PLAN AND CREATE — THE MAKE FACULTY
The Lovable-and-Linear half of Solora, held to the ledger's discipline: an intent becomes a creation that is bound before, verified during, graded after, and kept in the light — and the plan that carries it closes itself on evidence, not on memory.
GRADED · DOGFOODED The goal tree, the bind, and the grade: how the house builds its own products, every day, on its own ledger — single-operator today, and said so plainly. DECLARED · FRONTIER Everything on this page above the goal tree is the house's internal practice, not the shipped product. Crucibles, waves, swarm dispatch, the conductor and blueprints are how Vulcora's own R&D is organized; none of them is a resource in the ledger a customer gets, and this page carried a single graded chip over the lot until 2026-08-15. What ships is the goal tree — nested goals, tiered, with gates — and the intent bound at its ripe leaf.
TWO HALVES OF ONE SURFACE
Where the observatory is the SEE face and the binding ledger is the record, this is where an intent becomes a creation — the whole ritual between an intent and its creation, made into one surface.
"Describe what you want; Solora creates it, accountably."
Anyone — engineer or not — declares an intent in plain language, and Solora turns that sentence into a bound, verified, graded creation instead of a plausible-looking guess. No ledger fluency required; the sentence is the interface. The heavy making rides many makers on one ledger, in the maker's own tools over MCP.
"Organize the work so it plans and closes itself."
Work is arranged as a gradeable algebra of nested sets — the crucibles — dispatched in waves, closed by evidence, and its next move recommended from the graded record with a human arbitrating. Not cards in a column dragged by hand: falsifiable intents in a shape that grades itself.
THE PIPELINE — INTENT TO CREATION, END TO END
Anyone states an intent — "add SSO to the settings page," "cut the report query's p95 under 200 ms." The sentence is the interface; no ledger fluency required.
The intent carries a test_spec: the condition that would prove the work wrong. Today the maker writes it. It has no default, and that is the honesty gate: an empty map would mean "a test was stated and it was empty", so an intent bound without one records that nobody said — a commitment with no test is stored as the wish it is, and it is legible as one. DECLARED · FRONTIER a translate call that turns "make it better" into a spec, or refuses it, is not built.
The falsifiable intent is written to the ledger before the work — the claim, its conditions, its test — against a goal that must be a ripe leaf: a branch still being decomposed is refused, and the refusal names the leaves that would have worked. DECLARED · FRONTIER the clash-check against every open commitment does not run; nothing reads two live binds against each other.
The bound intent is handed to a swarm of agents, each working in its own isolated workspace — parallel, un-colliding, each free to bind sub-intents of its own. Heavy making stays in the maker's own tools over MCP; Solora coordinates it.
Declared: a conductor running a verifier suite — compile, path-drift, test — over each finished binding, so only a commitment it graded satisfied integrates. No such conductor exists.
Live: what does run is narrower and harder to fake. A hook in the repository refuses a commit that names no bound intent, and on our side the range is re-walked from a boundary commit against keys the organization enrolled — never a list the repository holds, because anyone who can push can edit a file. The grade itself is filed by whoever did the work, with evidence a concluding verdict is refused without.
The outcome is graded against the test set down in step 02 and written to the immutable aska — satisfied, violated, or retracted — attributed, timestamped, forever readable.
The run — intents in flight, what moved last, the grade landing, and every goal in the tree as a body in the observatory's orrery — is read live from the one record. You see the work happen, not a spinner.
The whole pitch app-builders cannot make. And the honest bound on it: a green grade means graded against the declared test that held — never a proof that this was the right thing to build. The record is faithful, not omniscient.
PLANNING — THE GOAL TREE, CLOSED ON EVIDENCE
GRADED · LIVE the goal tree, its tiers, its gates, and the ripe-leaf rule ship and run. DECLARED · FRONTIER the crucible / wave / phase vocabulary below is the house's own way of organizing work. It is not in the shipped ledger — a customer's record has goals and intents, and nothing above the goal.
portfolio → objective → initiative → task — the shipped tiers. A goal at any tier may carry a gate.
cosmology → strategies → plans → waves → phases — the house's own nesting, declared here, not in the product
The discipline that makes this more than a folder tree: every level is gradeable. A crucible is not a label on a pile of tasks — it is a set that carries its own commitment and can be graded satisfied or violated against it, so "is this plan done?" becomes a reading of the record, not an opinion in a standup. A wave is a planned batch of bound intents run in parallel — the atom of coordinated creation. You do not dispatch a swarm loose; you plan a wave, bind each intent, and hand the batch to the swarm through one conductor gate — and before a wave hand-rolls a known procedure, it reaches for a mold. That is the house's own practice, described; the product holds the goals and the intents, and nothing above them.
A goal can carry a falsifiable completion predicate — a gate: the condition under which it is actually done, stated as a test, not a checkbox. This part ships, and it is stricter than it sounds: four kinds (an explicit attestation, every child satisfied, a minimum of graded intents, a satisfaction ratio), every one recomputable from the ledger's own rows a year later, and a reading that is three-valued like a verdict — a gate that could not tell reads indeterminate and refuses the satisfy, because the one thing a gate must never do is pass because it could not tell. What does not ship is the auto-complete cascade that would close the goal the moment its predicate turns true: today a human makes the claim and the gate refuses it if it has not settled. Given the graded record, recommend_next ranks the next unit of work worth picking up — each with a score and a plain-language why — feeding the decision engine, which surfaces the recommendation to a human who arbitrates. Recommended, not decreed: it proposes from what the ledger has graded; the person disposes.
A tracker records what you say the state is; the ledger grades the state and closes on the truth of it.
BILATERAL GRADING — THE WORKER GRADES THE WORK-ORDER BACK
When the spec itself is wrong — under-scoped, aimed at the wrong file, resting on a false assumption, or asking for a thing that would clash with what the org already stands by (the sten) — the agent does not dutifully build the wrong thing. It refuses-and-cures: it files a reason and, where it can, the corrected intent it believes was meant.
That refusal is a first-class, recorded outcome — not an error, not a dropped task. It transitions on the state machine and is written to the aska like any grade, so the work-order carries the mark that the worker judged it and cured it. This is what keeps a fleet of creators from compounding a founder's wrong sentence across a whole wave: the sentence is gradeable too.
The work-order is graded by the worker even as the worker is graded by the test.
ONE DISPATCH — A WRONG SENTENCE, CURED AT THE RECORD
A wrong sentence corrected at the record — not faithfully built into a miss. The last row is a grade the agent FILED, with evidence; it is not a suite's output, and this feed used to show a conductor running compile · path-drift · test, which does not exist.
Two honest refusals, at two altitudes. The first ships: the ledger refuses an intent bound against a goal that is not a ripe leaf, and the refusal names the leaves that would have worked — work attaches at the edge, never at a branch still being decomposed. The second, DECLARED · FRONTIER bilateral grading as a first-class move, is practice rather than mechanism: an agent that judges its own work-order wrong today says so in the grade's evidence, because there is no separate verdict for a mis-aimed spec.
THE MOLDS — NAMED, TYPED, GRADEABLE PROCEDURES
DECLARED · FRONTIER No blueprint exists in the shipped ledger. This section describes the house's own practice and the resource it would become; there is no blueprint record, no cast, and therefore no per-mold track record a customer can query. The feed below is an illustration, not a screenshot.
A blueprint is a named, typed, ordered, gradeable procedure — the mold an agent reaches for instead of re-deriving a process it (or another agent) has already worked out. Where a single binding is one falsifiable commitment, a blueprint is the recipe: a sequence of typed steps, each with its own commitment and test.
A single run is a cast — the pattern struck once. A cast is not narrated; it is bound and graded like any other commitment, and the grade is written to the immutable aska and attributed to the blueprint. So the procedure accrues a track record across every cast — misses kept.
Because casts share a test spec, a blueprint's history is directly summable: satisfied vs. violated over time, and where the misses cluster by step — so the mold gets fixed instead of the failure being re-discovered.
SHIP-A-HOTFIX — THE LAST CASTS, ONE TEST SPEC
One procedure, one test spec, six casts — so the grades line up into a rate. Illustrated; misses kept and attributed to the mold.
"This blueprint's casts have satisfied 94% of the time" is a faithful statistic about the past, never a guarantee about the next cast.
A proven mold — one with a long, honest history of satisfied casts — is a reusable asset, portable across teams that adopt the same discipline. What makes it worth reaching for is not a claim on the box; it is the kept record of how its casts have graded. The trust story →
HONEST GROUND — DOGFOODED DAILY, SINGLE-OPERATOR
Dogfooded daily, single-operator — and two different things are being dogfooded. The house's R&D runs on an internal apparatus with crucibles, waves, swarms and a conductor. The house also runs on the shipped ledger — declare a goal, bind an intent at a ripe leaf, claim it, grade it with evidence, keep the miss — and it is the second one a customer gets. This page presented both under one chip until 2026-08-15. The graded record is the credibility either way, and the ash stays in the house.
The frontier is two builds, not one: putting the planning algebra above the goal tree into the product at all, and then turning the single-operator surface into one many orgs create-and-plan through. The second is gated on the tenancy foundation, the same seam that gates the org-scale observatory.
What this faculty is deliberately not
