RIS-1 · technical manual
RIS-1 technical manual
RIS-1, the Resquites Intelligence Scales, First Edition, is a broad, high-range intelligence battery online: 22 subtests and ten indexes, scored to IQ 225 (the full scale to 160+, Extended Reasoning to 225), in three sittings. This page says how it works, how every score is made, what is still provisional, and what its scores can and cannot claim. Updated 20 November 2026.
Status: a research edition, provisional norms
RIS-1 has no norming sample yet. Every score is provisional and model-based: the adaptive subtests are scored on difficulties assumed from how each problem is built, and the other subtests on assumed means and spreads. The first norms will be made from real first attempts, gathered by the norming drive; every score is then computed again from its own answers. Until then a RIS-1 score describes performance on these subtests under a model, not a percentile of a population.
What it measures
Ten indexes of the broad abilities of the Cattell-Horn-Carroll model, each from one to four subtests. The sitting each subtest belongs to is given; Session 1 (the first sitting, eight subtests) gives seven indexes and a short-form full scale.
| Index | Ability | Subtests | Sittings |
|---|---|---|---|
| Gf | Fluid reasoning | KYMO: complete a row that a hidden machine would leave at rest (the machine is made of one to three of thirteen listed moves); THOE: mend an invented world with the fewest changes so that every law holds | 1 |
| Gq | Quantitative reasoning | PLOTO: work out the count in a basin from sluices with hidden shares; DOTO: work out a meter's hidden wheel sizes, then count on or back | 1, 2 |
| Gv | Visual processing | HALIE: the pile that falling pieces make; SPEIO: the pegs a tightened loop touches; MENIPPE: a hidden pattern from glimpses through a moving window; GALATEA: a figure forming out of fog | 1, 2, 3 |
| Gwm | Working memory | ACTAEA: a grid after frames turn its cells over; PHERUSA: an unseen pile after tokens are laid, taken and turned | 1, 2 |
| Gs | Processing speed | EUNIKE, ERATO, EUAGORE: 90-second searches and matches, scored right minus wrong | 1, 2, 3 |
| Gt | Reaction and decision speed | EUDORA: simple reaction, choice reaction and inspection time | 3 |
| Gl | Learning and retrieval | POLYNOE: invented signs for sea creatures, learned a few at a time with the right answer shown after every miss, and asked again at the end of the sitting (attempts begun before 19 November 2026: KYMOTHOE, what glyphs do); PROTO: a short story, asked now and later | 1, 2 |
| Gr | Retrieval fluency | SAO: naming marks as fast as you can, 60 seconds; IONE: as many different figures that keep a rule as you can, 90 seconds | 3 |
| Ga | Auditory processing | LIMNOREIA: play a heard run of tones backward or moved; MELITE: the tone that comes next | 3 |
| Gc | Verbal knowledge | GLAUKE: order words from the widest kind to the narrowest; AGAUE: the word the meaning fits exactly | 1, 2 |
Every subtest is original: its problems are made by a generator, and every reasoning problem's single answer is proven by an independent solver that tries every possibility. Every subtest begins with three practice problems, not scored, each followed by its solution when the answer was not right (POLYNOE: a short practice with two signs).
How the scores are made
- Adaptive subtests (KYMO, THOE, PLOTO, DOTO, HALIE, SPEIO, MELITE, GLAUKE, AGAUE): each problem is chosen near the current estimate of ability; the estimate is the mean of its posterior under a one-parameter model (slope 1.7, a normal prior, each subtest's chance of guessing). The estimate is divided by the square root of 1 minus its error variance, so that the scale keeps its spread of 15.
- Staircases (ACTAEA, PHERUSA, MENIPPE, LIMNOREIA): from a start that follows the reasoning met so far, two levels up at a time until the first error, then one up after a right trial and one down after a wrong one, over 20 trials. The score is the level that best explains every trial (the mean of its posterior), given the spread of 15 as the adaptive subtests are, so it can lie above the top level. (Attempts begun before 19 November 2026: 16 trials from level 3, the mean level after the first four.)
- Timed subtests (EUNIKE, ERATO, EUAGORE, SAO, IONE, GALATEA): right minus a share of wrong within the time; a short tapping check before them measures each device's delay and gives it back.
- Reaction time (EUDORA): the median times of simple and choice reactions, and the shortest glimpse seen, each reversed so that faster is higher.
- Learning (POLYNOE, PROTO; KYMOTHOE in attempts begun before 19 November 2026): right answers now and again later in the sitting. POLYNOE's rounds follow how the examinee is doing: after a round with no miss the next teaches one more new sign (up to five), after a round with two or more misses one fewer (down to two). Its score counts every right answer in the rounds and in the late recall, so a quicker learner, taught more signs, can score higher; on assumed norms until the pilot.
- Each subtest becomes a scaled score (mean 10, SD 3, from 1 to 19). An index combines its subtests, and the full scale combines the indexes (at least seven of the ten), each on the IQ scale (mean 100, SD 15) and held between 40 and 160. A combination is further from 100 than the mean of its parts, as on the Wechsler scales. It is the sum of its parts over the square root of the sum of their correlations, so that every index and the full scale keep a spread of 15. Until the norming data measure them, the correlations are a model's (the provisional norms R1.1: about .41 between indexes, .52 to .62 between the subtests of an index), as are the means and spreads of the subtests that have no model of their own.
Table view
| Score | Percentile | Share at or above | Rarity |
|---|---|---|---|
| 55 | 0.13th | 99.9% | — |
| 70 | 2nd | 97.7% | — |
| 85 | 16th | 84.1% | — |
| 100 | 50th | 50.0% | — |
| 115 | 84th | 15.9% | 1 in 6 |
| 130 | 98th | 2.3% | 1 in 44 |
| 145 | 99.87th | 0.1% | 1 in 741 |
| 160 | 99.997th | 0.003% | 1 in 31,574 |
These are the normal curve's own figures. Until RIS-1 is normed, a RIS-1 score places a person on this curve under a model, not yet against a measured population.
Precision, by simulation (on the problems' assumed difficulties; real data will measure it): KYMO, 20 problems, reliability 0.92; THOE, 16 problems, 0.91; HALIE, 24 problems, 0.92; DOTO, 20 problems, 0.93; the staircases, 20 trials, about 0.97 (without the earlier form's ceiling above +2 SD); the speed subtests about 0.92 on the assumed times.
The charts below come from the same model-based norms (R1.1). The first follows simulated people across the whole range, scored by the site's own code. The second turns each score's reliability into the band that holds a person's true score 95 times in 100. Both describe what the model expects; the norming data will measure them.
Table view
| True level | Complete: error | Complete: average | Session 1: error | Session 1: average | Complete, difficulties off: error | People |
|---|---|---|---|---|---|---|
| 55 | 2.0 | +0.8 | 2.9 | +1.6 | 2.5 | 1,149 |
| 60 | 1.9 | +0.5 | 2.7 | +1.1 | 2.2 | 1,232 |
| 65 | 1.9 | +0.3 | 2.5 | +0.7 | 2.1 | 1,255 |
| 70 | 1.9 | +0.3 | 2.4 | +0.5 | 2.0 | 1,349 |
| 75 | 1.8 | +0.1 | 2.4 | +0.3 | 1.9 | 1,246 |
| 80 | 1.7 | +0.1 | 2.3 | +0.3 | 1.9 | 1,288 |
| 85 | 1.8 | −0.1 | 2.3 | 0.0 | 1.9 | 1,277 |
| 90 | 1.8 | −0.1 | 2.2 | 0.0 | 1.9 | 1,277 |
| 95 | 1.8 | −0.1 | 2.2 | 0.0 | 1.9 | 1,246 |
| 100 | 1.8 | −0.1 | 2.3 | −0.1 | 1.9 | 1,298 |
| 105 | 1.8 | 0.0 | 2.3 | 0.0 | 1.9 | 1,308 |
| 110 | 1.8 | 0.0 | 2.3 | −0.1 | 1.9 | 1,294 |
| 115 | 1.8 | −0.1 | 2.3 | −0.1 | 1.8 | 1,262 |
| 120 | 1.9 | 0.0 | 2.2 | +0.1 | 1.9 | 1,242 |
| 125 | 1.8 | 0.0 | 2.3 | +0.1 | 1.9 | 1,281 |
| 130 | 1.8 | +0.1 | 2.3 | +0.2 | 1.9 | 1,275 |
| 135 | 1.8 | +0.2 | 2.3 | +0.2 | 1.9 | 1,210 |
| 140 | 1.8 | +0.1 | 2.4 | +0.2 | 2.0 | 1,203 |
| 145 | 1.9 | 0.0 | 2.5 | +0.1 | 2.1 | 1,252 |
| 150 | 1.9 | 0.0 | 2.6 | −0.2 | 2.2 | 1,234 |
| 155 | 1.9 | −0.4 | 2.6 | −0.5 | 2.2 | 1,221 |
| 160 | 2.0 | −1.2 | 2.5 | −1.6 | 2.6 | 1,129 |
By simulation of the model-based norms: the error a score would have if the problems and the people behaved as the model assumes; real data will measure it. A score's 95% band is about twice its error. The full scale is held at 160, so true levels above it read a little low there. Average: how far the score lands above (+) or below (−) the true level.
Table view
| Score | Reliability | 95% band | Subtests |
|---|---|---|---|
| Full scale, complete profile | 0.986 | ±3.5 | all 22 |
| Gwm Working memory | 0.984 | ±3.7 | ACTAEA, PHERUSA |
| Full scale, Session 1 | 0.978 | ±4.4 | the first sitting's eight |
| Ga Auditory processing | 0.966 | ±5.4 | LIMNOREIA, MELITE |
| Gv Visual processing | 0.964 | ±5.6 | HALIE, SPEIO, MENIPPE, GALATEA |
| Gs Processing speed | 0.958 | ±6.0 | EUNIKE, ERATO, EUAGORE |
| Gq Quantitative reasoning | 0.953 | ±6.4 | PLOTO, DOTO |
| Gf Fluid reasoning | 0.943 | ±7.0 | KYMO, THOE |
| Gc Verbal knowledge | 0.935 | ±7.5 | GLAUKE, AGAUE |
| Gt Reaction and decision speed | 0.892 | ±9.7 | EUDORA |
| Gl Learning and retrieval | 0.890 | ±9.8 | POLYNOE, PROTO |
| Gr Retrieval fluency | 0.877 | ±10.3 | SAO, IONE |
By simulation of the model-based norms; real data will measure it. In Session 1 most indexes rest on one subtest, so their bands are wider (about ±4.7 to ±9.0). Two indexes differ beyond chance (at 95%) only when they are further apart than √(a² + b²), where a and b are their half-widths: about 1.4 times the half-width when the two are alike.
Integrity, and what verified means
Every sitting has the camera check, which watches on the examinee's own device for a phone, another person or an empty seat and keeps a few still pictures for the author's review. The page also notes what outside help leaves behind: leaving the test window, screenshot, copy, paste and print attempts, AI assistant add-ons. The server keeps its own times. Two checks are RIS-1's own: reasoning far above the timed and memory subtests, which a helper or an AI model cannot do for the examinee; and a timed run whose answers came later than its limit allows. The record never changes a score; it is a reason for the author to review.
The untimed reasoning problems can be solved by an AI model from a photo of the screen, as the site's own screens of the reasoning and word subtests (15 November) and of EUARNE (20 November) found; the timed and memory subtests mostly cannot. So a score is called verified only when the battery is taken under supervision with a clean record, or when a supervised re-check confirms an online attempt: six reasoning subtests on twelve new problems each, by live video with the author, confirming the attempt unless its reasoning is more than two standard errors above the re-check.
Extended Reasoning
The full scale stops at 160. An examinee whose four untimed reasoning subtests (KYMO, THOE, PLOTO and HALIE) together reach 145 or more is offered an extra sitting of harder problems, on eight levels planned at IQ 150 to 225, scored as a separate Extended Reasoning score up to IQ 225+. It has 60 problems: 12 from each of the same four subtests, and 12 from EUARNE, a subtest of solid shapes found only in this sitting. A score at a level's IQ or above must rest on success at that level (three right at it or above, and two thirds right of the problems tried at the level itself). Above about IQ 190 no test has a norming sample; such a score ranks performance on these problems.
EUARNE (added 18 November 2026) is visual-spatial reasoning in three dimensions. Its problems are of four kinds, taken in turn:
- Turn: which objects are the object shown, turned and not mirrored. The objects are arms of cubes joined at right angles.
- Fold: a sheet is folded, punched, and unfolded; where the holes are. At the top levels the punch is a triangle, so each hole also points a way.
- Net: a net of six marked squares is folded into a cube; which views the cube can show.
- Build: which piece completes a block, with pieces turned and never mirrored.
At the first three levels one option of eight is right. From the fourth, every right option must be marked (two to four of eight), or the holes tapped on the unfolded sheet, so a guess almost never succeeds.
Every problem is made by a program and proven by a second solver that works another way: quarter turns compared cube by cube; holes followed forward and unfolded backward; the net folded on its hinges and a cube rolled across it; and an exact search of every way to fill the block. An object is drawn only from a side that shows all of it, so that no other object could look the same.
In simulation (difficulties off their planned places by 0.6 SD, answers less sharp than the model), the sitting's reliability over IQ 145 to 230 is about 0.96, with an error of about 3 to 6 IQ points from IQ 150 to 225. EUARNE leaves that precision about where it was. What it adds is breadth: reasoning about solid shapes, beside HALIE's flat ones. Its levels, like the rest, are planned and not yet measured.
Table view
| True IQ | Planned: error | Planned: average | Harsher: error | Harsher: average |
|---|---|---|---|---|
| 150 | 2.7 | −1.1 | 3.0 | −0.3 |
| 160 | 3.0 | −1.4 | 4.7 | −1.9 |
| 170 | 3.4 | −1.6 | 5.5 | −2.1 |
| 180 | 3.5 | −1.7 | 5.3 | −1.9 |
| 190 | 3.2 | −1.8 | 5.3 | −2.3 |
| 200 | 3.5 | −0.5 | 5.4 | −1.2 |
| 210 | 4.1 | −1.8 | 5.4 | −2.8 |
| 220 | 4.0 | −1.7 | 5.6 | −3.3 |
| 225 | 3.1 | −1.8 | 5.9 | −3.6 |
| 230 | 5.5 | −3.1 | 6.4 | −4.2 |
By simulation (the Extended package's xsim13.mjs); the harsher world has each problem's true difficulty off its planned one by 0.6 SD and answers less sharp than the model (slope 1.3 instead of 1.7). Reliability over IQ 145 to 230, from 1,500 examinees spread over that range: 0.98 at the planned difficulties, 0.96 in the harsher world. Scores read a little low (average below 0). The error dips at 225, the top level's own IQ: a score held under a level by the success rule lands just under that level, which costs least at the level itself. Above 225 no harder problems remain, so scores read low. Above about IQ 190 no test has a norming sample: such a score ranks performance on these problems.
The certificate
A verified score can have a certificate, which the author issues by hand (RI-XXXX-XXXX-XXXX, on the Verify page). It gives the full scale, the indexes and, when it was taken under the same conditions, the Extended Reasoning score. It counts for the Resquites societies, at the higher of the full scale and the Extended Reasoning score, and for the Resquites Register.
Limits
- Provisional norms, as above: scores will change, some by several points, once the norms are made.
- The timed and heard subtests depend on the device: screen, speakers and touch delay vary. The tapping check corrects the touch delay; the pilot will show how well the rest holds.
- Ages 16 to 80. Not a clinical assessment, and not a diagnosis of anything.
- The CHC labels describe what each index is built to measure; whether the subtests group into these ten indexes is checked once the norming data are in.
- How RIS-1 compares with other tests is not yet known. After a finished attempt, an examinee may add up to three scores from other tests; with consent, they are used without names to measure it, and they never change a score.