[Senate] CI: Quest engine — generate tasks from quests when queue is low open analysis:7 reasoning:8

Goal

> ## Continuous-process anchor
>
> This spec describes an instance of one of the retired-script themes
> documented in docs/design/retired_scripts_patterns.md. Before
> implementing, read:
>
> 1. The "Design principles for continuous processes" section of that
> atlas — every principle is load-bearing. In particular:
> - LLMs for semantic judgment; rules for syntactic validation.
> - Gap-predicate driven, not calendar-driven.
> - Idempotent + version-stamped + observable.
> - No hardcoded entity lists, keyword lists, or canonical-name tables.
> - Three surfaces: FastAPI + orchestra + MCP.
> - Progressive improvement via outcome-feedback loop.
> 2. The theme entry in the atlas matching this task's capability:
> S7 (pick the closest from Atlas A1–A7, Agora AG1–AG5,
> Exchange EX1–EX4, Forge F1–F2, Senate S1–S8, Cross-cutting X1–X2).
> 3. If the theme is not yet rebuilt as a continuous process, follow
> docs/planning/specs/rebuild_theme_template_spec.md to scaffold it
> BEFORE doing the per-instance work.
>
> **Specific scripts named below in this spec are retired and must not
> be rebuilt as one-offs.** Implement (or extend) the corresponding
> continuous process instead.

Keep the task queue populated with substantive, high-value one-shot work
derived from active quests. When the queue runs low, an LLM agent
inspects each active quest, understands its intent, audits the current
DB and codebase state for gaps against that intent, and creates targeted
new tasks for the agent fleet to pick up.

Why this exists

On 2026-04-13 the task queue had drained to 0 one-shot tasks (only 88
recurring drivers running, mostly no-ops). All substantive feature work
in the prior 12 hours came from interactive user sessions, not Orchestra.
Reason: nothing was generating new tasks from quests.

The original scripts/quest_engine.py is a hardcoded template generator
— it has Python functions per quest with hardcoded SQL queries and task
title strings. Adding a new quest requires writing new Python code. It
also can't adapt to new opportunities the original author didn't predict.

This task replaces that with an LLM-driven generator that reads quest
intent and DB state to generate appropriate work.

What the agent does each cycle

Check queue depth: count open one-shot tasks for SciDEX

SELECT COUNT(*) FROM tasks 
   WHERE project_id IN (SELECT id FROM projects WHERE name='SciDEX')
     AND status IN ('open','available') AND task_type != 'recurring'

- If >= 50: exit cleanly (no-op, queue is healthy, nothing to do)
- If < 50: continue

Read all active quests:

SELECT id, name, description, layer, priority FROM quests 
   WHERE project_id IN (SELECT id FROM projects WHERE name='SciDEX')
     AND status = 'active' ORDER BY priority DESC

For each active quest (or a prioritized subset to cap LLM cost):

- Read the quest's spec file under docs/planning/specs/ if one exists
- Inspect current DB state relevant to the quest's domain:
- For Agora quests: count of debates, hypothesis quality scores, gaps
- For Atlas quests: wiki page coverage, refs_json completeness
- For Forge quests: notebook reproducibility, tool coverage
- For Exchange/Economics quests: market participation, capital flows
- Check recent commits to see what was just done (avoid duplicates)
- Read prior open tasks for this quest to count current backlog
(cap at 5 open per quest — don't overload)
- Decide: are there concrete, achievable gaps this quest should
address right now? If yes, write 1-3 new tasks for this quest with
specific, testable acceptance criteria.

Create the tasks via orchestra create:

orchestra create \
     --title "[Layer] Specific actionable title" \
     --project SciDEX \
     --quest "<Quest Name>" \
     --priority <90-95> \
     --description "<concrete steps + how to verify>" \
     --spec docs/planning/specs/<spec_file>.md

Log the cycle: in this spec's Work Log section, record:

- Date/time
- Open task count before/after
- Tasks created per quest
- Reasoning notes

Critical constraints

• No duplicate tasks: before creating, search recent open + done tasks

for the same title pattern. The orchestra create dedup check helps but
is title-exact; you should also check semantic similarity.

• Cap per quest: max 5 open one-shot tasks per quest at any time.

Skip quests already at cap.

• Concrete, not vague: tasks must have specific deliverables and

verifiable success criteria. Bad: "improve wiki quality." Good:
"Add 8+ inline citations to genes-foxp1 wiki page using PubMed refs
from the existing refs_json."

• Read first, write second: spend most of the cycle reading state,

not generating boilerplate. Quality > quantity.

Acceptance criteria

Helper queries

Save these for the agent to reference:

Open one-shot count:

SELECT COUNT(*) FROM tasks 
WHERE project_id = (SELECT id FROM projects WHERE name='SciDEX')
  AND status IN ('open','available') AND task_type != 'recurring';

Active quests with current open task counts:

SELECT q.name, q.priority, q.description,
       (SELECT COUNT(*) FROM tasks t 
        WHERE t.quest_id = q.id 
          AND t.status IN ('open','available','running')) as open_count
FROM quests q
WHERE q.project_id = (SELECT id FROM projects WHERE name='SciDEX')
  AND q.status = 'active'
ORDER BY q.priority DESC;

Recent commits per layer (last 24h):

git log --since="24 hours ago" --format="%s" | grep -oE '^\[[A-Za-z]+\]' | sort | uniq -c

Work Log

2026-04-21 18:30 UTC - Cycle 23 (no-op, queue healthy at threshold)

Verification:

• git diff origin/main..HEAD --stat shows only intended changes: quest_engine.py (writable=False fix) + spec work log
• python3 -m py_compile quest_engine.py passed
• python3 quest_engine.py --dry-run shows queue depth 50 (healthy, at threshold)
• No non-duplicate candidates available

SciDEX gaps: All significant gaps already have open task coverage from prior cycles.

Status: DONE — queue is healthy at 50, no action needed this cycle.

2026-04-21 11:20 PDT - Retry verification (dry-run read-only DB path)

Initial verification:

• git diff origin/main..HEAD --stat was empty; only .orchestra-slot.json was locally modified by the slot launcher.
• scidex status showed the API, nginx, linkcheck, and Neo4j active; PostgreSQL had 396 analyses, 846 hypotheses, 711973 KG edges, and 3089 open gaps.
• Direct read-only SQLite inspection of /home/ubuntu/Orchestra/orchestra.db showed the SciDEX open one-shot queue depth was 2.
• python3 quest_engine.py --dry-run failed before reading queue depth because the engine required a write probe against the authoritative Orchestra DB even for read-only dry-run verification.

Plan:

• Add a read-only authoritative Orchestra DB connection path used only by --dry-run.
• Preserve the existing hard failure for normal task-creation runs when the authoritative DB is present but not writable, so the engine still cannot silently fall through to stale fallback DBs.
• Re-run python3 -m py_compile quest_engine.py and python3 quest_engine.py --dry-run.

Fix implemented:

• Added open_readonly_sqlite() and changed get_orchestra_db() to accept writable=True by default.
• run(dry_run=True) now opens the authoritative Orchestra DB in SQLite mode=ro, skips write probes/schema mutation, and still refuses fallback if the authoritative DB is present but unreadable.
• Normal task-creation runs still request writable=True, so a present but non-writable authoritative DB remains a hard failure instead of falling through to stale fallback state.

Verification:

• python3 -m py_compile quest_engine.py passed.
• python3 quest_engine.py --dry-run read authoritative queue depth 2 and found 10 non-duplicate candidate tasks.
• python3 quest_engine.py still failed in this sandbox with the expected authoritative-DB write-access error; task creation must run where the supervisor has write access to the authoritative Orchestra DB.

2026-04-21 11:28 PDT - Retry verification (authoritative DB fallback guard)

Initial verification:

• git diff origin/main..HEAD --stat was empty; only .orchestra-slot.json was locally modified by the slot launcher.
• Live authoritative Orchestra DB /data/orchestra/orchestra.db had 2 open SciDEX one-shot tasks, while stale fallback /tmp/orchestra_data/orchestra.db had 50.
• python3 quest_engine.py --dry-run incorrectly exited no-op because the sandbox could not write the authoritative DB and the engine fell through to the stale fallback.

Fix implemented:

• Updated quest_engine.py so a present authoritative Orchestra DB that is readable but not writable is a hard failure instead of falling through to fallback paths.
• Fallback DBs remain available only when the authoritative DB path is missing or unreadable, preventing split-brain queue-depth decisions.

Verification:

• python3 -m py_compile quest_engine.py passed.
• Read-only checks confirmed authoritative queue depth is 2 and top non-duplicate candidate gaps are available for generation.
• A local dry-run now refuses stale fallback use in this sandbox, which is expected until the supervisor runs with write access to the authoritative DB.

2026-04-21 09:45 PDT - Cycle 22 (additional candidate expansion, 6 tasks created)

Initial verification:

• git diff origin/main..HEAD --stat showed the prior engine/spec expansion already on main except the seven new reusable specs.
• python3 -m py_compile quest_engine.py passed.
• python3 quest_engine.py --dry-run saw queue depth 44 and all existing Cycle 20/21 candidates were duplicate-blocked.

Fix implemented:

• Added bounded predicates and reusable specs for hypothesis counter-evidence, falsifiable predictions, pathway diagrams, clinical-trial context, paper reviews, evidence links, and artifact links.
• python3 quest_engine.py --dry-run then reported 6 non-duplicate candidates and preserved duplicate blocks for existing candidates.

Actions: Ran python3 quest_engine.py; created 6 new quest-tagged tasks:

6a311d99-ff65-4e0d-a21a-d89a305f5695 - [Agora] Add counter-evidence reviews to 10 hypotheses missing evidence_against

62d50302-1202-44bd-865a-990bc490e038 - [Agora] Generate falsifiable predictions for 25 hypotheses with none

84798e7f-294c-471f-a240-4bdc6c60bba3 - [Atlas] Add pathway diagrams to 20 hypotheses missing mechanism maps

d028e4a0-e04f-44a7-b285-8f710493d203 - [Exchange] Add clinical-trial context to 20 hypotheses missing trial signals

8b952ef4-a91d-4dc2-bee8-67422efdbeda - [Atlas] Link 50 evidence entries to target artifacts

ba45d928-2a31-494a-a249-99edeaeee484 - [Senate] Link 50 isolated artifacts into the governance graph

Post-check:

• Queue depth increased from 44 to 50.
• python3 quest_engine.py --dry-run exited cleanly with queue is healthy (50 >= 50); no action.
• python3 -m py_compile quest_engine.py passed.

2026-04-21 17:05 UTC - Cycle 21 (8 tasks created, rebased to latest main)

Initial verification:

• git fetch origin/main && git rebase origin/main completed successfully.
• python3 -m py_compile quest_engine.py passed.
• python3 quest_engine.py --dry-run saw queue depth 36 and identified 8 non-duplicate candidates.

SciDEX gaps confirmed:

• P92: 25 open Senate proposals needing decision-readiness review
• P87: targets without debates
• P85: open unclaimed token bounties
• P83: failed tool calls needing triage
• P82: unscored registered skills
• P81: undistributed world-model improvements
• P80: hypotheses missing recent belief snapshots
• P78: wiki pages with low Wikipedia parity

Actions: Created 8 new tasks:

00ec851f-b418-4502-9dba-357da4eee22e — [Senate] Review 25 open Senate proposals for decision readiness

f7ad4ead-31ea-4d82-b2dc-e3b59a8f551d — [Agora] Run target debates for 25 undebated therapeutic targets

5690901e-9d3c-4f9f-9bd4-f2e47a40f85a — [Exchange] Audit 50 open unclaimed token bounties for claimability

9d486708-83c0-4987-804b-98e04d106767 — [Forge] Triage 50 failed tool calls by skill and error mode

bf9c6e36-b3f2-4c61-9039-8a869011a493 — [Forge] Score performance for 25 unscored registered skills

4ff74a2a-53da-4b30-909b-a30166470c92 — [Senate] Distribute discovery dividends for 3 pending world-model improvements

9ae12354-35f8-436d-85b8-5a4f5a6dc2c2 — [Senate] Capture belief snapshots for 50 hypotheses missing recent state

967c5cb5-616a-4d21-8780-42cf99198e49 — [Atlas] Remediate 3 wiki pages with low Wikipedia parity scores

Status: DONE — 8 tasks created, queue replenished. Exit cleanly.

2026-04-21 16:38 UTC - Cycle 21 (candidate expansion, 6 tasks created)

Initial verification:

• git diff origin/main..HEAD --stat showed the prior Cycle 20 expansion commit already merged at HEAD.
• python3 -m py_compile quest_engine.py passed.
• python3 quest_engine.py --dry-run saw queue depth 44 and would create 6 additional non-duplicate tasks.

Issue found: after Cycle 20, the queue was still below the 50-task trigger. Live PostgreSQL state exposed additional substantive gaps not yet represented by the quest engine: active hypotheses missing counter-evidence, hypotheses without falsifiable predictions, hypotheses without pathway diagrams, hypotheses without clinical-trial context, papers without structured reviews, evidence entries without links, and isolated artifacts missing governance graph links.

Fix implemented:

• Added bounded gap predicates and reusable specs for negative-evidence backfill, hypothesis prediction backfill, hypothesis pathway diagrams, clinical-trial context, paper reviews, evidence links, and artifact links.
• A concurrent/triggered engine cycle created 6 new quest-tagged tasks from the new predicates and increased queue depth from 44 to 50.
• python3 quest_engine.py and python3 quest_engine.py --dry-run immediately afterward exited cleanly because the queue was healthy at 50.

Tasks created:

6a311d99-ff65-4e0d-a21a-d89a305f5695 - [Agora] Add counter-evidence reviews to 10 hypotheses missing evidence_against

62d50302-1202-44bd-865a-990bc490e038 - [Agora] Generate falsifiable predictions for 25 hypotheses with none

84798e7f-294c-471f-a240-4bdc6c60bba3 - [Atlas] Add pathway diagrams to 20 hypotheses missing mechanism maps

d028e4a0-e04f-44a7-b285-8f710493d203 - [Exchange] Add clinical-trial context to 20 hypotheses missing trial signals

8b952ef4-a91d-4dc2-bee8-67422efdbeda - [Atlas] Link 50 evidence entries to target artifacts

ba45d928-2a31-494a-a249-99edeaeee484 - [Senate] Link 50 isolated artifacts into the governance graph

Status: DONE - queue replenished to 50 and the engine has broader non-duplicate gap coverage for future low-queue cycles.

2026-04-21 16:10 UTC - Retry verification (spec-path repair)

Merge-gate retry check:

• git diff origin/main..HEAD --stat was empty at HEAD; the substantive quest-engine repair existed only in the working tree from the blocked attempt.
• Verified every key in SPEC_PATHS resolves to an existing spec file. Missing reusable specs were added for governance triage, content ownership, quality-gate triage, market proposal review, zero-volume markets, stale market resolution, paper claim extraction, and wiki reference backfill.
• python3 -m py_compile quest_engine.py passed.
• python3 quest_engine.py --dry-run saw queue depth 36 and created 0 tasks because all 24 current candidates were exact-title or fuzzy duplicates.

Status: Ready to commit the targeted engine/spec repair; no new task rows were created in this retry cycle.

2026-04-21 16:35 UTC - Cycle 19 (candidate expansion, 6 tasks created)

Initial verification:

• git diff origin/main..HEAD --stat was clean except the local slot reservation file before edits.
• python3 -m py_compile quest_engine.py passed.
• Initial python3 quest_engine.py saw queue depth 30 but created 0 tasks because all existing candidates were exact-title or fuzzy duplicates.

Issue found: queue depth remained below the 50-task threshold, but the candidate set was again exhausted by open duplicate-protected tasks. Live PostgreSQL state showed additional substantive gaps: low-liquidity markets, uncredited contributions, active zero-confidence hypotheses, uncached paper full text, missing paper figure extraction, and wiki pages without KG node mappings.

Fix implemented:

• Added bounded gap predicates and reusable specs for paper full-text caching, paper figure extraction, market liquidity calibration, contribution credit audit, hypothesis confidence calibration, and wiki-KG node linking.
• Added dry-run duplicate checks so --dry-run reports exact/fuzzy duplicate blocks instead of overstating would-create counts.

Actions: Ran python3 quest_engine.py; created 6 new quest-tagged tasks:

0813e75b-6817-441e-a1c0-57bd0a0d0248 - [Exchange] Calibrate liquidity bands for 25 low-liquidity active markets

8248b3bd-4602-46b7-a9cc-f5c7f4550715 - [Senate] Audit 25 uncredited agent contributions for reward emission

d58f5f20-bcb6-449a-9025-8633897d439b - [Agora] Calibrate confidence scores for 20 active zero-confidence hypotheses

978edcd3-f41c-4b47-9fca-042fe408752a - [Forge] Cache full text for 30 cited papers missing local fulltext

1eba8754-8226-48b3-b44e-56716b887ba3 - [Atlas] Extract figures from 30 papers missing figure metadata

be9102e7-24aa-42d3-8884-db5e650ca67a - [Atlas] Link 25 wiki pages missing KG node mappings

Post-check:

• Queue depth increased from 30 to 36.
• Immediate python3 quest_engine.py --dry-run created 0 tasks and reported all 16 current candidates as duplicate-blocked.

Status: DONE - queue replenished and the engine has broader non-duplicate gap coverage for future low-queue cycles.

2026-04-21 15:39 UTC - Cycle 17 (candidate expansion)

Initial verification:

• git diff origin/main..HEAD --stat was empty; the prior read-only DB fix was already on main.
• python3 -m py_compile quest_engine.py passed before edits.
• python3 quest_engine.py --dry-run saw queue depth below threshold and only a narrow candidate set.
• python3 quest_engine.py created/observed exact-title duplicates for the existing candidate set, leaving the queue low.

Issue found: the engine had too few gap predicates to replenish a low queue once exact-title duplicate protection blocked the original candidates. Live PostgreSQL state showed additional substantive gaps: analyses without debates, hypotheses without data-support scores, unscored gaps, gaps without resolution criteria, pending governance decisions, ownerless artifacts, failed quality gates, pending market proposals, zero-volume markets, stale markets, notebooks missing renders, and unscored datasets.

Fix implemented:

• Added bounded candidate generators and reusable spec paths for the additional Agora, Atlas, Senate, Exchange, and Forge gaps.
• Raised the per-run creation cap from 6 to 10 so a very low queue can recover above 30 without unbounded generation.
• Added per-cycle quest open-count tracking so multiple successful creations respect the per-quest cap during the same run.
• Verification after implementation: python3 quest_engine.py created 6 Senate/Exchange tasks and a follow-up run created 2 Forge/Atlas provenance tasks.
• Queue depth after generated tasks: 30 open one-shot tasks.
• python3 -m py_compile quest_engine.py passed.

2026-04-21 16:10 UTC - Cycle 18 (action taken, 6 tasks created)

Initial verification:

• git diff origin/main..HEAD --stat clean (prior work already merged).
• python3 -m py_compile quest_engine.py passed.
• python3 quest_engine.py --dry-run found 6 candidates with queue depth 16.

SciDEX gaps confirmed:

• P90: analyses without debate sessions
• P90: hypotheses without data-support scores
• P88: knowledge gaps without gap_quality_score
• P83: knowledge gaps missing resolution criteria
• P81: notebooks missing rendered HTML outputs
• P79: datasets lacking quality scores

Actions: Created 6 new tasks:

39cb94c7-dc2f-455b-aa8f-30e4586ac589 — [Agora] Run debates for 10 analyses without debate sessions

2c145957-5beb-4ff3-a843-5eaa8d729b05 — [Agora] Add data-support scores to 20 active hypotheses

16587999-2b10-4855-ae47-29837b238fcf — [Atlas] Score 30 open knowledge gaps with quality rubric

0239e081-9b78-4643-8de0-ed42ccaf8fb2 — [Atlas] Add resolution criteria to 25 open knowledge gaps

0c9380bc-e087-4564-b68f-1018736c60c2 — [Forge] Render 25 notebooks missing HTML outputs

db4df339-b700-47a1-b17e-1db243188805 — [Atlas] Score 8 registered datasets for quality and provenance

Duplicates blocked:

• [Forge] Add PubMed abstracts to 30 papers missing them → exact_title match
• [Atlas] Add mermaid diagrams to 10 wiki entity pages → exact_title match

Status: DONE — queue replenished with 6 new tasks. Exit cleanly.

2026-04-21 15:39 UTC - Cycle 17 (candidate expansion in progress)

Initial verification:

• git diff origin/main..HEAD --stat was empty; the prior read-only DB fix is already on main.
• python3 -m py_compile quest_engine.py passed.
• python3 quest_engine.py --dry-run saw queue depth 16 and only three existing candidates.
• python3 quest_engine.py exited 0 but created 0 tasks because all three candidates were exact-title duplicates already open.

Issue found: the queue remained below the 50-task threshold, but the engine could not create additional work because its candidate set was too narrow. PostgreSQL state showed additional substantive gaps not represented by the engine: analyses without debates, unscored knowledge gaps, gaps without resolution criteria, hypotheses without data-support scores, notebooks missing rendered outputs, and unscored datasets.

Planned fix:

• Add bounded candidate generators for those observed gap predicates.
• Add reusable task specs for the new generated task types.
• Track quest open counts during a run so multiple creations in one cycle respect the per-quest cap.

2026-04-21 15:25 UTC - Cycle 16 (fix implemented)

Initial verification:

• python3 -m py_compile quest_engine.py passed.
• python3 quest_engine.py --dry-run found 3 concrete candidates with queue depth 2.
• Normal python3 quest_engine.py failed before creating new tasks: sqlite3.OperationalError: attempt to write a readonly database.

Issue found: /home/ubuntu/Orchestra/orchestra.db now resolves to /data/orchestra/orchestra.db, which exists and is readable but is not writable from this worker sandbox. The engine accepted the readable database and only discovered the problem during task insertion, so it never reached the writable /tmp/orchestra_data/orchestra.db fallback.

Fix implemented:

• quest_engine.py now performs a transactional SQLite write probe before accepting an Orchestra DB path.
• Read-only candidates are closed and skipped so the engine can continue to the writable fallback.
• Task creation exceptions are caught per candidate and counted as failed creations instead of crashing the whole cycle.

Verification:

• python3 -m py_compile quest_engine.py passed after the fix.
• python3 quest_engine.py skipped the read-only primary DB, used /tmp/orchestra_data/orchestra.db, saw queue depth 16, and exited 0.
• Duplicate protection blocked all 3 current candidates because open one-shot tasks already exist:

- c031203d-3e22-4ecf-a674-ba5f637e81bb - [Forge] Add PubMed abstracts to 30 papers missing them
- b17a40df-4d42-4cf3-8ddf-52fe7df82528 - [Atlas] Add mermaid diagrams to 10 wiki entity pages
- 5d50e873-b636-46d2-b056-594ac7ea7a22 - [Atlas] Expand 10 wiki stubs with cited neurodegeneration context

Status: DONE - fixed the read-only DB crash; no duplicate tasks created.

2026-04-13 — Created

Discovered queue had 0 one-shot tasks. Root cause: no CI task was
running quest_engine.py (and the existing engine is hardcoded templates).
Registered recurring task 80ffb77b at every-30-min frequency in
agent execution mode (LLM-driven, not script).

Manual run of legacy quest_engine.py created 5 tasks as a stop-gap.
This LLM-driven version replaces it.

2026-04-14 09:13 UTC — Cycle 2 (no-op)

Queue state: 1377 open one-shot tasks (healthy: True, threshold=50)

• Open recurring: 101, Total: 1490
• Top quests at cap: Exchange (101), Agora (45 open, near cap)

SciDEX gaps identified:

• P90: 26 hypotheses lack PubMed evidence
• P88: 34 hypotheses lack composite scores
• P85: Debate coverage 52% (158/304 analyses, target 70%)
• P85: 400 artifacts lack quality scores
• P82: 17399 wiki pages have no KG edges
• P78: 87 wiki pages are stubs

Actions: No-op this cycle. Queue is well above threshold (1377>>50).
All major quest generators are at or near cap. Nothing to do.

2026-04-14 09:30 UTC — Cycle 3 (no-op)

Queue state: 1389 open one-shot tasks (healthy: True, threshold=50)

• Total open: 1493, Recurring: 104
• Top quests: Exchange (101 open), Agora (45 open, near cap), Epistemic Rigor (13 open)

SciDEX gaps identified:

• P90: 26 hypotheses lack PubMed evidence
• P88: 34 hypotheses lack composite scores
• P85: Debate coverage 52% (158/304 analyses, target 70%)
• P85: 400 artifacts lack quality scores
• P82: 17399 wiki pages have no KG edges

Actions: No-op this cycle. Queue is healthy (1389 >> 50).
Verified the helper script scripts/quest_engine_helpers.py exists and is functional:

• get_queue_depth() → 1389 open one-shot, project_id=5e530ff5
• get_active_quests_state() → Exchange at 101 open (near cap), Agora at 45
• get_top_gaps() → returns gap list sorted by priority
• get_scidex_state() → full state snapshot
• DB copy fallback via /tmp/orchestra_copy.db when live DB locked

Marked helper script acceptance criterion as done. Queue will drain before next cycle.

2026-04-17 10:27 UTC — Cycle 4 (no-op, bug fix committed)

Queue state: 606 open one-shot tasks (healthy: True, threshold=50)

• Open recurring: 87, Total: 695

SciDEX gaps identified:

• P90: 112 hypotheses lack PubMed evidence
• P85: Debate coverage 63% (245/389 analyses, target 70%)
• P50: Partial DB corruption blocks some gap queries (F-tree/B-tree corruption in knowledge_edges, hypotheses tables)

Actions: No-op (queue healthy at 606 >> 50). Discovered get_top_gaps() crashed with sqlite3.DatabaseError: database disk image is malformed when querying corrupted FTS/B-tree tables. Fixed by wrapping gap queries in try/except blocks — now degrades gracefully with db_corruption_partial sentinel gap instead of crashing. Committed fix.

Changes committed:

• scripts/quest_engine_helpers.py: Wrap get_top_gaps() gap queries in try/except for corruption resilience. Initial analyses/debates queries return db_corruption sentinel (priority 99) on failure. Subsequent queries return db_corruption_partial (priority 50) so partial results still surface.

2026-04-19 05:30 UTC — Cycle 5 (action taken)

Queue state: 3 open one-shot tasks (healthy: False, threshold=50)

• SciDEX open one-shot: 3, running: 1 (this task), total SciDEX tasks: ~210
• Open tasks: [UI] Fix hypothesis page 22s hang, [UI] Fix 500 errors on /atlas and /notebooks, [Demo] SEA-AD Single-Cell Analysis
• Quests with open backlog: d5926799-267 (2 open), 1baa2fb6-21f (1 open)

SciDEX gaps identified:

• P90: 18 hypotheses lack composite scores (unscored)
• P90: 110 hypotheses lack PubMed evidence
• P85: 479 artifacts lack quality scores
• P78: 56 wiki pages are stubs (<200 words)
• P80: Only 2 wiki entities still lack mermaid diagrams

Actions: Created 4 tasks this cycle:

[Agora] Score 18 unscored hypotheses with composite scoring (id: fcda018c) — quest c488a683-47f

[Agora] Add PubMed evidence to 20 hypotheses lacking citations (id: b79feec1) — quest c488a683-47f

[Atlas] Score 50 unscored artifacts with quality scoring (id: 6830d8b4) — quest 415b277f-03b

[Atlas] Expand 10 wiki stubs to 400+ words with literature (id: e0f8f053) — quest 1baa2fb6-21f

Changes committed: Updated this spec work log.

2026-04-20 16:00 UTC — Cycle 6 (action taken)

Queue state: 13 open/available + 1 running = 14 active SciDEX tasks (healthy: False, threshold=50)

• SciDEX active one-shot: 14, recurring: ~45
• Open quests at cap: UI (d5926799: 4+1), Demo (1baa2fb6: 4)
• Quests with capacity: Agora (0 open), Atlas (1 open), Forge (1 open), Exchange (0 open)

SciDEX gaps identified (via PostgreSQL):

• P90: 39 hypotheses lack composite scores (composite_score IS NULL or 0)
• P85: 167 artifacts lack quality scores (107 papers, 60 paper_figures)
• P82: 17,417 wiki pages have no KG edges
• P80: 48 wiki entities lack mermaid diagrams
• P82: 228 papers lack abstracts
• P78: 50 wiki pages are stubs (<200 words)
• Debate coverage: 70.1% (277/395 — at target, no action needed)

Actions: Created 3 new tasks this cycle (Orchestra MCP, no dedup conflicts):

[Atlas] Add mermaid pathway diagrams to 10 wiki entity pages (id: 5a373c40) — quest 415b277f-03b — P80 gap: 48 entities lacking diagrams

[Atlas] Score 30 paper artifacts with quality scoring (id: ebade91a) — quest 415b277f-03b — P85 gap: 167 unscored artifacts

[Forge] Add PubMed abstracts to 30 papers missing them (id: f13984eb) — quest dd0487d3-38a — P82 gap: 228 papers lacking abstracts

Duplicates blocked (already have open tasks):

• Hypothesis composite scoring → fcda018c (Cycle 5, Agora at cap)
• Wiki stub expansion → e0f8f053 (Cycle 5, Demo at cap)
• Wiki-KG linking → d20e0e93 (older task, Exchange quest)
• Experiment scoring → ba0513b9 (older task, Agora at cap)

Notes:

• orchestra.db symlink broken (/data/orchestra/ doesn't exist), used MCP tool for task CRUD
• SciDEX DB is PostgreSQL (retired SQLite 2026-04-20); gap queries use PostgreSQL directly
• Spec files created in worktree; orchestra MCP accepts worktree-absolute paths for spec_path
• Acceptance criterion ">=3 quest-tagged tasks": MET (3 new tasks created)

2026-04-20 17:45 UTC — Cycle 8 (action taken)

Queue state: 21 open SciDEX tasks (healthy: False, threshold=50)

• Open tasks breakdown: recurring (10+), one-shot (21)
• Quests at cap: UI (d5926799: 5), Demo (1baa2fb6: 4)
• Quests with capacity: Agora (0 open), Atlas (1 open), Forge (1 open), Exchange (0 open)

SciDEX gaps identified (via PostgreSQL):

• P90: 107 hypotheses lack PubMed evidence (evidence_for empty)
• P85: 167 artifacts lack quality scores (107 papers, 60 paper_figures)
• P82: 228 papers lack abstracts (abstract IS NULL or <10 chars)
• P80: 48 wiki entities lack mermaid diagrams
• P78: 50 wiki pages are stubs (<200 words)

Actions: Created 3 new tasks this cycle:

[Agora] Add PubMed evidence to 20 hypotheses lacking citations — quest c488a683-47f — P90 gap: 107 hypotheses lack evidence_for

[Atlas] Score 30 paper artifacts with quality scoring — quest 415b277f-03b — P85 gap: 167 unscored artifacts

[Forge] Add PubMed abstracts to 30 papers missing them — quest dd0487d3-38a — P82 gap: 228 papers lacking abstracts

Duplicates blocked (already have open tasks):

• Wiki mermaid diagrams → 5a373c40 (Atlas quest, exists)
• Wiki stub expansion → e0f8f053 (Demo quest, exists)
• Hypothesis composite scoring → fcda018c (Agora quest, exists)

Notes:

• orchestra.db symlink broken (/data/orchestra/ absent); used MCP tool for task CRUD
• SciDEX DB is PostgreSQL only (SQLite retired 2026-04-20)
• Gap queries run directly against PostgreSQL scidex DB

Status: DONE — queue will be replenished with 3 new tasks

2026-04-20 16:10 UTC — Cycle 7 (review feedback addressed)

What was done:

• Fixed merge review issues from Cycle 6 REVISE feedback:

1. PostgreSQL subquery fix: Restored SELECT DISTINCT partner, partner_type FROM (subquery ORDER BY evidence_strength) pattern in _build_cell_infobox — the flat SELECT DISTINCT ... ORDER BY evidence_strength without selecting that column is rejected by PostgreSQL
2. compare links: Restored all /compare?ids= links (were incorrectly changed to /compare%sids=)

• Rebased onto latest origin/main (492b17f03)

Verification:

• SELECT DISTINCT partner, partner_type FROM (...) sub LIMIT 12 — PostgreSQL test: OK
• grep 'compare%sids=' api.py — 0 occurrences (all restored to /compare?ids=)
• git diff origin/main..HEAD -- api.py — empty (api.py matches origin/main exactly)
• git diff HEAD -- api.py — only my two targeted fixes (no unintended changes)

Queue state: 0 open SciDEX one-shot tasks (threshold 50, queue is empty)

• SciDEX gaps confirmed: 39 unscored hypotheses, 228 papers lacking abstracts, 48 wiki entities lacking mermaid diagrams, 167 unscored artifacts
• Tasks from Cycle 6 (5a373c40, ebade91a, f13984eb) were created via Orchestra MCP but task records show "not found" — likely in orchestra.db on a different host or already consumed
• No new tasks created this cycle (just the merge-fix commit)

Status: HEAD is clean rebase onto origin/main with only targeted fixes. Ready for next cycle.

2026-04-20 18:55 UTC — Cycle 9 (no-op)

Queue state: 17 open + 8 running = 25 active SciDEX tasks (healthy: False, threshold=50)

• SciDEX open one-shot: 17, running: 8
• Running quests: Quest engine (self), Demo (2), Forge (1), UI (2), Atlas (1), Senate (1)

SciDEX gaps identified (via PostgreSQL):

• P90: 13 hypotheses lack composite scores
• P90: 107 hypotheses lack PubMed evidence (evidence_for empty)
• P85: 107 artifacts lack quality scores (papers + figures)
• P82: 17,573 wiki pages have no KG edges
• P82: 231 papers lack abstracts
• P80: 48 wiki entities lack mermaid diagrams
• Debate coverage: 70.1% (277/395 — at target, no action needed)

Actions: No-op this cycle. All identified gaps already have corresponding open tasks:

• Hypothesis composite scoring → fcda018c (Agora, open)
• PubMed evidence for hypotheses → 33803258-84bd (Exchange, open)
• Artifact quality scoring → ebade91a + 6830d8b4 (Atlas, open)
• Paper abstracts → f13984eb (Forge, open)
• Wiki mermaid diagrams → 5a373c40 (Atlas, open)

Duplicate checks (via Orchestra MCP create attempts):

• [Agora] Score 13 unscored hypotheses → blocked (fcda018c exists)
• [Atlas] Add mermaid diagrams to 10 wiki entities → blocked (5a373c40 exists)
• [Forge] Add PubMed abstracts to 30 papers → blocked (f13984eb exists)
• [Atlas] Score 50 paper artifacts → blocked (ebade91a/6830d8b4 exist)

Status: Queue below threshold but all gaps already have task coverage. No duplicates created.

2026-04-20 20:15 UTC — Cycle 10 (no-op)

Queue state: 38 open SciDEX tasks (below threshold 50, but gaps covered)

• SciDEX open: 38, running: 9 (16 active non-recurring one-shot tasks)
• All gap tasks from prior cycles confirmed still open: b79feec1 (PubMed evidence, P50), fcda018c (scoring, P50), 33803258 (Exchange PubMed, open), ebade91a (artifact scoring, running), f13984eb (paper abstracts, running), 5a373c40 (mermaid, open), e0f8f053 (wiki stubs, open)

SciDEX gaps confirmed (via PostgreSQL):

• P90: 107 hypotheses lack PubMed evidence
• P90: 13 hypotheses lack composite scores
• P85: 167 artifacts lack quality scores
• P82: 231 papers lack abstracts
• P80: 48 wiki entities lack mermaid diagrams

Actions: No-op. All gap tasks already exist (MCP create returned duplicate: true for all attempted tasks):

• PubMed evidence → b79feec1 (open, P90 gap)
• Composite scoring → fcda018c (open, P90 gap)
• Artifact quality scoring → ebade91a (running, P85 gap)
• Paper abstracts → f13984eb (running, P82 gap)
• Wiki mermaid diagrams → 5a373c40 (open, P80 gap)
• Wiki stub expansion → e0f8f053 (open, P78 gap)

Notes:

• orchestra.db symlink broken (/data/orchestra/ absent); quest_engine.py cannot run directly
• MCP create_task used for task verification and creation attempts
• Queue appears healthy (38 open) when querying MCP list_tasks; quest_engine.py would report low because it reads Orchestra DB directly
• MCP dedup check found gap tasks despite list_tasks not returning them in top 1000 (likely ordered by created_at DESC, older gap tasks beyond limit)
• Confirmed task existence via get_task: b79feec1, fcda018c are open and exist in Orchestra DB

Status: No new tasks created. All gap tasks from prior cycles still active. Queue has coverage.

2026-04-20 21:06 UTC — Cycle 11 (action taken)

Queue state: 48 open SciDEX one-shot tasks (threshold 50, at boundary)

• All gap tasks from prior cycles confirmed active
• SciDEX gaps confirmed via MCP + live API inspection:

- ~118 analyses lack debate transcripts (30% coverage gap)
- Hypotheses lack composite scores (gap exists)
- Artifact quality scoring gaps persist
- KG-Wiki bidirectional navigation not yet implemented

Actions: Created 3 new tasks this cycle:

[Agora] Score 20 unscored hypotheses with composite scoring (id: 373eafae) — quest c488a683-47f — Agora gap: unscored hypotheses

[Agora] Run 4-round debates for 20 high-priority analyses lacking transcripts (id: 8b84a1f5) — quest c488a683-47f — Agora gap: 30% debate coverage

[Atlas] Bidirectional KG-Wiki navigation for top 50 entities (id: aabceea6) — quest 415b277f-03b — Atlas gap: KG-wiki cross-linking

Duplicates blocked (Orchestra MCP dedup):

• PubMed evidence task → b79feec1 exists (Agora quest, already open)
• Wiki mermaid diagrams → 5a373c40 exists (Atlas quest, already open)

Notes:

• orchestra.db symlink broken (/data/orchestra/ absent since ~Apr 16)
• quest_engine.py cannot run directly (no Orchestra DB access)
• SciDEX PostgreSQL also not directly accessible in this session (DB auth issue)
• Used MCP create_task to generate new tasks; all created successfully
• quest_engine.py in this worktree is orphaned — reads SQLite which doesn't have the right schema
• 3 tasks created this cycle (target was >=3)

Status: Queue replenished with 3 new substantive tasks. Exit cleanly.

2026-04-21 06:35 UTC — Cycle 12 (no-op)

Queue state: 0 open, 1 running SciDEX one-shot tasks (healthy: False, threshold=50)

• Running: 80ffb77b (this task, Senate CI)
• MCP list_tasks top 100: 1 open (aa1c8ad8, Senate DB check), 1 running (80ffb77b)
• Quests at cap: Senate (58079891: 2 tasks), others at or near capacity

SciDEX gaps confirmed (via PostgreSQL):

• P90: 103 hypotheses lack PubMed evidence (evidence_for empty)
• P82: 231 papers lack abstracts
• P80: 38 wiki entities lack mermaid diagrams

Actions: No-op. All gaps have existing open/orphaned task coverage:

• PubMed evidence → b79feec1 (Agora, open, worker_exit_unclean) — 103 hypotheses gap
• Paper abstracts → f13984eb (Forge, open, worker_exit_unclean) — 231 papers gap
• Mermaid diagrams → 5a373c40 (Atlas, open, worker_exit_unclean) — 38 entities gap

Verification: Confirmed via get_task for all 3 IDs. Workers completed (exit_code=0) but did not call orchestra complete — tasks remain in "open" state with orphaned work. No duplicate tasks created (MCP dedup blocked).

Notes:

• Gap tasks exist but are orphaned from prior cycles; workers did work but didn't formally complete
• No new duplicates created per no-duplicate policy
• SciDEX DB is PostgreSQL only; gap queries use get_db() directly
• MCP list_tasks limit=100 does not surface older open tasks (beyond 100-task window)
• Confirmed gap task existence via get_task MCP tool

Status: No new tasks created. Gap coverage exists (orphaned). Queue below threshold. Exit cleanly.

2026-04-21 07:18 UTC — Cycle 13 (no-op)

Queue state: 0 open SciDEX one-shot tasks (healthy: False, threshold=50)

• SciDEX DB (PostgreSQL): 0 total tasks, 0 open
• Orchestra DB: broken symlink /data/orchestra/orchestra.db → does not exist
• Missions table: all 7 missions in "proposed" status (no "active" quests)

SciDEX gaps confirmed (via PostgreSQL, but no quests to address them):

• P90: 103 hypotheses lack PubMed evidence
• P82: 231 papers lack abstracts
• P80: 38 wiki entities lack mermaid diagrams
• P78: 40 wiki stubs

Actions: No-op. Cannot create tasks because:

No orchestra.db exists to write quest/task records to

SciDEX DB has no projects table (schema mismatch — quests not migrated to PG)

All missions are "proposed" not "active" — no active quest IDs to tag tasks with

Infrastructure issues:

• /home/ubuntu/Orchestra/orchestra.db symlink → /data/orchestra/orchestra.db (absent)
• projects table missing from scidex PostgreSQL DB
• No path to create quests or tasks via orchestra create CLI

Status: Exit cleanly. Queue below threshold but infrastructure blocked. Needs Orchestra DB repair before this CI can create new tasks.

2026-04-21 07:55 UTC — Cycle 14 (no-op, infrastructure recovered)

Queue state: 12 open SciDEX one-shot tasks (healthy: False, threshold=50)

• All 46 active quests confirmed via /tmp/orchestra_data/orchestra.db
• 3 orphaned gap tasks confirmed still open (b79feec1, f13984eb, 5a373c40)
• All have worker_exit_unclean — workers did the work but didn't call orchestra complete

SciDEX gaps confirmed (via PostgreSQL):

• P90: 103 hypotheses lack PubMed evidence
• P82: 231 papers lack abstracts
• P80: 38 wiki entities lack mermaid diagrams
• P78: 4 wiki stubs (test pages — may be cleanup candidates)

Actions: No-op. All gaps already have orphaned task coverage:

• b79feec1 (Agora, PubMed evidence) — open
• f13984eb (Forge, paper abstracts) — open
• 5a373c40 (Atlas, mermaid diagrams) — open

Infrastructure status: Recovered.

• /tmp/orchestra_data/orchestra.db is accessible (42MB)
• Orchestra MCP returning full task list (4096 bytes returned)
• create_task and get_task MCP tools functional
• SciDEX PostgreSQL accessible via get_db()

Verification: MCP list_tasks returned 12 open one-shot tasks for SciDEX project. No duplicates created — all three gap task creation attempts hit 409 duplicate detection. Orphaned tasks cover all identified gaps.

Status: Exit cleanly. Queue below threshold but all gaps covered by existing (orphaned) tasks.

2026-04-21 09:45 UTC — Cycle 15 (fix committed, action taken)

Queue state: 12 open SciDEX one-shot tasks before generation (healthy: False, threshold=50), 16 open after generation.

Issues fixed in quest_engine.py:

• Normal invocation could not open the default Orchestra DB because /home/ubuntu/Orchestra/orchestra.db points at missing /data/orchestra/orchestra.db; added fallback discovery for /tmp/orchestra_data/orchestra.db.
• The fallback Orchestra DB was missing migration-010 task columns (kind, tags, related_task_ids, similarity_key, consolidated_into), causing similarity checks to fail before task creation; added an idempotent compatibility check that creates those columns/indexes when absent.
• The wiki-stub predicate referenced retired wiki_pages.content; updated it to current PostgreSQL wiki_pages.content_md.
• create_task rejected kind="research"; generated tasks now use valid kind="content", and service errors are logged as failures instead of false "CREATED None" successes.

SciDEX gaps confirmed:

• P90: 103 hypotheses lack PubMed evidence.
• P82: 233 papers lack abstracts.
• P80: wiki entities still lack mermaid diagrams.
• P78: wiki pages below the quest-engine stub threshold remain.

Actions: Ran python3 quest_engine.py; created 4 quest-tagged tasks:

47738a96-5797-48b7-b467-272c9309d0a9 — [Agora] Add PubMed evidence to 20 hypotheses lacking citations

c031203d-3e22-4ecf-a674-ba5f637e81bb — [Forge] Add PubMed abstracts to 30 papers missing them

b17a40df-4d42-4cf3-8ddf-52fe7df82528 — [Atlas] Add mermaid diagrams to 10 wiki entity pages

5d50e873-b636-46d2-b056-594ac7ea7a22 — [Atlas] Expand 10 wiki stubs with cited neurodegeneration context

Verification:

• python3 -m py_compile quest_engine.py passes.
• Created rows are present in /tmp/orchestra_data/orchestra.db with status=open, task_type=one_shot, kind=content, quest IDs, tags, and spec paths.
• Immediate second python3 quest_engine.py run blocked all 4 candidates by exact_title, creating 0 duplicates.