Why Krillion Rejects Answers That Look Right: Validation, Normalization & the Six Rarity Tiers

Written by Marina Voss, Founder & Editor · Published September 14, 2026 · Last verified: September 14, 2026

Fast answer: Krillion rejects your answer because it validates by exact match against a closed, pre-built answer catalog — not by judging whether your answer is actually correct. Every prompt ships with a human-reviewed list of canonical answers and registered aliases, each already carrying a fixed rarity tier. If your normalized text isn't on that list, you get a clean rejection, even when your answer is genuinely right. Here's how that pipeline works, the seven ways a valid-looking answer falls out of it, and what you can do about each one.

• ✓ How Krillion's closed-catalog validation and normalization pipeline actually work
• ✓ The seven reasons a correct-looking answer gets rejected — with a first-hand capture
• ✓ How the six fixed rarity tiers are scored, and why they don't move mid-game
• ✓ Why catalog size decides how likely an honest omission is
• ✓ A copy-paste triage checklist for your next rejection, and how to report real omissions
• ✓ What the sources can't tell us — the honest limits

In a hurry? Check whether your word counted on past prompts →

Why Krillion rejects answers: the closed-catalog rule

Most daily word games check your input against a dictionary or a live language model. Krillion does neither. According to the official FAQ ↗, answer sets are curated in advance "by aggregating datasets (dictionaries, wiki pages, encyclopedias, whatever sources the question calls for), then merging and cleaning the result into one answer set." Rarity tiers are attached during that same editorial pass — some prompts scored mechanically against a familiarity database, others with LLM assistance and a manual review. When you play, the game isn't thinking about your answer at all; it's looking it up.

Two consequences follow. First, catalogs are prompt-scoped: the method page ↗ states plainly that "the same word can be accepted under one question and rejected under another." Second, every judgment call — which spellings count, whether a stretch answer is clever or just wrong — was made by the editor before the dive went live, and frozen into a versioned catalog. If your idea isn't in that version, no amount of correctness rescues it mid-game. You can see exactly what made each past catalog in our answer archive, and check whether a specific word counted on any logged prompt with the answer checker.

How Krillion validates what you type, step by step

The method page documents a textbook normalization pipeline, and each step maps to a rejection you've probably hit:

1. Structural rejections. Empty submissions, anything over 80 characters, and punctuation-only text are thrown out before matching begins.
2. Unicode NFKC normalization. Compatibility variants — fullwidth characters, ligatures, superscripts — fold onto their plain equivalents, the standard trick for loose matching described in Unicode UAX #15 ↗.
3. Lowercasing. Case never decides acceptance.
4. Punctuation becomes spacing. Hyphens and apostrophes turn into word separators rather than vanishing, so "spider-man"-style names stay token-separated instead of gluing into one word.
5. Whitespace collapse, then exact match. The method page's worked example shows "Storm," "storm!" and a version padded with extra spaces all reducing to the same lookup key.

The decisive line is what the pipeline omits: "no fuzzy matching, no thesaurus, no translation, and no runtime language models." A near-miss typo or an unregistered synonym fails cleanly. That's deliberate — it keeps validation deterministic and auditable — but it means the burden of matching the catalog's exact form sits entirely on you.

Seven reasons a valid-looking answer gets rejected

After logging every daily sheet since launch and comparing them against the documented pipeline, every rejection I've seen decomposes into one of these:

1. Catalog omission. The answer is valid but the editor never found it. The developer's own one-line FAQ answer to "why was my answer not listed?" is disarmingly honest: "Because I didn't find it, sorry. Sometimes I miss a source."
2. Unregistered spelling variant. With no spell-correction at runtime, a genuinely different spelling is rejected rather than treated as a rare answer. Diacritics only pass where both forms were registered — a German-dish prompt on August 19 accepted both Käsespätzle and Spaetzle because someone listed both by hand.
3. Plural vs. singular. Normalization doesn't stem or lemmatize, so each word form needs its own catalog entry. Archived sheets show both forms listed only where the editor chose to accept both.
4. Multi-word phrasing. Punctuation folds to spacing, but word order and completeness must still match the canonical form or an alias. Half a name is a different string, not a partial credit.
5. Proper-noun conventions. First name vs. surname vs. full name is an editorial decision encoded per answer — type the wrong granularity and you're out.
6. Prompt-scope mismatch. Country vs. territory, actor vs. character, sport vs. competition — your word can be real, spelled perfectly, and still answer a slightly different question than the one asked.
7. The unanticipated loophole. A prompt-bending stretch the editor saw coming lands in the catalog at the punitive 15-point Too Clever tier; one they didn't is simply rejected. There is no published rule for which loopholes pass, which is why chasing them is a lottery.

Reason #6 is the one we can show you first-hand. Playing today's dive on krillion.io while writing this (September 14, 2026), prompt 1 asked us to "name a modern country that Alexander the Great's empire reached." We typed Persia — historically a perfect fit, since Alexander's empire swallowed it whole — and the game rejected it on the spot: red error state, score frozen at 0, depth still 0 meters. Persia answers the history but not the adjective; it isn't a modern country, so it isn't in this prompt's catalog no matter how correct it feels.

One mercy worth knowing: rejection is silent and sterile. The method page says an unknown but well-formed guess "returns a normal rejection without saving the raw text, changing statistics, or advancing the answer screen" — failed guesses don't leak into the data used for future calibration, and they don't cost you anything but time. Our Persia miss behaved exactly like that: no penalty, no record, just a red flash and the clock still running.

How Krillion scoring works: six fixed rarity tiers

The biggest misconception I see — and one most community guides get wrong — is that Krillion recomputes rarity from each day's submissions. It doesn't. Per the method page, "the active catalog version fixes that mapping before players begin," and "the first accepted player receives the same fixed result as later players." Player aggregates are collected (one counted contribution per participant per prompt, stored as privacy-safe hashes), but they only ever inform future, human-reviewed catalog revisions — never a mid-game rewrite.

So the ladder — Plankton 10, Too Clever 15, Schooler 30, Rare 60, Deep Cut 85, One in a Krillion 100, with each point worth 10 meters of depth toward a 700-point, 7,000-meter perfect dive — is editorial, not statistical. The full score table is on our rarity reference page, and the arithmetic behind the daily total is in the scoring-math breakdown. One quirk straight from the FAQ: the 100-point gem isn't chosen on rarity alone — past a certain depth rarity gets hard to disambiguate, so it's also picked "for fun factor." Our own logs bear that out: the gem is occasionally less obscure than some Deep Cuts on the same sheet.

Bottom line: the points and depth in this table are the published, fixed mapping — your tier never depends on when you play or how many people played before you. What you control is only which rung your answer was filed under, so the skill is steering toward answers an editor would file deep. Tier meanings per the official FAQ ↗.

This is what a fixed tier looks like at reveal time. Later in the same September 14 dive we answered a prompt with Stapler, and the game flipped to its verdict card: PLANKTON — "Stapler" — +10 PTS · sink 100m, with the gentle dig "the answer everyone blurts out." The tier, the points and the depth were all decided before we typed a letter; the reveal just reads them back.

Playing today's dive and want to sanity-check an idea before you burn it? Look the word up in the answer checker to see which tier it earned on past prompts, or compare against today's full answer sheet after your dive closes.

Why some prompts reject far more answers than others

Catalog size varies by more than two orders of magnitude with category breadth, and it directly sets your odds of an accidental omission. A constrained prompt like "name a country starting with M" ships with 19 accepted answers; a broad pop-culture prompt runs into the thousands. September 7's "name a Mario character" holds 3,147 — the biggest roster we've logged — while a broad mammals prompt carried 2,502, almost all of them Deep Cuts.

The practical reading: on a 19-answer factual prompt, the catalog is probably complete and a rejection almost always means a scope or spelling problem. On a 3,000-answer pop-culture prompt, the long tail was assembled from wikis and fan lists, and honest omissions are far more likely — which is exactly the case the report-for-review channel exists for.

What to do when Krillion rejects a correct answer

1. Re-read the prompt's exact scope. Most avoidable rejections are category traps — territory for country, character for actor. Fix the scope before the spelling.
2. Respell canonically. Use the most standard reference-work spelling; drop or keep diacritics to match how an encyclopedia lists the entry.
3. Try the other number or form. Singular vs. plural, full name vs. surname — each is a separate catalog key.
4. Check the precedent. The answer checker shows whether your word (or its variants) counted on any logged prompt, which tells you whether you're fighting an omission or a convention.
5. Report real omissions. The method page is explicit that a plausible omission should be reported for human review rather than silently guessed at runtime — fixes ship as reviewed, versioned catalog updates with the same alias-conflict checks as the original release.

Same routine, as a checklist you can copy into your notes and run in the 25 seconds the official homepage currently gives each prompt:

KRILLION REJECTION TRIAGE — run before you blame the game
[ ] Scope: does my answer fit every word of the prompt
    (modern / country / character, not just the topic)?
[ ] Spelling: is this the standard reference-work spelling?
[ ] Form: would singular/plural or full name/surname match better?
[ ] Length: under 80 characters, not punctuation-only?
[ ] Precedent: has this word (or a variant) counted on a past
    prompt? → answer checker
[ ] All five pass and it's still rejected → likely a catalog
    omission: report it for human review, then let it go.

Honest limits: what we know and what we don't

A few sourcing caveats, because precision matters on this topic. The detailed validation documentation (normalization, exact matching, counted-once aggregation) is published on a Krillion-affiliated domain, krillion.live, rather than the main krillion.io FAQ — it is internally consistent with the FAQ and with the behavior we observe daily, but read it as the developer's description of intent, not an audited specification. The player-share percentages attached to tiers elsewhere online (">35% of players land Plankton," and so on) come from a third-party fan guide and are estimates, not official figures — Krillion does not publish per-answer player counts. The distribution data and examples here come from fan-maintained archives, including the daily sheets we log ourselves. And the obvious one: this site is an unofficial companion with no affiliation to the game's developer — when the game and this article ever disagree, the game is right.

Before your next dive: run your riskiest idea through the answer checker → — it shows the exact tier that word earned on every logged prompt, so you can tell a catalog omission from a convention before it costs you a prompt.

Free fallback: drill the triage checklist in practice mode → — unlimited rounds built from real past sheets, no daily limit, no stakes — or peek at today's spoiler-free hints if you just want each tier's answer count before you commit.