the site, so the box spots phrasings it already knows rather than understanding you.

Compare

Every card records intensity, duration, tone, typical cause, definition and an example, so any pair can be set against each other field by field. A pair people often confuse, such as dejected and despondent, gets a sentence written for it instead: one expects to recover, the other has stopped expecting to.

Every word notes how strong it is, how long it lasts, how formal it sounds, what usually causes it, what it means and an example, so any two can be lined up point by point. Words people often confuse, like dejected and despondent, get a sentence written just for them.

Words Not There Yet

Built and tested, not yet wired to the page: the server answers from the entry store, or queues the word ahead of the backlog and returns an estimate from measured throughput plus any usage-limit pause. Planned next is the offline generator: a language model writes each entry once, kept only if it passes the entry rules.

This part is built and tested but not yet connected to the page. A missing word goes to the front of the waiting list, and the wait is worked out from how fast words really get done. Still to come: an AI that writes each entry once, ahead of time, keeping only the ones that pass a set of checks.