Question
A bounded uncertainty whose answer can change structure, priority, or project scope.
Research · 22 August 2026
Research here exists to change what we know, test, preserve, or build.
Each question is tied to a decision that more evidence can still overturn: revise the standing, change the experiment, keep or narrow a boundary, materialize something needed to learn or act, or deliberately leave implementation unchanged.
- open / testing public questions
- 9
- answered / reframed public questions
- 19
- project domains represented
- 11
- dated public arguments
- 36
Start here
Three research results define the current frontier.
Begin with the work that can still change the system, the boundary most recently answered, and the experiment that removed the most proposed machinery.
Most important question under test
When should a finite intelligence expand its interface to Reality instead of only reasoning harder?
Wait for a natural owner-level failure that satisfies an already-frozen strict burden: a real certificate/language or grounded-meta-closure case where the current closure is prospectively explicit and the repair cannot be reconstructed as ordinary composition inside the admitted substrate. Do not manufacture E1b/E2 merely because apparatus exists.
1 supporting publications ↗Recently answered boundaryWhich Agent-era world-model laws survive historical out-of-distribution pressure?
Answered for the HD0–HD9 programme. The frozen 1760–2026 campaign closed with 320 trajectories and 96 Deep Anchors across eight chronological waves. Cross-era recurrence survived only as a typed seven-family structural map plus explicit demotions, merges, replay negatives, and measurement/currentness results; the evidence did not justify ten universal success invariants or eight universal innovation laws.
Read the accepted answer ↗Experiment that most changed the architectureWe Cut 203 Observations to 8. The Agent Still Refused to Edit.
How Harness P5 moved from exhaustive repository traversal toward Agent-owned high-information discovery, why minimizing observation count alone failed, and why the successful result was an evidence-backed refusal to mutate source.
5 min research report ↗Full research index
Browse every active, open, answered, and reframed question.
Use the explorer after the current priorities are clear. It groups the same research by question, project, publication date, or status.
Research atlas
The same work, organized by the question you need to answer.
When should a finite intelligence expand its interface to Reality instead of only reasoning harder?
Distinguish failures of a callable action interface, representation/hypothesis language, physical affordance, installed executor capability, and artifact/grounding so repair happens at the level that is actually closed.
Current judgmentTesting under a contracted theory. External prior art falsifies the claim that action/interface expansion is absent, but it also weakens any absolute transcendence claim: callable, representational, affordance, and executor expansions can be new relative to one closure while remaining compositional inside a deeper supplied substrate. No universal basis hierarchy or generic basis-escape object is admitted.
When should an Agent act, wait, observe, refuse, reconcile, or request a decision?
Treat non-action as a timed, evidence-bearing decision rather than the absence of Agent capability.
Current judgmentTesting with a stronger real-domain non-action case. Finance QB6 let one Agent request exact read-only evidence rather than infer missing execution facts, then update from more_research to no_op when the current minimum contract exceeded both the frozen 10% target and existing 25% canary bound; the branch ended before simulation, Decision, Proposal, or financial write.
Can Agent capital judgment remain separate from financial effect authority?
Pressure quantitative research, proposal/decision semantics, signer and credential authority, one-shot dispatch, and venue reconciliation without turning a backtest or model conclusion into permission to trade.
Current judgmentTesting at a later boundary than the original QB experiments. Finance now preserves real FinancialResources independently of current risky positions, aggregates component portfolio state with explicit FX/temporal evidence, binds owner-managed capital through replayable CapitalState/Exposure context, and blocks both stale state-changing Decisions and unsupported owner-contribution shortcuts. Context v21 routes only work pulled by current Finance consumers into authoritative obligations while external financial submission remains independently fail-closed.
Which game form deserves the next Ordivon Game product—and where do Agents add Player Value?
Search game form before Agent role: use cheap falsifiers and bounded playable proofs to find a form worth playing, then test whether Agent participation creates value a simpler design cannot.
Current judgmentTesting in the Pre-G0 direction-search phase. Station Zero v2 remains the only registered current product; v3, Casefile, and other treatments remain references/research surfaces rather than selected product stages. R1–R29 foundation expansion is provisionally frozen. Since the earlier A/D/I proof wave, v3 Product Value pressure has become independently runnable and currentness-fenced: natural dogfood rejected two attractive but ineffective fixes, then found and corrected a Core extraction successor defect, moving current deterministic Core completion from 1/6 to 5/6 without changing Rescue or Hive basins. This is machine Product Value evidence, not a population or human-experience claim, and no next product has been selected.
How do we know a creative improvement is not just a searched winner or model preference?
Separate richer perception, evidence-grounded meaning, real encounter, typed consequence, search history, and independent holdout before promoting a creative prior.
Current judgmentTesting after R4–R6. Rich perception earned medium-local equipment, grounding proved more stable than fine labels, and R6 showed a 5,000-candidate zero-effect search could manufacture an impressive winner that failed stronger evidence. One reveal-order effect replicated directionally on pristine content for one Agent-observer class.
Can the public site explain changing research judgment without becoming a second fact system?
Use a stable project map, dated publications, Feynman-style causal explanation, and exact owner links while keeping rendered encounter evidence separate from technical truth.
Current judgmentTesting with a direct 2026-08-22 public-currentness audit. Source-bound Game, Harness, and Security projections exposed exact owner review obligations and were refreshed without copying their implementation databases; their current review set is now empty. Manual project judgments for Host, Runtime, World, Finance, Computing, Human, and Media required editorial re-adjudication, and fast-moving Finance drifted again within the same audit because it lacks the public-authority contract used by the source-bound owners. The audit also caught two different downstream currentness failures: a stale static `out/` artifact and smoke tests that still asserted superseded public wording. Home, Now, System, Research, and Writing were then refreshed while dirty/uncommitted owner state was deliberately excluded.
Which Task Runtime objects are required by asynchronous waiting and Join semantics?
Determine the smallest durable objects required when work waits, fans out, and later joins.
Current judgmentOpen. Round 1 showed that mature workflow engines can carry durable work mechanics, but no interrupted fan-out and fan-in workload has yet proved which semantic Join facts must remain visible above those mechanics.
Which repeated physical operation deserves a dedicated structured Runtime contract?
Require stable identity, enforceable preconditions, receipts, replay/reconciliation, and ambiguity behavior before promoting another physical Tool contract.
Current judgmentOpen, but trigger-gated rather than under active expansion. Structured Patch and a small set of Runtime-owned lifecycle operations earned dedicated contracts; the later C1–C10 operational-realization programme did not produce a generic Effect layer, external-consequence authority, cleanup service, or semantic-completion primitive.
When does persistent opponent history earn a Security-owned OpponentModel?
Treat opponent state as a trigger-gated hypothesis: require held-out repeated encounters where history changes decisions beyond current observations and ordinary trajectory before adding durable strategic state.
Current judgmentOpen under the CA7 negative admission gate; no positive transfer standing exists. Earlier Round 1 evidence was mixed and did not establish causal held-out advantage, while CA6 handled deliberate Blue policy change using current consequences and replanning without a stored opponent hypothesis. CA7 therefore leaves persistent OpponentModel NOT ADMITTED and records opponent-history gain as one explicit reopen condition.
Which responsibilities survive materially different workloads without becoming one mandatory stack?
Use unrelated owners and applications to distinguish genuinely recurring responsibility boundaries from repository-local packaging or a frozen Host→Harness→Runtime architecture.
Current judgmentReframed and answered at the responsibility level. Game, Finance, Security, World, Web, Host, Harness, and Runtime now provide materially different consumers and deletion tests. The recurring result is not a universal Host→Harness→Runtime stack: Computing's canonical Core explicitly treats subsystem names as non-invariants and retains only responsibilities that remain unowned after mature mechanisms and owner-native composition. Cross-domain pressure has repeatedly localized, deleted, or kept different packages while preserving a smaller set of ownership/identity/currentness/recovery distinctions.
Which game structures become possible only when Agents persist and act through one World history?
Historical Agent-first research that separated strategy existence, policy accessibility, model realization, and simpler classical substitutes before the Game programme moved to GameForm-first direction search.
Current judgmentReframed after GX1 and the later Game direction work. GX1 showed materially different reachable futures and separated strategy existence from model realization; the model still failed the frozen Core-carrier continuation while a deterministic continuous-carrier policy could succeed, and a generic structured primaryObjective prototype was deleted after presentation-order sensitivity. Those results remain valid research pressure, but the programme no longer assumes that proving an Agent-native mechanic selects the next game.
What semantic work should Host preserve when broader work continues across sessions?
Retain durable Task identity and the bounded semantic frontier without turning Host into a Goal coordinator, executor, Runtime proxy, or domain-completion authority.
Current judgmentReframed by the 0.3 contraction. The original ‘broader Goal completion’ target was too broad: Host deliberately removed shared Goal coordination, caller-neutral foreign execution, Runtime integration residue, generic capability policy, and legacy execution workloads. Current 0.4.x preserves the semantic work claim and revalidation hints while cognition, execution, source currentness, external occurrence, and domain completion remain owner-native.
Which Agent-native responsibilities remain after strong classical baselines?
Use mature workflow, retrieval, idempotency, and Provider systems to delete shared abstractions that do not own a distinct failure.
Current judgmentAnswered within Round 1. Strong baselines reduced the proposed core; later Host and Harness extraction clarified the surviving ownership split without adding a universal Agent platform.
What is the smallest explanation that preserves the causal boundaries an Agent needs?
Compare compact owner-native causal prose with richer cards, relation vocabularies, and explicit diagnostic grammars before promoting another mandatory representation.
Current judgmentAnswered on the EX3–EX7 Agent-facing surfaces. All treatments reached the same exact-action ceiling across 1,326 accepted decisions; richer representations added token burden, so the preregistered smallest-non-inferior rule selected compact prose.
When does valid evidence stop being current enough to govern a decision?
Separate historical validity, current applicability, and downstream commitment; revalidate the exact load-bearing dependency with its owner before a new consequence rather than using age or one global freshness service.
Current judgmentAnswered for the current evidence horizon. Security EC1 separated byte identity from semantic applicability; World independently separates historical occurrence from current relation/Presence; Web source-bound publication distinguishes exact owner revision from current editorial judgment and exposed stale static/test consumers; Finance further showed that an already coherent state-changing Decision loses Proposal authority when its PortfolioSnapshot baseline is no longer current. Computing Knowledge now records both HistoricalValidity != Currentness and AdmissionCurrentness != CommitmentCurrentness. These cases support owner-native revalidation and have not forced a generic Freshness, impact-graph, or automatic rebase service.
Does canonical evidence order make an Agent understand the evidence better?
Separate deterministic byte identity from semantic tractability, and falsify wire/evaluator ambiguity before attributing Agent errors to presentation order.
Current judgmentAnswered for P11. Six unordered datasets collapsed 720 raw permutation digests to one stable identity, but after invalidating a truncated Tool-wire campaign and clarifying an ambiguous evaluator target, raw presentation scored 44/48 strict versus 42/48 stable list and 43/48 identity map. No semantic normalization advantage survived.
Which Agent-era world-model laws survive historical out-of-distribution pressure?
Freeze the modern theory, reconstruct history before coding it, preserve failures and anti-cases, and let independent epochs narrow or contradict the candidate laws before any cross-era promotion.
Current judgmentAnswered for the HD0–HD9 programme. The frozen 1760–2026 campaign closed with 320 trajectories and 96 Deep Anchors across eight chronological waves. Cross-era recurrence survived only as a typed seven-family structural map plus explicit demotions, merges, replay negatives, and measurement/currentness results; the evidence did not justify ten universal success invariants or eight universal innovation laws.
When does a useful research constitution earn default Agent context?
Freeze context treatments before holdout reveal and require replicated marginal benefit before turning persuasive domain knowledge into mandatory prompt state.
Current judgmentAnswered for the current Quant Bedrock default-context claim. Constitution-only looked less bad than control on QB3b but remained negative; the unchanged treatment became materially worse than control on the second sealed QB3c carrier and used about 27.5% more Provider tokens. Mandatory injection has not earned default status.
Which Harness objects survive live Provider replacement without duplicating Host or Runtime state?
Preserve Provider-faithful execution while retaining only shared Run facts required for recovery and completion.
Current judgmentAnswered by H1–H5 and later extraction. The durable Run boundary survived Codex and Hermes replacement; a universal Provider lifecycle did not.
Does the caller-neutral Harness boundary remain valuable across different workloads and execution strategies?
Test whether one bounded Agent-Run authority can preserve cognition, Provider/Tool continuity, recovery, action admission, and completion proposals across callers without becoming a second Task/domain store or universal planner.
Current judgmentAnswered for the current pre-1.0 product boundary. Harness is now an independent, Host-free, caller-neutral durable cognitive execution substrate with its own Journal/CAS and supported HarnessAgentRun path. Real consumer pressure retained exact Run/turn capability separation, WorkingSet/WorkingView versus canonical history, Provider/Tool UNKNOWN recovery, structured completion as proposal rather than domain truth, and bounded capability composition; other candidate machinery such as a universal planner, generic Memory/RAG, Provider-continuation core, capability registry, and automatic knowledge/promotion services was deleted or kept local. The latest research tournament reports no next branch selected: remaining semantic frontiers are blocked on new consumers, owner transitions, or implementation diversity rather than an unresolved proof that Harness has any reusable boundary at all.
Which Provider-required state must survive a Tool turn without becoming Agent cognition?
Separate model-visible semantic history from opaque Provider continuation bytes whose authority comes from an exact prior Provider Tool result and lineage.
Current judgmentAnswered within Harness X2. A live Gemini falsifier returned 200 with the exact continuation and 400 INVALID_ARGUMENT when only the signature was stripped. A 940-byte continuation then survived crash/reopen, remained hidden from Agent messages, completed LIVE_CONTINUATION_OK, and caused zero Runtime redispatch. ProviderToolContinuation was the only new primitive earned across five pressured frontiers.
What operational surface is required for durable semantic Host continuity?
Determine the smallest inspection, recovery, backup/restore, deployment, and MCP surface needed to operate Host without expanding it into an operations platform.
Current judgmentAnswered for the current 0.4.x product scope. Host now has schema-v5 Journal/CAS, full-history validation, Doctor, backup/restore, receipt-bound deployment/rollback, WorkingCheckpoint recovery, and exactly six public MCP Tools. Runtime health, source currentness, Provider state, and domain verification are intentionally not folded into Host operations.
What practical path can increase personal economic autonomy without materially reducing life quality?
Translate first-principles economic reasoning into a conditional, practical path with explicit trade-offs and no individualized advice claim.
Current judgmentAnswered for the first admitted cycle. HUMAN-ECON-001 produced the E0–E9 practical path and six-case review; conclusions remain conditional and do not automatically create E10.
Which remaining friction belongs above Runtime rather than becoming another primitive?
Prevent semantic convenience, environment closure, external-effect policy, or cleanup pressure from expanding physical execution authority without a reproduced unreconstructable fact.
Current judgmentAnswered for the current P0–P5, HP0, and C1–C10 evidence horizon. Storage pressure produced an owner/retention correction rather than a generic cleanup service, and later operational-realization pressure kept currentness, retry, evidence scope, identity, authority, retention, and reclamation separable without reopening Runtime Foundations.
Which Agent-level adversarial distinctions survive after mature security capability is imported?
Separate the Security-owned semantics that remain decision-relevant after ordinary scanners, fuzzers, execution carriers, defensive mechanisms, providers, Host, Harness, and Runtime are consumed directly.
Current judgmentAnswered for the CA0–CA7 capability programme. CA6 showed a fixed script fail three of four held-out worlds while both a generic observation-driven deterministic adaptive policy and the canonical Harness/model Actor succeeded all four with the same oracle-regret vector; P1 reproduced the core adaptive-selection result under real provider/authority friction. The model did not establish a stronger residual than the thin adaptive policy. CA7 therefore closed by contraction: Campaign, Organization, persistent OpponentModel, coevolution, provider gateway, and cross-fidelity strategic law are not admitted by current evidence.
Did a shared World Interaction improve correlated external recovery?
Record the negative result that direct owner-local integration carried the tested external recovery responsibilities.
Current judgmentAnswered negatively. W1 and subsequent experiments did not justify a shared World authority; the active repository retains Provider-native and machine-local capabilities only.
What remains valuable after the shared World layer is removed?
Separate retained operational capability from the rejected semantic project thesis.
Current judgmentAnswered for the current repository. Two capability families remain; the shared semantic authority is rejected and archived.
Which World responsibilities still earn permanent ownership after deletion pressure?
Archive stale active-history material, delete research-only APIs, replay the same mixed failures, and retain only cross-owner reality relations whose absence recreates a current failure.
Current judgmentAnswered for the current HP0–HP8 boundary. Current retrieval improved 7/9→9/9 after historical contraction, research-only ForeignEgress/EffectPath machinery was deleted, and the same 96 mixed-failure invocations passed after deletion.
Research contract
Current judgments may change; dated evidence must not.
Question summaries evolve as evidence accumulates. Publications preserve the complete argument at a date. Repositories, tests, releases, receipts, and observed behavior remain authoritative.
Publication
A dated complete argument that records evidence, limits, conclusions, and the next test.
Source
The owning repository, receipt, release, or report that remains authoritative for exact technical facts.
System context
Questions explain why the architecture is still moving.
Use the System page to follow one work trajectory through the current boundaries. Return here when the useful unit is the uncertainty that could cause a component to change or disappear.
Follow the work trajectory