Without a distinguishing relation
Three candidates, one key
Akey q
Bkey q
Ckey q
Position, ordering or an extra settle pass cannot create information the field never observed. A tie is the honest answer, and the system abstains.
01 · The thesis
One substrate, one landscape, one settle path.
The homepage now opens with the origin question: what did evolution actually discover? This page is the architecture that follows if that question is taken seriously. Almost every design decision in Hylaean follows from four commitments. They are restrictions, not features. Each one removes a place where a system could look intelligent without being intelligent.
-
01
One substrate
Everything cognitive lives in one continuous state called
state.S. The question, the candidate answer, the world model and every specialised region are slices of the same material. There is no second store, no side memory and no planner holding state of its own. -
02
One landscape
Everything that pushes the state enters one composed energy landscape called
master_energy. A new behaviour has to arrive as an energy term, an operator, a bus signal or a commit gate. Anything else would be a second physics. -
03
One settle path
One production descent, the joint fixpoint pass, moves the state each tick. Bounded candidate programs may compete, but they compete inside that same landscape and the survivor is a settled state, never a choice made by the host.
-
04
One dumb readout
Readout projects the settled state and formats it. It may report abstention. It may not repair, vote, or resolve an ambiguity. If a cleverer decoder could recover the answer without the field, the result does not count.
Why the restrictions are the point. The hardest question to ask of any cognitive system is simple to state and usually impossible to answer: did the system do the work, or did its scaffolding do it? With one substrate and one descent path there is exactly one place a result can come from, so the question becomes an experiment. Switch the mechanism off and the capability has to disappear with it. Every proof on this page is built that way, and several of them came back negative.
Abstention is physics, not politeness.
When the field cannot separate its candidates, no commit forms. That is not a safety policy bolted on afterwards, it is what an unresolved landscape does. The clearest case is also the oldest one on this page.
With a real third relation
Observed structure gives each a role
Akey q₁
Bkey q₂
Ckey q₃
Now the candidates differ because the episode contains a third relation. The distinction comes from what was observed, not from a label supplied by the host.
These are architectural commitments. The category profile below assesses how far the evidence supports them. Historical verification labels remain with their dated experiments.
Hylæan / The shared evaluation
An architecture measured on its own terms.
Modularity, causal closure, plasticity, stability, scaling, efficiency and necessity are separate questions. These are the same category assessments shown in Evaluation.
Read the canonical evaluation profile. The same assessment will appear here when it has loaded.
Historical evidence and research context
This is the earlier research narrative. Dates and claim references identify its scope. Its old ladders and badges are historical instruments, not current category assessments. The shared profile above supplies the current assessment.
02 · Proof state · 18 August 2026
The architecture proof ladder.
D1 to D4 names architecture proof obligations. These scoped results feed the evaluation of architecture and causal autonomy; they do not form an overall intelligence score. Open the shared assessment.
The proof ladder is stricter than a count of experiments. Each rung names one frozen end condition. D1, D3 and D4 are closed. D2 remains open because ARC 2 evaluation is still not measured sealed against the fixed threshold of 12 of 120. Off-seal law tasks read 10 of 120, resident laws 18, hold fulfilled 31 Aug. D3 carries the owner qualification in full: proven under harness attested material, a production attestation is missing.
-
D1
Learn and transfer a lawClosed. Three law forms survive holdout, Fresh Boot and ablation.closed
-
D2
Move ARC 2 evaluationOpen. Sealed field produced remains not measured against the frozen threshold of 12 of 120. Off-seal law tasks read 10 of 120, resident laws 18, hold fulfilled 31 Aug. The representation times law form matrix is measured terminal in every cell. The park that followed ended on 29 August: the named primitive, a field native object to grid paint from resident correspondence, now has a measurement through the production descent path, the first exactly field painted grids, 2 of 3 on the executable carves, default off and not an organ, with the scoped external arm at 0 of 5 and the sealed cell unchanged.open
-
D3
Initiate and close an experience cycleClosed. Full circle 4 of 4, timer control 0 of 4, second world transfer 3 of 3, under harness attested material.closed
-
D4
Replicate independentlyClosed. A content addressed capsule reproduced both declared surfaces.closed
Open the exact proof references and limits
D1: VERIFY.FIELD.TRIADIC_ENDOGENOUS_RELATION_LAW_GLYPH_HOLDOUT_RETRY.01
plus the breadth claim VERIFY.FIELD.ENDOGENOUS_LAW_CLASS_DIVERSITY_TRIAD.01.
D2: VERIFY.ARC.ARC8_PUBLIC_ANCHOR_SEALED_REMEASURE.01, with the
sealed field produced ARC 2 evaluation cell still not measured. Off-seal law tasks read 10 of 120, resident laws 18, hold fulfilled 31 Aug. Seven design censuses on
17 and 18 August measured the complete representation times law form matrix
to honest NO_GO verdicts, from row and grid atoms through object programs
over a proven scene graph. The park that followed ended on 29 August, when the
named primitive, a field native object to grid paint from resident
correspondence, was measured through engine.jfp_wrap at
2 of 3 exact on the executable carves,
default off and not an organ
(VERIFY.ARC.TRANSITION_CHART_SECTION_TO_GRID_RASTER_DESIGN_CENSUS.01,
VERIFY.ARC.OBJECT_TO_GRID_PUSHFORWARD_JFP_WRAP_MINI_DYNAMICS.01)
(VERIFY.ARC.OBJECT_LEVEL_SCENE_REPRESENTATION_DESIGN_CENSUS.01,
VERIFY.ARC.OBJECT_PROGRAM_SCENE_GRAPH_DESIGN_CENSUS.01). D3:
VERIFY.WORLD.AUTONOMOUS_EXPERIENCE_NEED_RECEIPT_TIMER_CONTROL_TRANSFER.01.
D4: VERIFY.REPRO.SEALED_DUAL_SURFACE_CAPSULE.02. The proof manifest
binds this state across the public surfaces. None of these rungs is an FIQ or a
human normed intelligence score.
VERIFY.WORLD.SELF_MOTION_LAW_PERSISTENCE_PRODUCTION_BUILD.01 and
VERIFY.VISION.COVARIANT_CHANNEL_B3B_PRODUCTION_BUILD.01.
Neither is a benchmark score, and the vision numbers are measured under the
declared measurement chain activation, not the production default.
03 · The one loop
Intelligence is the whole circulation.
No station is a hidden second brain. Select a station to see what it contributes. The moving pulse shows the intended circulation, not a claim that every return edge is already closed in production.
Perception becomes a disturbance
Vision, world state, text and audio enter through declared sensory surfaces. These surfaces stimulate the field. They do not reason, remember or select an answer.
Historical experiment: a measured episode through the loop
A bounded recording from the cited run. Its native measurements describe that episode and do not replace the current architecture assessment.
The same loop, once, as it was actually measured
The diagram above is the intended circulation. It does not claim that every return edge is closed in production, and most of them are not. Below is one episode in which the return edge did close: the field formed a law, acted on it, and then formed a different law out of the world its own action had left behind, three times over. Every law, every click and every changed world in the picture is a value read out of that run's own artifact rather than a label placed by hand.
What this does not say: that the loop is closed. One cell, one environment, shipped switched off, and the switched off arm reproduces the landed baseline exactly. Two laws of the same class in a row did not happen, because the harvest of the second law was empty and the collector stays unlicensed. That is the next wall, and it is a question about material rather than about mechanics.
04 · The five layers
One system, with strict boundaries between roles.
The loop above is what happens. The five layers are who is allowed to do it. Each layer is defined as much by what it may not do as by what it does, because that is where the architecture is intended to be enforced. This is a map of responsibilities; current evidence for closure, stability and integrity appears in the shared category assessment.
05
Verification
May measure, control, ablate, refuse and publish a verdict either way.
May not touch the answer path, or count a clean measurement as progress.
04
Communication
May project a settled state, format it, and report abstention.
May not repair an answer, vote between answers, or hold a second reasoning path.
03
Core cognition
May hold operators, energy terms, organs, memory admission, leases and bounded candidate competition.
May not run a second optimiser, a hidden relax loop or an answer repair pass.
02
Perception and material
May encode an observation or a constraint into declared sensory slices. Sources: vision, worlds and games, text, audio.
May not reason, remember, route by task type or select an answer. Each source must prove its own path from observation to settled field material, and they are not equally deep.
01
Substrate and dynamics
Is state.S with the K, L and T operators inside one composed energy landscape, moved by the joint fixpoint pass.
Guarantees that every cognitive write is owned, recorded and auditable. An unauthorised write aborts the run.
The next empirical tests are recorded in the evaluation profile. Each test must show which boundary or capability it changes, under which controls.
Hylæan / The shared evaluation
One field. Six kinds of question.
The map explains what each part is for. Capability and confidence are assessed separately in the shared profile.
The shared cognitive substrateOne field. Different questions.
state.SObservations, stored structure and decisions meet in the same evolving geometry.
Vision
What is here?
Identity across views and transformations.
World
What happens if I act?
Consequences, prediction and intervention.
Grounding
What does it mean?
Connect a representation to its effects.
Memory
What should persist?
Experience changes what the field can retain.
Language
How can it be expressed?
A route from the field to a reportable answer.
Reasoning & transfer
What carries to another problem?
Reuse relations under changed conditions.
Read the canonical evaluation profile. The same assessment will appear here when it has loaded.
Historical evidence and research context
This is the earlier research narrative. Dates and claim references identify its scope. Its old ladders and badges are historical instruments, not current category assessments. The shared profile above supplies the current assessment.
05 · The organ map · 18 August 2026
Six organ questions, one field, one connecting chain.
Each organ answers one question about the world, and every answer must settle in the same field. The ring around them is the autonomous chain: the field notices a need, experiments, seals a concept, stores it, uses it and transfers it. The markers use the vocabulary from the top of this page and none of them is rounded up.
Vision E3 · sealed twice
The covariant content channel makes the instance read transformation tolerant. Armed covariant reads land 315 of 320 across ten geometric variants, from horizontal flip through rotations to zoom, against a paired channel off baseline of 2 of 320, sealed in two independent runs at zero false commits. Honest scope: measured on the declared vision measurement pin, with the production pointer untouched and the channel still default off.
VERIFY.VISION.COVARIANT_CHANNEL_B3B_PRODUCTION_BUILD.01
Build order stage 4
World E3 · live
Law, action and outcome close on the production surface. The hard cell stands at 21 of 24 scored episodes and the easy cell moved from 11 to invalidated of 24 through self motion law persistence, sealed twice at zero false terminals. The causal control is explicit: the same keys without the persisted law fall back to 12 of 24. The residual episodes are measured and parked with named causes.
VERIFY.WORLD.SELF_MOTION_LAW_PERSISTENCE_PRODUCTION_BUILD.01
VERIFY.WORLD.POST_E3_RESIDUAL_EPISODE_ANATOMY.01
Build order stages 2 and 3
Grounding built · off
Concepts form from effect equivalence: five concept classes sealed from the regular world percept transition seam, held out membership separation 1.0 after a genuine fresh boot, and a triple causal ablation. Remove the producer and nothing is acquired. Built and verified, every key default off.
VERIFY.GROUNDING.WORLD_EFFECT_CONCEPT_RECEIPT_PRODUCTION_BUILD.01
Build order stage 3
Memory measured · licensed
The primitive class is measured, not guessed: a conjunctive factor address whose disjointness thesis holds across boots, and an exclusive retrieval lease that scopes both the K drift and the L metric on the rows it owns. On 18 August the last resisting touchstone converted on a second independent strong realisation, which closed the K2 half of the build contract and produced a build licence. That licence is not a build: the slot stays queued behind the question answering channel build.
VERIFY.MEMORY.LEASE_FULL_SCOPE_STRONG_T0_REPLICATION_CENSUS.01
VERIFY.MEMORY.LEASE_TRANSPORT_SCOPE_DESIGN_CENSUS.01
VERIFY.MEMORY.CONJUNCTIVE_FACTOR_BINDING_DESIGN_CENSUS.01
Build order stage 7
Q&A cell measured
The taught association era 220 item cell is legacy with a measured hard ceiling of 29. Its named successor is the structured codomain cell: the organ door chain answers 40 of 46 at zero false as a claim overlay, while the unchanged production decode still commits none of them. Both follow up builds came back red and said why: the workcell build failed at its frozen outcome gates, and the count percept mechanism is proven but stopped at two named channel walls.
VERIFY.QA.STRUCTURED_CODOMAIN_EXTERNAL_CELL_CENSUS.01
VERIFY.QA.STRUCTURED_ANSWER_WORKCELL_PRODUCTION_BUILD.01
VERIFY.QA.FAITHFUL_COUNT_PERCEPT_PRODUCTION_BUILD.01
Build order stage 5
ARC parked
Fully terminally mapped: seven design censuses measured every representation times law form cell to an honest NO_GO at zero false commits. In 52 of 74 same shape cases the real transitions merge, split and transform objects at the same time, which no typed step hypothesis expresses. ARC therefore parks until an external primitive supplies object transition identity, exactly the class the vision class chain or the grounding transfer could produce. Kept and reusable: the object scene graph, the proof carrying licence, the program nucleation machinery.
VERIFY.ARC.OBJECT_LEVEL_SCENE_REPRESENTATION_DESIGN_CENSUS.01
VERIFY.ARC.OBJECT_PROGRAM_SCENE_GRAPH_DESIGN_CENSUS.01
Build order stages 8 and 9
Historical D3 control experiment
The ring is not decoration. It is an inventory of eight links from need to transfer,
of which one exists and four are open
(VERIFY.CHAIN.AUTONOMOUS_CONCEPT_DISCOVERY_END_TO_END_DESIGN_CENSUS.01).
Its first link is genuinely closed: on a teacher built action diverse world the field
issued its own experience need receipts, staged its own experiments and sealed its
first concept from fully autonomously initiated material, with the timer control
producing zero need receipts and zero autonomous attributions
(VERIFY.CHAIN.ACTION_DIVERSE_WORLD_AUTONOMOUS_SEAM_CENSUS.01). The next
link is measured absent today: no consumer reads a stored concept before acting. The
census names the exact seam where that read belongs and shows the signal is already
available, class specific and ablatable
(VERIFY.CHAIN.CONCEPT_TO_ACTION_CONSUMPTION_DESIGN_CENSUS.01). The end to
end composite claim is registered and open
(VERIFY.CHAIN.AUTONOMOUS_CONCEPT_DISCOVERY_END_TO_END.01).
The four arm control behind D3. An experience need receipt appears only when causal residue remains, no active law explains the situation, and a bounded intervention could reduce uncertainty. The decisive comparison is full circle minus timer at equal episode budgets: plus 1.00. Law ablation removes the actions, outcome ablation removes the transfer, and false terminals plus unauthorised writes stay 0 of 0.
One lease class instead of many special cases. A lease is temporary
dynamics authority. A receipt opens a bounded, exclusively owned region of the field.
That region may briefly hold its own local equilibration and may scope its transport
operators, and the lease ends with global semantics restored. It never commits and
never writes persistently. In the World cell the lease mechanism was causally proven
on its claim surface, and the missing primitive it named, self motion law persistence,
became the build that moved the easy cell
(VERIFY.WORLD.TYPED_EXPLORATION_LEASE_PRODUCTION_BUILD.01 Failed,
VERIFY.WORLD.SELF_MOTION_LAW_PERSISTENCE_PRODUCTION_BUILD.01 Verified).
In memory retrieval the same class is the converting mechanism, and the question
answering workcell design builds on it. The owner amendment of 18 August 2026 fixes
its transport scoping form. That is why it appears as the second of the three
primitives below.
Hylæan / The shared evaluation
The next questions are empirical.
A component matters when it changes a controlled outcome. These next tests come directly from the evaluation profile.
Read the canonical evaluation profile. The same assessment will appear here when it has loaded.
Historical evidence and research context
This is the earlier research narrative. Dates and claim references identify its scope. Its old ladders and badges are historical instruments, not current category assessments. The shared profile above supplies the current assessment.
06 · The three primitives · owner review 18 August 2026
The open questions between components.
The organs above fail in different places, and for a long time each failure looked like its own problem. The 18 August owner review reached the opposite conclusion. Every remaining wall across every front resolves into three shared architectural primitives, and the whole programme is now ordered as one twelve stage build sequence through them. This replaces the earlier reading, which treated the gap as a four part address stack.
Directed Causal Binder
A directed, contrastive, law scope bound address that ties a concept to a law. It has to arise from evidence rather than from a label or a host key, and it has to carry the settled target.
- 1Causal contrast binder on GCRGate: 27 of 27 in all orders, separation ratio at least 3, false 0.Measured
- 2Concept coverage consumerA consumer that reads a stored concept before acting.Measured
- 3Autonomous concept circleGate: at least two concepts, fresh boot, second world.Open
- 4Vision effect joinGate: area under curve at least 0.70, out of distribution 20 of 20, false at most 3.Measured
Typed causal lease
The typed exclusive cohort lease, generalised into one carrier principle for counting, deciding and memory scope. One class instead of one special case per organ.
- 5Question answering count leaseTarget cell: 29 of 220.Building
- 6Factorised decision leaseGate: 3 then 6 then 12 commits.Open
- 7Scoped memory leaseGate: paired five boot test with identical seeds, no old item lost to the lease alone, paired net balance at least zero.Licensed
Generative program grammar
Candidates generated from residual directions instead of enumerated from lists. A counterfactual relational program lattice, with ARC training closures as the external cell that judges it.
- 8Counterfactual program latticeGate: at least 5 of 23 real ARC training closures, four candidates drawn from residual directions.Open
- 9Exactly one ARC evaluation sealOne sealed measurement, not a sweep.Open
- 10Runner and profileReentry gate: one run identifier from opportunity through to a second surface.Open
- 11Self discoveryCapabilityProfile self_discovery_v1 active; Self Address persists across boot (E3 cell invalidated/3).Live
- 12External replicationSomeone outside the lab reproduces the chain.Open
Stages 10 to 12 need all three primitives at once, which is why they are one band
rather than a column. Stage 11 is no longer empty prose: the dedicated self discovery
E3 cell and boot persistence of the Self Address are live under production profiles
(VERIFY.SELF.SELF_DISCOVERY_V1_PROFILE_ACTIVATION.01,
VERIFY.SELF.SELF_ADDRESS_PERSISTENCE_FLIP.01). That does not close
stages 10 or 12, and it does not claim the field already proposes its own next research
question.
Where stage 1 actually stands, in full. The binder design census landed on 18 August and returned an honest NO_GO under a result contract frozen before the run. It names three structural findings rather than noise. The whitening formula produces a feature block only axis by construction, so deflating along it removes almost nothing and instead boosts the shared mass it does not touch. The comparison arm that sheds that shared mass reaches 27 of 27 in all five orders with every control green, and fails only the separation ratio at 2.40 against the required 3. The pure address does carry the direction in both pair orders, but its landing ceiling sits just under the untouched floor, and the record admission blend sits under its declared fidelity floor. The consequence recorded by the owner is a combined build recipe, and the missing ratio needs a named new primitive rather than an edited threshold.
VERIFY.BINDER.CAUSAL_CONTRAST_ADDRESS_DESIGN_CENSUS.01
Stage 4 has its material. The world and vision coupling census joined both axes on the same instances with full provenance binding, 40 of 40, which flipped the joint material cell from zero to measurable and put a class probe within reach on held out material. That is a measurability proof on designed teacher material, not a discovered capability, and the census says so.
VERIFY.COUPLING.WORLD_VISION_FORM_EFFECT_JOINT_MATERIAL_DESIGN_CENSUS.01
Failed Two of the rows the folds below cite are landed honest negatives rather than results: the settled identity read found the correct row 27 of 27 and the commit organ still abstained 27 of 27, because no row carried a declared commit target. Both are named with their claim ids inside the second fold.
What happened to the earlier four part address stack
Until this review the same gap was described as a contextual address stack in four transient parts: a scene frame receipt to choose the value free relational frame, a candidate address atom to emit one typed address per candidate, an exchangeable address set to hold several identities without ordering them, and a decision landing lease to reform the decision at the legal late read point. The reading was not wrong about the problem, and one part of it was built and measured honestly.
- 01SceneFrameReceiptChoose the value free relational frame.now inside primitive 03
- 02CandidateAddressAtomEmit one typed address per candidate.built, no production traffic
- 03ExchangeableAddressSetKeep several identities without ordering them.now inside primitive 01
- 04DecisionLandingLeaseReform the decision at the legal late read point.now inside primitive 02
The candidate carrier was built default off and then screened on the production
path. All nine source cases reached opportunity, call, emit and consume, and then the
effect stayed at zero of nine with the production consumer matrix at zero of 162, so
carrier authorisation was suspended
(VERIFY.FIELD.CANDIDATE_ADDRESS_ATOM_PRODUCTION_TRAFFIC_INERTNESS_SCREEN.01).
The three primitives above are what that negative result, together with the ARC
terminal mapping and the memory lease chain, turned into. None of them licenses a
new persistent store, a host planner, a second settle path or a smart decoder.
The earlier building blocks this rests on, in short
Five results in early August widened what a learned law can be and made the proof package fail closed. Three law forms were learned endogenously, one to one, one to many and many to one. Meaning joined across modalities by equal causal effect, 5 of 5 after a fresh boot. A six part rewrite language made typed structure, provenance, bounded quantification and composition executable. A law orbit canonicaliser reduced 18 reordered candidates to 3 stable identities. The public proof envelope gained a semantic roundtrip, a canary, a proof sync and bundle refusal.
The first complete producer to material to consumer chain closed in the lab in the same period. Family B material joined all 27 expected structures, the settled identity read selected 27 of 27 at minimum margin 1.0 with all 162 permutations holding, all 9 foreign cases abstaining and the binary control tying 6 of 6. The boundary was named at once: serve found the correct row 27 of 27 but none carried a declared commit target, so commits abstained 27 of 27. That is a structural proof, not a benchmark lift.
References:
VERIFY.WORLD.TRIADIC_RELATIONAL_ASYMMETRY_GENESIS_FAMILY_B_FULL_GO_CENSUS.01,
VERIFY.QA.TRIADIC_W3A_IDENTITY_READ_CONSUMER.01,
VERIFY.QA.TRIADIC_IDENTITY_SERVE_FIELD_COMMIT_CONSUMER.01,
VERIFY.QA.FAMILY_A_BOUND_W3A_IDENTITY_READ_CONSUMER_REPLICATION.01.
07 · How a claim earns a place on this page
Every claim must be able to fail.
The registry is an audit trail, not a score. A large number of verified claims means the measurements were clean, not that the system improved. This section is the part of the architecture that decides what the rest of the page is allowed to say.
-
01
Register a claim
State the mechanism, the expected movement and the condition that would refute it.
-
02
Build the harness
Bind the claim to the real field path, its controls, its source material and its consumer.
-
03
Measure
Run positive and negative controls. Shuffle, impossible and ablation arms expose false wins.
-
04
Book the verdict
Verified and Failed are both final evidence. A Failed claim names the missing primitive.
-
05
Mirror it publicly
The status log, fronts, research map and progress view are rebuilt from the recorded sources.
What counts as progress
Results are booked on one maturity ladder, and only its top two rungs are progress. Everything on this page is labelled by where it sits.
- E0EvidenceA census, a measurement, a diagnosis.
- E1MechanismA local build with a causal ablation.
- E2Production chainOpportunity through to effect on a regular production surface.
- E3External outcomeA paired improvement of an external result cell.
- E4TransferThe same mechanism improves a second domain.
The rules that protect the difference
These are the standing rules that separate an attractive demonstration from evidence that the field itself did the work.
No teach reporting
A result is never reported as intelligence if the answer was shown immediately before scoring. Taught recall is mechanics. A warmup round that teaches sibling pairs is teaching too, and is labelled as such.
Zero false commits
Scored acceptance gates require zero false commitments. Difficult or impossible cases have to end in abstention, not in a confident guess.
No smart decoder
The readout may project and format a settled state. A benchmark that a cleverer decoder could win is an invalid benchmark, whatever the score says.
The kill rule
If a mechanism reduces to an adapter side proposer, a host side enumerator or a smart decoder, it is rejected immediately and regardless of its numbers. Guard erosion is not traded against results.
Sieves before building
A build has to pass screens before it exists: can the proposed term move the field at all, does its material survive a frozen replay through the unchanged consumer, does the regular seam actually drive that consumer. Each screen is necessary and each one is explicitly capable of a false positive, so a green screen is a licence to try, never a promise of effect.
Predict before you run, and replicate sealed
An authorised build commits its prediction before the first outcome run, and the outcome line is written afterwards even when the answer is that nothing moved. Replication is the same discipline at the level of the whole system: D4 is closed by a content addressed capsule that reproduces the declared frozen cell and ARC device arms from code, checkpoint, seeds and sealed inputs with no undeclared reads. That proves reproducibility of those surfaces, not general intelligence.
VERIFY.REPRO.SEALED_DUAL_SURFACE_CAPSULE.02
How the work is organised
Many experiments move at once without their evidence or verdicts overwriting one another.
Parallel lanes, isolated workspaces
Separate lanes investigate distinct claims and material families at the same time, each with a narrow question and a named acceptance set fixed before it runs, each editing and measuring in its own workspace.
Serial landing
Finished work enters the shared history through one guarded queue that checks that registry rows, sources and generated public mirrors survive the landing together.
Ownership and envelopes
A session ledger shows who owns each front and claim, so a verified result is consumed instead of repeated. Each verdict travels with its configuration, inputs, controls, metrics and file digests, which is what connects a public sentence to the evidence that earned it.
Honest negatives are the point, not an embarrassment. A red result narrows the search: it records the exact missing primitive and stops the same idea from returning under a new name. Three of the six organs above are where they are because a build came back red and said why, and the twelve stage order in the previous section exists because seven ARC censuses in a row said no.
Continue the story
From architecture to evidence, then open fronts.
Read the Vision evidence next, then the science and its falsification rules. Status is the last step: the live proof ladder, outcomes and open fronts.