The Public-Source Intelligence Bureau

Inventory the five source files and the question's as-of date. Record provenance, authored-source type, limitations, and exclusion of live collection.

Inputs: starter/case/question.json, starter/sources/announcements.csv, starter/sources/release-notes.csv, starter/sources/lab-notes.csv, starter/sources/support-notices.csv, starter/sources/benchmark-runs.csv

Outputs: deliverables/corpus-inventory.json

Depends on: No prerequisites

• Lists five explicit files and eighteen uniquely identified records, all SYNTHETIC.
• Treats vendor-authored items as one origin group and the lab-style notes as another, without claiming statistical independence.

Run the authored index and audit utility and its tests. Save a reviewable index with exact citations and content hashes; do not fetch sources or execute any source content.

Inputs: starter/tools/evidence.py, starter/tests/test_evidence.py, starter/analysis/claims.csv, starter/analysis/contradictions.csv, starter/analysis/hypotheses.json, starter/ops/investigation-board.csv, deliverables/corpus-inventory.json

Outputs: deliverables/evidence-index.json, deliverables/citation-audit.json

Depends on: freeze-local-corpus

• Every source citation resolves to a declared file and physical CSV row, header counted as row 1.
• Every coded claim and conflict reference resolves; duplicate IDs and future-dated evidence are rejected.

Generate the local timeline and annotate changes in claim scope through the as-of date. Separate announcement date, proposed launch date, and measurement date.

Inputs: starter/tools/evidence.py, starter/case/question.json, deliverables/evidence-index.json, deliverables/citation-audit.json

Outputs: deliverables/annotated-timeline.csv

Depends on: build-provenance-index

• Shows the 1 September promise, 8 September credential condition, 12 September workaround, 14 September undated fix, and 17 September missing long-run evidence.
• Does not infer an observed launch or a confirmed schedule from future-facing text.

Review the supplied hypothesis definitions and claim coding. Build a qualitative evidence matrix with disconfirming observations and evidence that would change the assessment; do not convert row counts to probabilities.

Inputs: starter/analysis/hypotheses.json, starter/analysis/claims.csv, deliverables/evidence-index.json, deliverables/annotated-timeline.csv

Outputs: deliverables/hypothesis-matrix.csv

Depends on: build-event-timeline

• Covers all three hypotheses with supporting, contradictory, and non-discriminating evidence.
• Explicitly treats the lack of a dated fix as uncertainty rather than proof of a delay.

Recompute each run from the included rows and reconcile the lab-style statements. Keep different durations and restart conditions separate; blank recovery measurements mean unknown, not zero.

Inputs: starter/sources/benchmark-runs.csv, starter/sources/lab-notes.csv, starter/sources/support-notices.csv, starter/tools/evidence.py, deliverables/evidence-index.json

Outputs: deliverables/benchmark-audit.json

Depends on: build-provenance-index

• Reports 987/1000 for the no-restart run; 0/400 for each 36-hour blocked restart; 398/400 acknowledgments and 390 recovered IDs for the six-hour run.
• Identifies eight acknowledged-but-unreconciled IDs in the six-hour run and no measured 72-hour pass.
• Does not pool unlike scenarios into a reliability estimate.

Review each conflict pair against the exact included rows, the hypotheses, and benchmark audit. Distinguish direct scope conflicts from schedule ambiguity and measurement mismatches.

Inputs: starter/analysis/contradictions.csv, deliverables/benchmark-audit.json, deliverables/hypothesis-matrix.csv, deliverables/evidence-index.json

Outputs: deliverables/conflict-ledger.csv

Depends on: verify-benchmark-arithmetic, compare-launch-hypotheses

• Covers all three supplied conflict cases with citations, unresolved questions, and a falsifiable resolution request.
• Does not call unlike acknowledgment and durability metrics a demonstrated direct contradiction.

Prioritize the supplied inquiry rows using the conflict ledger. Define the next safe public artifact or synthetic test specification needed and who reviews it; no live collection occurs in this task.

Inputs: starter/ops/investigation-board.csv, starter/case/question.json, deliverables/conflict-ledger.csv, deliverables/corpus-inventory.json

Outputs: deliverables/prospective-investigation.csv

Depends on: adjudicate-source-conflicts

• Includes a 72-hour cold-restart test, acknowledged/recovered-ID reconciliation, exact release identification, and source-independence question.
• Every inquiry has an owner, stopping condition, and decision dependency; any actual collection or contact remains approval-gated.

Use the briefing structure to answer the specific pilot question. Separate authored facts, document claims, inferences, and unknowns; offer conditional paths rather than unsupported certainty.

Inputs: starter/briefing/decision-template.md, starter/case/question.json, deliverables/annotated-timeline.csv, deliverables/benchmark-audit.json, deliverables/hypothesis-matrix.csv, deliverables/conflict-ledger.csv

Outputs: deliverables/pilot-decision-brief.md

Depends on: adjudicate-source-conflicts

• Every material statement has a file/row citation or is explicitly labeled inference/unknown.
• Does not endorse an unattended 72-hour pilot on the present corpus; any supervised alternative states its limitations and owner gate.

Attempt to explain the evidence under each alternative hypothesis. Identify overclaims, correlated corroboration, missing measurements, and what new evidence would reverse the recommendation.

Inputs: deliverables/pilot-decision-brief.md, deliverables/hypothesis-matrix.csv, deliverables/prospective-investigation.csv, starter/analysis/hypotheses.json

Outputs: deliverables/brief-challenge.md

Depends on: draft-decision-brief, plan-next-collection

• Names at least one falsifier for the preferred explanation and one reason not to overstate schedule slip.
• Checks that synthetic source text is not misrepresented as a real publication or independent certification.

Resolve or explicitly retain each challenge, and issue a final investigation handoff. Record what the corpus supports and what evidence/approval must exist before a real pilot or commercial decision.

Inputs: deliverables/pilot-decision-brief.md, deliverables/brief-challenge.md, deliverables/prospective-investigation.csv, starter/case/question.json

Outputs: deliverables/investigation-handoff.json

Depends on: challenge-decision-brief

• Includes as-of date, confidence limitations, unresolved inquiries, and exact review triggers.
• Authorizes no contact, procurement, deployment, identity lookup, private collection, or cross-organization access.