GQ-009 · The God Questions Observatory

How should the Great Library make God Questions visible, composable, continuously updated, and connected to evidence, decisions, approved action, and capability learning without becoming the corpus warehouse or execution runtime?

SISO Open Source Foundation

active

researching

Review due · assessed 2026-08-03 · due 2026-08-17

Stand up the operating loop on always-on hardware: a single agent whose domain is improving the Library/Foundry/people-graph system, with a public dossier surface showing data bucket counts, question states, and what it changed and when. Evidence selection must be question-driven given the measured extraction ceiling.

gls:release:df8a236d-b092-4d51-9e17-d1016671db64 · program-metadata-1.1.0+0066cfa

Not released — the selected Release contains question-program metadata, not an answer artifact.

hybrid

Which registry fields, Library projections, evidence interfaces, and action-return contracts should become the standard infrastructure for every SISO God Question?

A cold human or agent can find every standing question, its state, decision target, evidence gaps, falsifiers, watch triggers, and current answer lineage from one generated Library surface. · Existing repositories, books, videos, research, and operational receipts can attach to a question through owned evidence systems without being copied into the public registry. · A question can justify a bounded action candidate and later receive a verified learning return without the Library authorizing or executing the action. · At least one repeated mechanism discovered through a question can be evaluated and promoted into its independently owned capability Work.

The design requires a second question registry or central operational database to remain coherent. · The visible dossier accumulates metadata but cannot improve a real research, routing, or investment decision. · Useful evidence cannot be connected without leaking private corpora, machine paths, client data, or operational topology into the public Library. · Maintaining the question contract costs more attention than the repeated research and decision errors it prevents.

No evaluated end-to-end user flow yet proves that the portfolio view helps a cold operator choose the next valuable investigation. · Claim, contradiction, assumption, experiment, mandate, observation, and learning-return objects do not yet share one proven cross-system receipt contract. · No three-question cohort yet measures authoring overhead, evidence reuse, decision impact, or capability promotion rate. · MEASURED 2026-08-04: extraction does not scale to the corpus. Claim extraction over the 932-book philosophy shelf alone needs ~400M tokens against a 150M budget (267%); the full 41.5M-passage corpus needs ~33,200M (22,133%). The budget covers ~187,500 passages. So evidence selection must be question-driven; 'extract everything and index it' is arithmetically impossible, not merely wasteful. · The retrieval unit is unresolved. Passages are the evidence substrate but not the search surface -- a paragraph is not a thing you can agree or disagree with. Claims (position + grounding quote + confidence) are ~2 orders of magnitude fewer and match the Evidence Engines ingest contract, but the extraction layer that produces them does not exist. · No standing agent operates the loop. Every piece of infrastructure built to date is supply; nothing continuously reads a question, finds bearing evidence, and updates an answer. This is the gap between a warehouse and an organ.

A new God Question cannot express a decision target, falsifier, evidence gap, or refresh condition in the current contract. · Foundry or Knowledge ships a question-addressable export that can replace a manual evidence link. · The first approved experiment returns an Observation Receipt and exposes a missing Library lineage field. · Three heterogeneous questions complete a full research-to-decision loop and justify stabilizing or deleting provisional fields. · A standing agent proposes a change to the Library or Foundry and it survives independent review -- the first evidence that self-improvement is real rather than asserted.

the existing Frontier Question Works, metadata seeds, answer-lineage model, and question-driven research architecture · the generated Research index and individual Question Work reading surfaces · publication-safe Foundry, Knowledge, Evidence Engine, Event, ADR, Release, and Snapshot interfaces · privacy-safe receipts from real research, routing, repository, book, and implementation work

A versioned God Questions operating architecture with a generated Observatory, machine-readable question-program contract, evidence and action-return interfaces, ownership boundaries, acceptance fixtures, and explicit promotion and kill gates.

Re-answer when a real question cannot use the contract, a source or action interface changes, three-question evaluation evidence lands, or measured operating overhead exceeds the value created.

public metadata only

The existing Work, Release, Snapshot, and Event spine can support question-program visibility without a second registry. Scope: Library identity and projection · Falsifier: A heterogeneous question cannot express and validate its current program or answer lineage without a separate source of identity truth. · Depends on: none · Evidence: EC-GQ009-CONSTITUTION, EC-GQ009-CHECKPOINT, EC-GQ009-OPERATOR-PASS, EC-GQ009-ESTATE-RECONCILIATION · Review: 2026-08-02

A compact assumption, evidence, and action-learning projection improves cold-reader reconstruction enough to justify its authoring overhead. Scope: question reconstruction economics · Falsifier: Cold readers or agents cannot reconstruct a decision path more accurately, or the repeated authoring cost exceeds the errors and rediscovery avoided. · Depends on: QA-GQ009-CANONICAL-EXTENSION · Evidence: EC-GQ009-CHECKPOINT, EC-GQ009-ESTATE-RECONCILIATION · Review: 2026-08-02

Typed public references can connect question demand to returned learning without granting approval, execution, or evidence-adjudication authority to the Library. Scope: action and learning authority · Falsifier: A real action-to-learning chain cannot remain reconstructable unless the Library stores private payloads or represents a proposal as approval or observation as accepted truth. · Depends on: QA-GQ009-CANONICAL-EXTENSION · Evidence: EC-GQ009-JOINT-BRIDGE, EC-GQ009-PRIVACY-FAIL, EC-GQ009-ESTATE-RECONCILIATION · Review: 2026-08-02

The pushed constitution preserves question identity, organ boundaries, action authority, learning proposals, and explicit later or rejected scope. Owner: The Great Library of SISO · Supports: QA-GQ009-CANONICAL-EXTENSION, QA-GQ009-OWNER-BOUNDARY · Challenges: none · Observed: 2026-08-02

The exact branch checkpoint passed the complete Library gate and published the authored and generated constitution plus Research navigation. Owner: The Great Library of SISO · Supports: QA-GQ009-CANONICAL-EXTENSION · Challenges: QA-GQ009-METADATA-VALUE · Observed: 2026-08-02

The public Event records the typed, approval-gated bridge while the Library remains limited to safe identity and lineage. Owner: SISO Evidence Engines · Supports: QA-GQ009-OWNER-BOUNDARY · Challenges: none · Observed: 2026-08-02

Independent cold-reader and contract lanes reconstructed the operator path and verified canonical human and agent projections at the exact pushed checkpoint. Owner: The Great Library of SISO · Supports: QA-GQ009-CANONICAL-EXTENSION, QA-GQ009-METADATA-VALUE · Challenges: none · Observed: 2026-08-02

The privacy lane found that free-text metadata and temporal or truth-state edges needed stronger validation; the mixed receipt is verified observation, not accepted truth. Owner: The Great Library of SISO · Supports: none · Challenges: QA-GQ009-OWNER-BOUNDARY · Observed: 2026-08-02

Four public repository observations resolved to two existing Work links, one historical no-change result, and one owner and license-review deferral; no source or registry identity was automatically promoted. Owner: The Great Library of SISO · Supports: QA-GQ009-CANONICAL-EXTENSION, QA-GQ009-METADATA-VALUE, QA-GQ009-OWNER-BOUNDARY · Challenges: none · Observed: 2026-08-02

GQ-009 defines the decision, falsifiers, evidence gaps, and expected answer without authorizing implementation. Owner role: question steward · Status: proposed · Recorded: 2026-08-02T00:33:07+07:00 · Truth: not applicable · Predecessors: none · Assumptions: QA-GQ009-CANONICAL-EXTENSION, QA-GQ009-METADATA-VALUE, QA-GQ009-OWNER-BOUNDARY

The active initiative is represented as a bounded action candidate; approval and execution receipts remain separately owned. Owner role: operating architecture · Status: proposed · Recorded: 2026-08-02T00:33:08+07:00 · Truth: not applicable · Predecessors: AL-GQ009-DEMAND · Assumptions: QA-GQ009-CANONICAL-EXTENSION, QA-GQ009-METADATA-VALUE, QA-GQ009-OWNER-BOUNDARY

Shaan approved a six-hour bounded infrastructure, verification, and closeout envelope; the private authorization payload remains owner-held and only its digest is public. Owner role: decision owner · Status: approved · Recorded: 2026-08-02T01:31:58+07:00 · Truth: not applicable · Predecessors: AL-GQ009-CANDIDATE · Assumptions: QA-GQ009-CANONICAL-EXTENSION, QA-GQ009-METADATA-VALUE, QA-GQ009-OWNER-BOUNDARY

The mixed independent observation passed cold-reader and contract reconstruction but failed privacy hardening, producing a bounded repair requirement rather than an accepted answer. Owner role: independent verifier · Status: verified · Recorded: 2026-08-02T01:58:30+07:00 · Truth: not adjudicated · Predecessors: AL-GQ009-MANDATE · Assumptions: QA-GQ009-CANONICAL-EXTENSION, QA-GQ009-METADATA-VALUE, QA-GQ009-OWNER-BOUNDARY

The Learning Return proposes the privacy and temporal repairs now under verification while keeping GQ-009 researching, publishing no Answer Release, and promoting no capability. Owner role: question steward · Status: no change · Recorded: 2026-08-02T01:59:30+07:00 · Truth: not adjudicated · Predecessors: AL-GQ009-OBSERVATION · Assumptions: QA-GQ009-CANONICAL-EXTENSION, QA-GQ009-METADATA-VALUE, QA-GQ009-OWNER-BOUNDARY

Owner-reviewed estate reconciliation supports the existing identity and authority boundaries, records no new Work or inventory, keeps the deferred candidate outside promotion, and proposes no Answer Release. Owner role: question steward · Status: no change · Recorded: 2026-08-02T05:07:20+07:00 · Truth: not adjudicated · Predecessors: AL-GQ009-OBSERVATION · Assumptions: QA-GQ009-CANONICAL-EXTENSION, QA-GQ009-METADATA-VALUE, QA-GQ009-OWNER-BOUNDARY