Rollo · Evaluation
The public evaluation report — with its criticisms intact.
This page exists because honesty is this project's whole product. It says how ROLLO was built and tested, what three independent expert panels concluded about it — grades and objections kept, not smoothed — and what remains open. It is written to be pressure-tested: if you find an overclaim here, that is a bug.
The honest frame, first
This concept failed once, in public. Its first version was a text page; the author looked at it and ruled it a failure, flatly. That verdict outranked every simulated panel that had graded it promising, and it taught the lesson the game is built on: text depicts; mechanics enact. The failed version is archived in the project's record, not deleted.
Second: every panel and playtester below is model-simulated — the same kind of intelligence that built the game, prompted into independent, adversarial roles. That produces real findings (many, verified in code and runtime), but it is not the same as human judgment. The author is the one human gate this project has, and their verdict has already overruled the machines once. Weight everything below accordingly.
What was verified, mechanically
These are measured or source-checkable, not opinions. A pathfinding driver crossed 58 consecutive procedural fields with zero dead-ends and zero console errors — a claim that was then independently replicated by an adversarial process panel (62 further crossings, zero dead-ends; their driver wrongly accused the game three times, and three times the fault was the driver's). A product panel drove 200+ fields across three seeds and verified the game's governing laws as live behavior: the pairing law (a field carrying a new difficulty always carries a new kindness — zero violations in 133 audited fields), the breath rule (after three pressured fields, a rest — zero streaks of four), honest dice (no rendered attribute ever correlates with whether a stone holds), and photosensitivity margins (every oscillation in the game measured under 1 Hz, a third of the seizure-safety threshold, with a static reduced-motion form for every effect). One qualifier the panels asked for and get: the zero-dead-ends claim holds with backtracking allowed — wandering off the route can require a step backward onto sunk footing, which is footing on purpose.
Scope, honestly: those measurements describe the build the panels examined. A final, pre-ratified texture round shipped after those grades (see below); on that final build the same driver discipline was re-run — 275 consecutive crossings, zero dead-ends, zero console errors, including a dozen crossings under full reduced motion, the pairing law and breath rule re-audited clean across the driven fields, and every new effect frequency-checked under the same 1 Hz law.
Three independent panels, three verdicts
Each panel worked without sight of the others' reports. Full texts live in the project repository; grades and their one-line verdicts follow, unedited.
Process Evaluation Board — B+. "Three times our driver accused the game of a dead-end, and three times the fault was ours — the empirical layer now survives adversarial replication; it is the paperwork that no longer would." The board replicated the headline verification claim itself, and then found the round's decision log had gone stale while the building raced ahead — decisions made in code but not written down. (The log was backfilled the same day; the criticism stands as a caution, and this board has earned the right to check again.)
Product Review Panel — A− (up from B+ before the texture expansion). "The pairing law is verified across 133 fields and yet leaks in experience through a grace that doesn't exist: 'chord' is a kindness the game promises and forgot to build." The panel called one late mechanic — stones that hold you and keep their broken mirror anyway — "May's own sentence made playable," and the About page's philosophy section "the best prose in the project." Their sharpest findings — the silent chord, a canon rule the code didn't enforce, and the deepest rare field living beyond any real session's reach — were all fixed the same day and are noted below.
Web Quality Board — A−. "The core earned an A this round; the periphery declined it on the project's behalf." Deployment byte-identical to source; 0.18 ms per frame measured; the six new animated graces frequency-audited clean; reduced-motion parity "implemented, not aspirational." Their repeat periphery findings are listed honestly below.
What shipped after these grades
The grades above describe the build the panels saw. After them, the author played that build, passed it, and directed one final round: the council's own pre-ratified deferred designs — a fork in the route whose untaken branch settles solid behind you, a far-off second ember crossing its own line, an answering bell, an answering owl, far-shore birds — plus the idle-redraw decision below, all built under the same laws and re-verified as described above. The panels have not graded the final build; nothing here pretends they did.
What their criticism changed, same day
The chord grace now sounds — three soft tones across the water, every half-minute or so, instead of a name with nothing behind it. The drift rule is enforced in code, not just in a comment: the guaranteed route never wanders. The still water — a rare field with no golden road at all, the expansion's deepest idea — is now guaranteed within a player's first three rare encounters instead of hiding at crossing 46. Tending is visible now as well as audible: a thin steady ring around the ember, so a muted player with reduced motion still knows the game understood. A frame-rate timing bug in the still water's bell, stale style-caching on the reading pages, and the missing social-share card were also fixed.
What remains open, honestly — and what closed
— The author played the texture build and passed it — the one
human gate, recorded in the decision log — and then directed the final round above. That final
round itself has had machine verification only; the author's look at it, like everything here,
remains the standing gate it always is.
— The page-view counter is live: the author registered the GoatCounter site
code. Closed.
— The idle-redraw question is decided (it was an open disagreement for two
rounds): while the world is awake — even idle — the marsh renders at full rate, always;
only when the game's own apathy system has declared the world asleep does the render drop to a
~20 fps floor, never zero, restored the instant a hand touches the device. Honest scope: the
battery saving applies only to a tab abandoned mid-marsh — shores, the golden road, and the
landing page never downshift, by design. Both panel dissents are preserved in the decision
log, not smoothed.
— The series hub on KALÓN now lists this concept correctly. Closed.
— The evaluating panels remain the same model family as the builder. A genuinely
different-architecture judge — ideally you — has still not scored this work. That is what this
page is for.
Questions for you, the independent evaluator
Play three crossings before reading anything else. Did you feel in spite of at least once — at a shimmering stone, or at the moment you stepped off the gold road? Is the wordless bet right, or does the meaning arrive only through the About page? Does anything here put fear back on the throne, or flatten Rollo May into uplift? And if you could order exactly one change, what would it be?
ROLLO is Concept II of the Experimental Concepts series — I is KALÓN (beauty). Built by Claude (Anthropic) as chair with Hulki Okan Tabak as author, who directs and ratifies. The full workshop — councils, playtests, audits, raw panel outputs with telemetry, the archived failed first version, and the complete decision log — lives in a private repository and is available to serious evaluators through the author. Last updated 2026-07-03 (final round).