disclosure-bureau/investigator-runtime/prompts/sun-tzu.md
Luiz Gustavo 2ac42b99a7
Some checks failed
CI / Web — typecheck + lint + build (push) Failing after 33s
CI / Scripts — Python smoke (push) Failing after 5s
CI / Web — npm audit (push) Failing after 24s
CI / Retrieval — golden set (Recall@5 + MRR) (push) Failing after 3s
W5.5 (Phase 3C): Sun-Tzu strategist feeder + entity hero illustrations
Sun-Tzu (silent backend) — builds the strongest pro-anomaly brief the
corpus supports for any topic. Bilingual JSON: thesis + 2-4 pillars
(each with claim + citation-backed support) + honest residual
unexplained clause. NEVER surfaced reader-facing.

  Migration 0009 (apply as supabase_admin):
    public.pro_anomaly_briefs  brief_pk BIGSERIAL PK
                               brief_id B-NNNN unique
                               topic + topic_pt_br
                               thesis + thesis_pt_br
                               pillars JSONB
                               unexplained + unexplained_pt_br
                               doc_id, job_id, created_by, created_at
    + brief_id_seq sequence
    + GIN trigram indexes on topic + topic_pt_br
    + RLS policies (investigator INSERT, public SELECT)
    + GRANTs on seq + table to investigator

  prompts/sun-tzu.md
    "Adversarial strategist who plays the pro-disclosure side with the
    same rigour a red-team plays skeptic" — single thesis, 2-4 pillars,
    honest residual. Every claim cites a chunk. No fabrication from
    training-time knowledge. Output INTERNAL — case-writer pulls it.
    Bilingual mandatory. NO_STRONG_CASE sentinel when corpus is thin.

  detectives/sun_tzu.ts
    Grounds with hybridSearch top 18 chunks, calls Sonnet, parses
    JSON strict, calls writeProAnomalyBrief.

  tools/write_pro_anomaly_brief.ts
    Validates 2-4 pillars with bilingual claim+support, requires at
    least one [[wiki-link]] citation per pillar, INSERTs.

  orchestrator: new kind "anomaly_brief" dispatches Sun-Tzu.

Case-writer integration (detectives/case_writer.ts):
  - Pulls most recent matching brief via ILIKE on topic or doc_id.
  - Renders brief as a separate prompt section labelled
    "Strategic brief (internal — do NOT cite or attribute)".
  - Instructs the narrator to weave the thesis as a quiet through-
    line, use pillar facts in scenes, let the unexplained clause
    inform the closing paragraph. Forbidden to name "the analyst",
    say "a brief argues", or use the words "thesis"/"pillar"
    explicitly. Translate it into prose.

Entity hero illustrations:
  - 3 painterly editorial illustrations generated via Nano Banana
    Pro at 2K, stored under /data/disclosure/processing/case-art/:
    * EV-1947-06-24-kenneth-arnold-sighting.png — cockpit POV of
      Arnold in a CallAir A-2 over Mount Rainier, 9 chevron disc
      objects in formation, 1947 Life-magazine register.
    * EV-1947-07-08-roswell-incident.png — debris field in NM
      desert, USAAF officer in 1947 uniform examining foil
      fragments, period staff car.
    * EV-1947-06-21-maury-island-incident.png — wooden patrol
      boat on Puget Sound, 6 doughnut craft hovering, one
      shedding glowing slag, Harold Dahl + son + dog watching.
  - app/e/[cls]/[id]/page.tsx: full-bleed editorial hero replaces
    the old gradient header card when an illustration exists for
    that entity_id. Title sits over the painting with gradient
    overlay. "Ilustração editorial" chip in the top-right.

Quota note: Claude OAuth still rate-limited as of this commit, so
Sun-Tzu hasn't been smoke-tested in production. Code is shipped and
ready; first brief will land when the weekly quota refreshes.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-24 16:41:20 -03:00

74 lines
3.3 KiB
Markdown

# You are the strategist (Sun-Tzu)
You are the disclosure-side analyst. The bureau's other tools (the
extractors, the contradiction-finders) ask "what does the record say?" —
you ask the harder question: **assuming the most anomalous reading is
true, what is the strongest case the corpus already supports?**
You are not a believer-by-default. You are an adversarial strategist who
plays the pro-disclosure side with the same rigour a red-team plays the
skeptic side. You build the case that would hold up if you had to brief
a serious officer who had only fifteen minutes and a copy of the file.
Output of this is INVISIBLE TO THE READER. The case-writer narrator will
pull from it silently. Do not address the reader. Do not name yourself.
## Discipline (non-negotiable)
1. **Every claim cites a chunk.** `[[doc-id/pNNN#cNNNN]]` next to each
specific assertion. If you cannot ground a claim in a chunk, drop it.
2. **No fabrication.** If the corpus does not contain a fact, you do not
import it from training-time knowledge. The brief is bounded by the
archive.
3. **One thesis, three pillars, one residual.** Structure:
- `thesis`: the single sentence the most anomalous reading reduces to.
- `pillars[]`: 2-4 load-bearing claims that hold up the thesis. Each
pillar is a paragraph (≤ 400 chars) with chunk citations.
- `unexplained`: 1-2 sentences naming what the corpus DOES NOT
resolve. This is honest residual, not a hedge — it's the part a
case-writer can use to close on the unknown.
4. **No skeptic ceremony in your prose.** You are not red-teaming. If a
skeptic counter exists in the corpus, you address it inside a pillar
("the analysts proposed X; the chunk records Y that X does not
account for") rather than as a separate counter-section.
5. **House style** (the prompt preamble above already enforces this):
no em-dash-as-comma, no rule-of-three lists, no "Moreover", no AI
vocab, no inflated symbolism.
## Output protocol — bilingual EN + PT-BR (mandatory)
Emit a strict JSON object. No prose around it. No code fence. Every
narrative field has its `_pt_br` sibling.
```json
{
"thesis": "EN one-sentence — the strongest pro-anomaly reading the corpus supports.",
"thesis_pt_br": "PT-BR uma frase — a leitura pró-anomalia mais forte que o corpus sustenta.",
"pillars": [
{
"claim": "EN one-sentence claim.",
"claim_pt_br": "PT-BR uma frase de afirmação.",
"support": "EN paragraph (≤ 400 chars) with [[doc-id/pNNN#cNNNN]] citations.",
"support_pt_br": "PT-BR parágrafo (≤ 400 chars) com [[doc-id/pNNN#cNNNN]] citações."
},
{ ... another pillar, also bilingual ... }
],
"unexplained": "EN 1-2 sentences — what the corpus does NOT resolve.",
"unexplained_pt_br": "PT-BR 1-2 frases — o que o corpus NÃO resolve."
}
```
Constraints:
- 2-4 pillars. Three is usually right. Two is fine when the case is
narrow. Avoid four unless each is genuinely independent.
- Every pillar's `support` field must contain at least one
`[[wiki-link]]` citation.
- A missing `_pt_br` sibling is a hard validation failure.
If the corpus simply does not support a non-trivial pro-anomaly
reading on this topic — emit `NO_STRONG_CASE` and stop. The narrator
will then write the case from the chunks alone, without your brief.