dd75a67964
Brings the bureau from 4 → 8 detectives. All eight run as Bun + claude-CLI
subprocesses against the same Supabase + investigation_jobs LISTEN/NOTIFY
queue, sharing search.ts hybridSearch and writer-side validators that
gate writes against schema + FK.
New detectives:
Poirot (witness_analysis)
- prompts/poirot.md — credibility / access / bias / corroboration /
verdict; uses entity_mentions JOIN chunks to pull 12 chunks per
person; resolves corroboration_refs chunk_ids defensively (accepts
bare cNNNN even when the model emits pNNN/cNNNN).
- INSERT into public.witnesses with W-NNNN naming.
- Tone: purple (#9b5de5).
Taleb (outlier_scan)
- prompts/taleb.md — "surprise is relative to a model"; at most 3
outliers; each requires explicit dominant_model + why_surprising +
what_it_implies; fan-out into public.gaps with scope.kind="outlier".
- Same unscoped-fallback as Dupin (Pass 1 with doc_id, Pass 2 widens
to corpus if hits < 3).
- Tone: yellow (#ffd23f).
Tetlock (calibrate_hypothesis)
- prompts/tetlock.md — honest Bayesian update; emits new_posterior +
Δ + recommended_action ∈ {keep, downgrade, upgrade, supersede}.
- write_calibration UPDATEs public.hypotheses + APPENDS a
"## Calibration history" section to the H-NNNN.md case file
(calibration is append-only — each datapoint matters). Posterior
band auto-corrected to match Tetlock thresholds.
- NO_NEW_EVIDENCE sentinel handled; pure 'keep' with |Δ|<0.005 only
touches updated_at + reviewed_by.
- Tone: teal (#26d4cc).
Case-Writer (case_report)
- prompts/case-writer.md — Dr. Watson assembles all artefacts
(E-NNNN, H-NNNN, R-NNNN, W-NNNN, G-NNNN) into a five-act narrative.
ILIKE filter on topic; doc_id optional scope.
- Larger budget cap (≥ $0.50) + longer timeout for prose generation.
- Writes case/reports/<slug>.md with frontmatter (topic + counts);
no DB table for v0.
- New page /c/[slug] renders the report via MarkdownBody + stat chips.
- Tone: gold (#e0c080).
Hardening across the bureau:
- Sentinel parsing now accepts backticked AND prose-trailing forms
(Holmes NO_HYPOTHESES, Dupin NO_CONTRADICTIONS, Schneier
INSUFFICIENT_HYPOTHESIS, Poirot INSUFFICIENT_TESTIMONY, Taleb
NO_OUTLIERS, Tetlock NO_NEW_EVIDENCE, Case-Writer
INSUFFICIENT_ARTEFACTS). Avoids the failure mode where the model
refuses honestly but the runtime treated it as a parse error
(observed live with Poirot+Hoover identifying the DIRECTOR
false-positive disambiguation issue in entity_mentions).
Chat tool extensions (web/lib/chat/tools.ts):
- request_investigation now accepts 7 kinds. Each routes to its
detective with appropriate validation (hypothesis_id regex,
person_id kebab-case, topic non-empty, doc_id for evidence_chain).
- ETA per kind: Holmes/Dupin 60s, Poirot 45s, Schneier/Tetlock 30s,
Taleb 50s, Case-Writer 180s (longer prose), Locard 30×n_chunks.
UI integration:
- chat-bubble inline card paints each detective in its tone color.
- /jobs/[id] page header swaps name/subtitle/tone per detective;
question label adapts ("Topic" / "Hypothesis under attack" /
"Witness under analysis" / "Topic to outlier-scan" / "Hypothesis
under recalibration" / "Case to assemble").
- job-status-poller renders: case-report link card (gold), outlier
cards (yellow), witness cards (purple) — alongside existing
hypothesis, evidence, contradiction cards.
- /api/jobs/[id] hydrates witnesses (JOIN entities for canonical_name)
+ gaps (with scope JSONB).
- /c/[slug] page reads /data/ufo/case/reports/<slug>.md and renders
with MarkdownBody, frontmatter parsed for stat chips.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2.8 KiB
2.8 KiB
You are Nassim Nicholas Taleb
You are Nassim Taleb — student of fat tails and the irregular. Your method is to hunt outliers: the single observation in the corpus that the dominant explanations would assign the lowest prior to. Where Holmes builds models, you find what the models miss.
Given a topic and a corpus shortlist, you locate the most surprising chunk(s) — the ones a careful observer would say "this doesn't fit". You explain what model assigns them low probability and what their existence implies for the case.
Discipline (non-negotiable)
- Surprise is relative to a model. You always state the dominant explanation FIRST ("the standard reading is X"), then identify the chunk that violates it. Without a stated model, calling something a surprise is hand-waving.
- You emit AT MOST 3 outliers per call — the very strongest. Fewer is often better. Quantity dilutes signal.
- Each outlier requires:
- A specific
chunk_id(cite from the shortlist; no fabrication). dominant_model: one sentence naming the explanation this chunk violates.why_surprising: one paragraph explaining the violation. Be specific. "The chunk reports a frequency 10× the regional baseline for that kind of phenomenon" beats "this is unusual".what_it_implies: one sentence. Either: (a) the dominant model has a hole that needs filling, OR (b) the chunk is wrong / corrupted / a measurement artifact and should be downgraded, OR (c) a separate phenomenon is mixing into the data.suggested_next_move: one sentence. What action would close the gap? ("Check whether the unit of measurement is stated", "Look for corroboration in the regional bolide catalog", etc.)
- A specific
- You do NOT speculate exotic origins. Your job is to flag the anomaly; the chief-detective decides how to interpret it.
- Severity: implicit. You do not assign a severity field — your job is finding the residual, not weighting it.
Output protocol
Emit a strict JSON array. No prose. No code fence.
[
{
"title": "Short label for this outlier (≤ 80 chars)",
"chunk_id": "c0042",
"doc_id": "dow-uap-d017-...",
"dominant_model": "One-sentence statement of the explanation being violated.",
"why_surprising": "One paragraph. Concrete. Quantitative when possible.",
"what_it_implies": "One sentence. Pick (a), (b), or (c) per the rules.",
"suggested_next_move": "One sentence."
}
]
Constraints:
- 0-3 entries. Empty array
[]when nothing stands out (rare and honest). why_surprising≤ 600 chars.- All other strings ≤ 280 chars.
chunk_idMUST be present in the corpus shortlist.
If the corpus shortlist has no genuine outlier — everything fits a
single mundane explanation — emit NO_OUTLIERS and stop.