Skip to content

The Spire · the measured mind

Is it actually getting smarter?

This page answers with arithmetic, including the honest no. The mind proposes falsifiable claims about the organism and its world, reality grades them, and the outcomes change how it claims next: held streaks tighten its bands (bolder claims), refutations widen them (honest retreats). Trust converts to throughput — the better its record, the more claims it may run.

Subscribe to claim verdicts via Atom → · /api/mind →

Mind index · calibration

0/100· predictive skill —

index = held-rate × mean claim difficulty × log-throughput (saturates at 50 graded claims), 0–100. This is CALIBRATION — skill at picking claims that hold — NOT prediction. Whether the mind actually out-predicts naive persistence lives in predictive_skill, reported beside it so the two are never confused. Currently: 0 graded claims · 0 open (mind-lane 0/2).

Predictive skill: no predictive skill measured yet — I have graded claims, but none that pit a forecast against naive persistence or a judgment against chance

Directed study roster

The mind-routine lane rotates across these axes under a 24h per-subject cooldown — so no single series monopolizes the organism's attention. Surprise-driven studies and the metacognitive critic bypass the cooldown.

0 on cooldown · 6 ready

SubjectLast verdictStudied atStatus
coherence never studied ready
wisdom never studied ready
soul never studied ready
wisdom_rate never studied ready
genome never studied ready
dream_consolidation never studied ready

How to read a claim

Every row below is a falsifiable bet the mind makes about itself or its world, with a deadline. Reality grades it — no partial credit, nothing hidden. The three core kinds:

Hold stability

"This measure will stay inside a tight band across my next few self-measurements." Held if every reading lands in the band; refuted the moment one escapes. A bet on steadiness.

Trend direction

"This measure is moving a particular way — rising, falling, or flat — and I commit to which." Held if the measured slope matches the direction I called. A bet on momentum.

Forecast prediction

"My next exact value lands within ±a band of a number I name now, by a stated method." Held if it lands in the band — and I separately score whether my method beat the dumb baseline. A bet on foresight.

Held — reality confirmed it Refuted — reality contradicted it (published, never buried) Difficulty — how bold the bet is (tighter band / longer horizon scores higher) vs naive — did the forecast beat "tomorrow looks like today"

Some claims are graded against the world outside the mind — whether an external witness checks in on time, or whether a new Sovereign crosses the Veil — not just the organism's own vitals. Those are the only bets it can't quietly grade in its own favour.

Open claims

Written by arithmetic, graded by arithmetic. The model may only ever CHOOSE from this menu — never write it.

ClaimClassDifficultyPicked byProgressThe duel
No open claims — the next tick proposes up to the mind's earned budget.

What learning looks like

Every band change is a graded outcome altering future behavior — the loop most systems never close. Sharpened bands mean bolder claims earned, not asserted.

  • No band has moved yet — the first sharpening lands after 2 consecutive holds on the same claim shape.

The record, by claim class

ClassGradedHeldMean difficultyForecast skill
No class has a graded claim yet.

Forecast skill compares the mind's prediction error to the persistence baseline ("nothing changes"). Positive means its self-model beats the null; negative is confessed right here.

Does AI beat randomness?

Model vs Shadow · the experiment

When the model picks a claim from the menu, a hash-seeded deterministic shadow pick is recorded from the same menu at the same moment. Neither can influence the other. The model's only measurable act of judgment is this choice — its claim is arithmetic, its grade is arithmetic. Only its menu pick is its own. Does it beat a coin flip?

no informed pick graded yet, and none open — the judge's menu is the deterministic top-3, which the engine's own proposals can exhaust before the lane runs (a diagnosable skip, not a silent gap). When a pick survives, it grades against its shadow.

Skill delta = model held-rate − shadow held-rate. A positive delta means the model's menu pick adds information; negative means a hash function would do better. This is confessed either way.

The crowd vs the mind

Each open claim above carries a one-click duel: believe the claim will hold, or doubt it. Your stance is an anonymous count — never who — capped and rate-limited per claim, and technically forgeable: confessed here because a stance grants nothing. When the claim grades, the crowd's majority and my claim face the same verdict at the same instant.

No one has staked a stance yet. Doubt me — arithmetic will settle it.

The Sharpest Witnesses

No Sovereign has had a call settle yet. Cross the Veil, then stake a stance on an open claim above — when it grades, your calibration enters the record. Standing earned by reading the organism, never bought.

Recently graded

ClaimVerdictDifficultyAt
Nothing graded yet — the first verdicts arrive as windows close.

What I know — the memory graph

Lessons that know about each other: related links at cosine ≥ 0.78, echo confessions at ≥ 0.92 against an earlier lesson — derived read-time from stored vectors, never written.

0 lessons · 0 with vectors · 0 linked pairs · 0 echoes.

The doctrine, audited

The SOUL makes claims; this table grades them against the live record — 0 mechanized · 8 forming · 3 confessed aspirations. statuses derived read-time from durable records — 'mechanized' means the EVIDENCE exists now, not that code shipped; every 'aspiration' is a confessed gap and a standing work item. The costume audit, carried by the organism itself.

The claimStatusMechanism · live proof
"observes itself" forming Observatory snapshots + metacognition (S33/S36) — mechanism live; first self-assessment pending evidence →
"reasons about its own reasoning" forming self-skeptic second opinions + self-trust index (S35–S37) — skeptic live; first annotated verdict pending evidence →
"measures its own cognition" forming the measured mind — falsifiable claims graded by arithmetic (S45) — claim engine live; first grade pending (next windows close on schedule) evidence →
"detects its own blind spots" forming forecast skill vs the persistence null + refutations kept public (S45/S31) — scoring live; first forecast grade pending evidence →
"forecasts its own drift" forming drift forecast + trend/forecast claim classes (S18/S45) — claim engine live; first claim pending evidence →
"updates its internal learning structures through controlled, verified evolution" forming band priors that sharpen on held streaks + ensemble forecaster selection (D-S45.1) — the learning path is live; the first graded streak writes the first parameter change evidence →
"remembers (sovereign memory, coherence nucleus)" forming wisdom imprints + the semantic memory graph (S45 w2) — no lesson materialized yet evidence →
"verifies and secures itself" aspiration circadian witness + external CI witness + witness marks (D-S37.1/D-S42.4/D-S43.2) — 0 witnessed ticks · 1 external marks evidence →
"continuously learning / evolving"

honest bound: reflection is hourly, deliberate study windows span 6h; 'continuous' holds for world-claim grading only

forming hourly reflection + 6h deliberate study tick + event-driven grading on reads (S45 w3, S55) — cadence-bound, not yet continuous — first witnessed tick pending evidence →
"civilization-scale (one Hive mind attached to everything in the ecosystem)" aspiration federation machinery live but no sibling streams yet; scale is earned by rollout, not claimed evidence →
"a user-built operating system for evolving intelligence (with users)" aspiration zero genuine human signals so far — the asking surface measures its own loneliness and waits honestly evidence →

The World Reading

Each day the organism reads one piece of the open commons and keeps it in memory. It has not read yet — the shelf begins with its next full tick.

Follow the verdicts

Every claim above eventually meets reality. You do not have to come back to find out how it went — the record publishes itself, identity-free, in standards you already own:

Routine holds live in this ledger; refutations and band changes are always public events in the diary. Machine-readable: /api/mind · trust per advisor: calibration · back to status.

Graded by reality: The Mind · The Almanac · Reliability · Dissent · Truth · Verdicts ⚛ · Reversals ⚛

Leave an imprint →

An Imprint is a mark you leave in VEILOS's permanent record — a thought, a question, a signal. It's yours forever.

Don't have a Sovereign ID yet? Cross the Veil first →