A bounded experimental protocol that detects an internal signal, formulates a falsifiable question, arbitrates whether to investigate it, produces a read-only proposal, and submits it to independent verification.
19 · Cognition & Control
Endogenous Thought Test
in GenOS.
A bounded experimental protocol that detects an internal signal, formulates a falsifiable question, arbitrates whether to investigate it, produces a read-only proposal, and submits it to independent verification.
DEFINITION AND SCIENCE
What it means
in GenOS.
ETT operationalizes a narrow functional question about internally triggered inquiry. It separates signal detection, falsifiable question formation, attention allocation, bounded proposal generation and independent classification. Observable software states can test that chain; they do not establish subjective experience, intention or consciousness.
Detect and question → Arbitrate and queue → Run under authority → Verify independently → Measure before claiming
The detector uses s=clamp(0,1,0.25a+0.20n+0.30g+0.20l−0.025c−0.025r), then applies separate integrity, mandate, tenant, freshness, budget, cooldown and concurrency gates. The preregistered primary metric is independently verified useful discoveries divided by normalized token, time and USD cost.
ANIMATED DIAGRAM · INTERACTIVE MODEL
01 / 03Explore the mechanism: Endogenous Thought Test
A qualitative teaching model of this concept. It does not run GenOS or validate its implementation.
01Detect and formulate a falsifiable questionINPUT
02Arbitrate and run a bounded proposalMECHANISM
03Classify independently without implicit promotionCONDITIONAL OUTCOME
Current transition · Step 1Detect and formulate a falsifiable question
The modeled transition may proceed. Change the assumption and move through the stages to inspect the decision path.
PROCESS AND USE
How it works
at a glance.
Detect and question
Validate a fresh, in-scope internal signal, then persist an immutable detection receipt and a falsifiable question with method, falsifier and expected evidence.
Arbitrate and queue
AGOW compares investigation with planned abstention. An admitted question becomes a blocked Ontogenesis task and still has no execution authority.
Run under authority
A one-use SHEV authorization permits a deterministic, read-only runner to persist an unverified proposal without invoking a worker or external effect.
Verify independently
The protocol requires a distinct verifier to sign a supported, refuted or inconclusive verdict over the exact proposal. This stage is specified but not present in the pinned published tree.
Measure before claiming
The preregistered campaign must compare the full loop with controls and ablations on reserved seeds before any benefit claim can be evaluated.
RUNTIME IMPLEMENTATION
What is connected,
and how far.
ETT v1 currently implements a deterministic, read-only chain from detection through a persisted unverified proposal. The runner launches no model or worker, performs no network or repository mutation, and grants no implicit promotion. Although the concept document specifies independent classification, its verifier service and test are absent from the pinned published tree. The reserved comparative campaign, real resource measurement, learning transition and production loop are also not implemented. ETT therefore demonstrates a partial auditable control chain, not the benefit claimed by its preregistered hypothesis, phenomenal consciousness or general autonomy.
A deterministic, read-only chain detects an internal signal, formulates and triages a falsifiable question, queues it through Ontogenesis and persists a bounded unverified proposal. The concept document specifies independent classification, but the verifier service and test are absent from the pinned published tree. Learning and the reserved comparative campaign are also not implemented; ETT is not a test of phenomenal consciousness. Source: docs/01-concepts/endogenous-thought-test.md.
A plausible proposal can be mistaken for verified evidence. · An independently classified result can be misreported as support when the verdict is refuted or inconclusive. · Development seeds can leak into threshold tuning and invalidate the reserved comparison. · A functional indicator can be overgeneralized into a claim about consciousness or autonomy.
backend/src/services/endogenousThoughtDetectionService.js↗backend/src/services/endogenousThoughtAdmissionService.js↗backend/src/services/endogenousThoughtQueueService.js↗backend/src/services/endogenousThoughtRunnerService.js↗backend/tests/test_endogenous_thought_contract.js↗backend/tests/test_endogenous_thought_runner.js↗benchmarks/endogenous-thought/protocol-v1.json↗
PRIMARY RESEARCH · MECHANISM LEVEL
The mechanism
behind the analogy.
The design of experiments
Ronald A. Fisher · Oliver and Boyd
Original experimental-design treatment; GenOS comparisons need matched inputs and a declared measurement protocol.
Open publication ↗Primary research describes the cited mechanism; specific technologies may instead cite their original technical documentation. These sources provide context and do not validate the GenOS implementation.
EVIDENCE AND LIMITS
A bounded functional protocol, not a consciousness test
ETT v1 currently implements a deterministic, read-only chain from detection through a persisted unverified proposal. The runner launches no model or worker, performs no network or repository mutation, and grants no implicit promotion. Although the concept document specifies independent classification, its verifier service and test are absent from the pinned published tree. The reserved comparative campaign, real resource measurement, learning transition and production loop are also not implemented. ETT therefore demonstrates a partial auditable control chain, not the benefit claimed by its preregistered hypothesis, phenomenal consciousness or general autonomy.