golangLAKEHOUSE/scripts/multi_coord_stress/main.go

// Multi-coordinator stress harness — Phase 1 of the 48-hour mock.
//
// Three coordinators (Alice, Bob, Carol) each own a contract with a
// different demand profile. They run queries against the matrix
// indexer with separate playbook namespaces. The harness fires
// scenario phases (baseline → surge → merge → handover → split) and
// captures every response so we can verify:
//
//   1. Diversity — different (coord, contract, role) triples should
//      surface DIFFERENT top-K worker IDs. If everything returns the
//      same handful of workers, the system is "cycling" not "locking
//      into scenarios."
//   2. Non-determinism — same query reissued should return near-
//      identical top-K (controlled variance from HNSW + judge, if any).
//   3. Learning — after Alice records playbook entries for her
//      contract's queries, Bob takes over the same contract using
//      Alice's playbook namespace; Alice's recordings should surface
//      in Bob's results.
//
// Phase 1 deliberately skips: time-based event clock (events fire
// sequentially), email/SMS ingest (no integration yet), Langfuse
// tracing (would need Go-side wiring). Those are Phase 2/3.
package main

import (
	"bytes"
	"context"
	"encoding/json"
	"flag"
	"fmt"
	"io"
	"log"
	"net/http"
	"os"
	"path/filepath"
	"sort"
	"strings"
	"time"
)

// ── data shapes ──────────────────────────────────────────────────

type Demand struct {
	Role     string   `json:"role"`
	Count    int      `json:"count"`
	Skills   []string `json:"skills"`
	Certs    []string `json:"certs"`
	InRoster *bool    `json:"in_roster,omitempty"` // nil = assume true
}

type Contract struct {
	Name     string   `json:"name"`
	Client   string   `json:"client"`
	Location string   `json:"location"`
	Shift    string   `json:"shift"`
	Demand   []Demand `json:"demand"`
}

type Coordinator struct {
	Name           string
	PlaybookCorpus string
}

// ── matrix.search wire shapes ────────────────────────────────────

type matrixSearchReq struct {
	QueryText      string   `json:"query_text"`
	Corpora        []string `json:"corpora"`
	K              int      `json:"k"`
	UsePlaybook    bool     `json:"use_playbook,omitempty"`
	PlaybookCorpus string   `json:"playbook_corpus,omitempty"`
	ExcludeIDs     []string `json:"exclude_ids,omitempty"`
}

type matrixResult struct {
	ID       string          `json:"id"`
	Distance float32         `json:"distance"`
	Corpus   string          `json:"corpus"`
	Metadata json.RawMessage `json:"metadata,omitempty"`
}

type matrixResp struct {
	Results          []matrixResult `json:"results"`
	PerCorpusCounts  map[string]int `json:"per_corpus_counts"`
	PlaybookBoosted  int            `json:"playbook_boosted,omitempty"`
	PlaybookInjected int            `json:"playbook_injected,omitempty"`
}

// ── event capture ────────────────────────────────────────────────

type ResultRef struct {
	Rank     int     `json:"rank"`
	ID       string  `json:"id"`
	Corpus   string  `json:"corpus"`
	Distance float32 `json:"distance"`
}

type Event struct {
	Phase             string         `json:"phase"`
	Hour              int            `json:"hour"` // operational-narrative time label, not real wall clock
	Coordinator       string         `json:"coordinator"`
	Contract          string         `json:"contract"`
	Role              string         `json:"role"`
	Query             string         `json:"query"`
	SurgeMultiplier   int            `json:"surge_multiplier,omitempty"`
	UsePlaybook       bool           `json:"use_playbook"`
	PlaybookCorpus    string         `json:"playbook_corpus,omitempty"`
	ExcludeIDs        []string       `json:"exclude_ids,omitempty"`
	TopK              []ResultRef    `json:"top_k"`
	PerCorpusCounts   map[string]int `json:"per_corpus_counts,omitempty"`
	PlaybookBoosted   int            `json:"playbook_boosted,omitempty"`
	PlaybookInjected  int            `json:"playbook_injected,omitempty"`
	// JudgeRating: 1-5 quality score on top-1 result against the
	// original inbox body (not the LLM-parsed query). Lets us flag
	// the case where LLM parsing produces a tight-distance match
	// but the result doesn't actually fit the original ask.
	// 0 = unrated, 1-5 = judge verdict.
	JudgeRating       int            `json:"judge_rating,omitempty"`
	Note              string         `json:"note,omitempty"`
	TimestampUnixNano int64          `json:"ts_ns"`
}

type Output struct {
	Coordinators []string  `json:"coordinators"`
	Contracts    []string  `json:"contracts"`
	Events       []Event   `json:"events"`
	Diversity    Diversity `json:"diversity"`
	Determinism  Determ    `json:"determinism"`
	Learning     Learning  `json:"learning"`
	GeneratedAt  time.Time `json:"generated_at"`
}

// Diversity = how distinct are top-K worker sets across (coord,
// contract, role) triples that SHOULD differ. We compute mean Jaccard
// similarity for matched-role-across-contracts pairs (lower is more
// diverse) and matched-coord-different-roles pairs.
type Diversity struct {
	SameRoleAcrossContractsMeanJaccard       float64 `json:"same_role_across_contracts_mean_jaccard"`
	DifferentRolesSameContractMeanJaccard    float64 `json:"different_roles_same_contract_mean_jaccard"`
	NumPairsSameRoleAcrossContracts          int     `json:"num_pairs_same_role_across_contracts"`
	NumPairsDifferentRolesSameContract       int     `json:"num_pairs_different_roles_same_contract"`
}

// Determ = same query reissued — top-K should be near-identical.
// Jaccard close to 1.0 = stable / deterministic, < 0.95 = some HNSW
// or judge variance.
type Determ struct {
	MeanJaccard      float64 `json:"mean_jaccard"`
	NumReissuedPairs int     `json:"num_reissued_pairs"`
}

// Learning = handover signal. After Alice records playbooks for her
// contract, Bob runs the same queries with Alice's playbook namespace.
// We measure: do Alice's recorded answer IDs surface in Bob's top-K?
//
// Two modes:
//   - Verbatim handover: Bob runs Alice's exact queries (trivial case).
//   - Paraphrase handover: Bob runs paraphrased queries against Alice's
//     playbook (the hard case — does cosine on paraphrase find the
//     recorded query's vector?). This is the multi-coord analog of the
//     paraphrase reality test in playbook_lift.
type Learning struct {
	HandoverQueriesRun       int     `json:"handover_queries_run"`
	RecordedAnswersTop1Count int     `json:"recorded_answers_top1_count"`
	RecordedAnswersTopKCount int     `json:"recorded_answers_topk_count"`
	HandoverHitRate          float64 `json:"handover_hit_rate"`

	// Paraphrase handover — only populated when --with-paraphrase-handover.
	ParaphraseHandoverRun       int     `json:"paraphrase_handover_run,omitempty"`
	ParaphraseTop1Count         int     `json:"paraphrase_top1_count,omitempty"`
	ParaphraseTopKCount         int     `json:"paraphrase_topk_count,omitempty"`
	ParaphraseHandoverHitRate   float64 `json:"paraphrase_handover_hit_rate,omitempty"`
}

// ── main ─────────────────────────────────────────────────────────

func main() {
	var (
		gateway       = flag.String("gateway", "http://127.0.0.1:3110", "gateway base URL")
		contractsDir  = flag.String("contracts", "tests/reality/contracts", "directory of contract JSON files")
		corporaCSV    = flag.String("corpora", "workers,ethereal_workers", "comma-separated matrix corpora")
		k             = flag.Int("k", 8, "top-k from matrix.search per query")
		out           = flag.String("out", "reports/reality-tests/multi_coord_stress_001.json", "output JSON path")
		ollama        = flag.String("ollama", "http://127.0.0.1:11434", "Ollama base URL (only used if --with-paraphrase-handover)")
		judgeModel    = flag.String("judge", "qwen2.5:latest", "Ollama model for paraphrase generation (only used if --with-paraphrase-handover)")
		withParaphraseHandover = flag.Bool("with-paraphrase-handover", false, "after the verbatim handover phase, run a paraphrase handover phase: Bob runs paraphrased versions of Alice's queries against Alice's playbook")
	)
	flag.Parse()

	contracts, err := loadContracts(*contractsDir)
	if err != nil {
		log.Fatalf("load contracts: %v", err)
	}
	if len(contracts) < 3 {
		log.Fatalf("need ≥3 contracts in %s, got %d", *contractsDir, len(contracts))
	}

	// First three contracts → coord assignments. Names are fixed so
	// playbook corpora are stable across runs (rerun lands on same
	// namespaces, exercising the persistence path indirectly).
	coords := []Coordinator{
		{Name: "alice", PlaybookCorpus: "playbook_alice"},
		{Name: "bob", PlaybookCorpus: "playbook_bob"},
		{Name: "carol", PlaybookCorpus: "playbook_carol"},
	}

	// Initial assignment: alice→alpha, bob→beta, carol→gamma.
	assignments := map[string]*Contract{
		"alice": &contracts[0],
		"bob":   &contracts[1],
		"carol": &contracts[2],
	}

	corpora := strings.Split(*corporaCSV, ",")
	hc := &http.Client{Timeout: 30 * time.Second}
	ctx := context.Background()
	_ = ctx

	output := Output{
		Coordinators: []string{"alice", "bob", "carol"},
		Contracts:    []string{contracts[0].Name, contracts[1].Name, contracts[2].Name},
		GeneratedAt:  time.Now().UTC(),
	}

	log.Printf("[stress] 3 coords, 3 contracts, k=%d, corpora=%v", *k, corpora)

	// ── Phase 1: baseline ───────────────────────────────────────
	// Each coord runs their own contract's role queries. Records
	// playbook entries (top-1 of each as a synthetic "successful
	// match" outcome) into their personal namespace.
	log.Printf("[stress] phase 1: baseline")
	for _, coord := range coords {
		c := assignments[coord.Name]
		for _, d := range c.Demand {
			q := buildQuery(c, d, 1)
			resp := must(matrixSearch(hc, *gateway, q, corpora, *k, true, coord.PlaybookCorpus))
			ev := captureEvent("baseline", 0, coord.Name, c.Name, d.Role, q, 1, true, coord.PlaybookCorpus, resp)
			output.Events = append(output.Events, ev)
			// Record top-1 as a successful playbook entry for this coord.
			if len(resp.Results) > 0 {
				if err := playbookRecord(hc, *gateway, q, resp.Results[0].ID, resp.Results[0].Corpus, 1.0, coord.PlaybookCorpus); err != nil {
					log.Printf("  record (%s/%s): %v", coord.Name, d.Role, err)
				}
			}
		}
	}

	// ── Phase 1b: new-resume injection (Hour 6) ─────────────────
	// Mid-day, three new resumes arrive — workers with no prior
	// history. We embed + add them to the workers vectord index,
	// then verify they're findable by their unique skill marker.
	// Tests the substrate's ability to absorb fresh candidates
	// without restart.
	log.Printf("[stress] phase 1b: new-resume injection (3 fresh workers, verify findable)")
	// Each fresh worker has a SEMANTIC query that should surface them
	// based on the actual content of their resume — role + skills +
	// location. nomic-embed-text is dense/semantic, NOT lexical, so a
	// "find me FRESHTAG_..." style unique-substring query does NOT
	// surface the fresh worker; the embedder weights rare substrings
	// as low-information noise. The semantic query below represents
	// what a real coordinator would actually issue.
	freshWorkers := []struct {
		ID     string
		Resume string
		Verify string // semantic query expected to surface this worker
	}{
		{
			ID:     "fresh-001",
			Resume: "Senior rigger with 12 years tower-crane signaling experience. NCCCO crane signal/rigger certification active. Chicago IL metro, available immediately. Construction-site rigging specialist.",
			Verify: "Senior tower crane rigger NCCCO certification Chicago construction signaling",
		},
		{
			ID:     "fresh-002",
			Resume: "Bilingual safety coordinator (Spanish + English). OSHA trainer credentials, 8 years manufacturing safety training delivery. Indianapolis IN. Manages multilingual crew safety briefings and incident documentation.",
			Verify: "Bilingual Spanish English OSHA trainer safety coordinator Indianapolis manufacturing",
		},
		{
			ID:     "fresh-003",
			Resume: "FAA Part 107 certified drone pilot. UAV site surveying with GIS mapping output for construction site progress reports. Chicago IL metro. 5 years aerial surveying for general contractors.",
			Verify: "FAA Part 107 drone surveyor UAV pilot GIS construction site mapping Chicago",
		},
	}
	for _, fw := range freshWorkers {
		if err := ingestFreshWorker(hc, *gateway, fw.ID, fw.Resume, map[string]any{
			"name":   fw.ID,
			"role":   "fresh-resume",
			"source": "phase-1b-injection",
		}); err != nil {
			log.Fatalf("ingest fresh worker %s: %v", fw.ID, err)
		}
	}
	for _, fw := range freshWorkers {
		resp := must(matrixSearch(hc, *gateway, fw.Verify, corpora, *k, false, ""))
		ev := captureEvent("new-resume-verify", 6, "system", "fresh-resume-pool", "fresh", fw.Verify, 1, false, "", resp)
		// Find the fresh worker's rank in top-K (rank 0 = top-1).
		freshRank := -1
		for i, r := range resp.Results {
			if r.ID == fw.ID {
				freshRank = i
				break
			}
		}
		switch {
		case freshRank == 0:
			ev.Note = fmt.Sprintf("fresh worker %s at top-1 — semantic absorption working", fw.ID)
		case freshRank > 0:
			ev.Note = fmt.Sprintf("fresh worker %s at rank %d (in top-K but not top-1)", fw.ID, freshRank)
		default:
			ev.Note = fmt.Sprintf("fresh worker %s NOT in top-K (top-1 was %s) — embedder didn't surface fresh-resume content over existing population", fw.ID, resp.Results[0].ID)
		}
		output.Events = append(output.Events, ev)
	}

	// ── Phase 1c: inbox burst (Hour 9) ──────────────────────────
	// Mid-morning, 6 incoming signals arrive — emails + SMS — each
	// carrying a structured demand for the system to act on. Events
	// fire in PRIORITY ORDER (urgent → high → medium). For each, we:
	//   1. POST to /v1/observer/inbox so the witness loop records it
	//   2. Run matrix.search using the embedded demand
	//   3. Capture both as events
	//
	// Priority weighting matters because real coordinators triage
	// urgent client-side asks before medium-priority background
	// signals. The substrate doesn't enforce ordering today (callers
	// fire in their preferred order); this phase verifies the
	// recording surface and the search-from-inbox flow work.
	log.Printf("[stress] phase 1c: inbox burst (6 events, priority-ordered)")
	type inboxEvent struct {
		Priority string // "urgent" | "high" | "medium" | "low"
		Type     string // "email" | "sms"
		Sender   string
		Subject  string
		Body     string
		Coord    string
	}
	inboxEvents := []inboxEvent{
		{
			Priority: "urgent", Type: "email", Sender: "ops@northstar.com",
			Subject: "URGENT: 50 forklift operators Cleveland Monday",
			Body:    "Need 50 forklift operators in Cleveland OH for Monday day shift. OSHA-30 + active forklift cert required. Current Milwaukee batch cannot relocate.",
			Coord:   "alice",
		},
		{
			Priority: "urgent", Type: "email", Sender: "client@crossroads-mfg.com",
			Subject: "URGENT: Production line down — need 30 production workers tonight",
			Body:    "Production line failure at Indianapolis IN site. Need 30 production workers swing shift starting tonight. Assembly + machine operation experience required.",
			Coord:   "bob",
		},
		{
			Priority: "high", Type: "email", Sender: "supervisor@loop-construction.com",
			Subject: "Need crane operator Chicago for 2-week project",
			Body:    "Crane operator with NCCCO certification needed for 2-week Chicago IL site project. Day shift. Mobile crane experience preferred.",
			Coord:   "carol",
		},
		{
			Priority: "high", Type: "sms", Sender: "+1-555-0142",
			Body:  "Bilingual safety coord needed Indy plant ASAP. Spanish + English. OSHA trainer credential.",
			Coord: "bob",
		},
		{
			Priority: "medium", Type: "sms", Sender: "+1-555-0188",
			Body:  "Drone surveyor for Chicago site progress mapping. FAA Part 107.",
			Coord: "carol",
		},
		{
			Priority: "medium", Type: "email", Sender: "scheduling@northstar.com",
			Subject: "FYI: warehouse worker capacity check Milwaukee",
			Body:    "Routine capacity check on Milwaukee warehouse worker pool — anyone with cold storage experience for next week?",
			Coord:   "alice",
		},
	}
	// Sort by priority (urgent < high < medium < low for ordering).
	prioRank := map[string]int{"urgent": 0, "high": 1, "medium": 2, "low": 3}
	sort.SliceStable(inboxEvents, func(i, j int) bool {
		return prioRank[inboxEvents[i].Priority] < prioRank[inboxEvents[j].Priority]
	})
	for _, ie := range inboxEvents {
		// 1. Record inbox event at observerd
		if err := postInbox(hc, *gateway, ie.Type, ie.Sender, ie.Subject, ie.Body, ie.Priority, ie.Coord); err != nil {
			log.Printf("  inbox record failed (%s): %v", ie.Priority, err)
			continue
		}
		// 2. LLM parses the body into a structured demand. Real
		// production: a small local model extracts {role, count,
		// location, certs, skills, shift} from email/SMS bodies
		// instead of a coordinator typing it into a form. Test
		// captures both raw body and parsed structure for review.
		parsed, perr := parseInboxDemand(hc, *ollama, *judgeModel, ie.Body)
		if perr != nil {
			log.Printf("  inbox demand parse failed (%s): %v", ie.Priority, perr)
			continue
		}
		// 3. Build a query string from the parsed demand and search.
		query := parsed.AsQuery()
		coord := coordByName(coords, ie.Coord)
		resp, err := matrixSearch(hc, *gateway, query, corpora, *k, true, coord.PlaybookCorpus)
		if err != nil {
			log.Printf("  inbox-triggered search failed (%s): %v", ie.Priority, err)
			continue
		}
		ev := captureEvent("inbox-triggered-search", 9, ie.Coord, "inbox-burst", ie.Subject, query, 1, true, coord.PlaybookCorpus, resp)
		parsedJSON, _ := json.Marshal(parsed)
		ev.Note = fmt.Sprintf("inbox %s/%s from %s · LLM-parsed demand: %s", ie.Type, ie.Priority, ie.Sender, string(parsedJSON))
		// 4. Judge re-rates top-1 against the ORIGINAL body — not the
		// parsed query. Catches the case where parsing dropped a
		// constraint (or where the corpus has no real match for the
		// asked specialist, e.g. "drone surveyor" against a corpus
		// of warehouse workers — the closest semantic neighbor will
		// have a tight distance but not actually fit).
		if len(resp.Results) > 0 {
			rating := judgeInboxResult(hc, *ollama, *judgeModel, ie.Body, resp.Results[0])
			ev.JudgeRating = rating
		}
		output.Events = append(output.Events, ev)
	}

	// ── Phase 2: surge ──────────────────────────────────────────
	// Each coord's contract demand doubles. URGENT phrasing.
	log.Printf("[stress] phase 2: surge (2x demand, urgent phrasing)")
	for _, coord := range coords {
		c := assignments[coord.Name]
		for _, d := range c.Demand {
			q := buildQuery(c, d, 2)
			resp := must(matrixSearch(hc, *gateway, q, corpora, *k, true, coord.PlaybookCorpus))
			ev := captureEvent("surge", 12, coord.Name, c.Name, d.Role, q, 2, true, coord.PlaybookCorpus, resp)
			output.Events = append(output.Events, ev)
		}
	}

	// ── Phase 2b: 200-worker swap (Hour 18) ──────────────────────
	// Alpha's client says "the 200 workers you placed are unavailable
	// — find replacements." We capture the top-K from the warehouse
	// query, then re-issue the same query with those IDs excluded.
	// Real product test: does the system find genuinely different
	// candidates, or does it sit on the same population?
	log.Printf("[stress] phase 2b: 200-worker swap (alpha warehouse — exclude originally placed)")
	warehouseDemand := contracts[0].Demand[0] // slot 0 is warehouse worker by contract design
	swapQuery := buildQuery(&contracts[0], warehouseDemand, 1)
	origResp := must(matrixSearch(hc, *gateway, swapQuery, corpora, *k, false, ""))
	placedIDs := make([]string, 0, len(origResp.Results))
	for _, r := range origResp.Results {
		placedIDs = append(placedIDs, r.ID)
	}
	origEv := captureEvent("swap-original", 18, "alice", contracts[0].Name, warehouseDemand.Role, swapQuery, 1, false, "", origResp)
	origEv.Note = fmt.Sprintf("captured %d originally-placed worker IDs", len(placedIDs))
	output.Events = append(output.Events, origEv)

	swapResp := must(matrixSearch(hc, *gateway, swapQuery, corpora, *k, false, "", placedIDs...))
	swapEv := captureEvent("swap-replace", 18, "alice", contracts[0].Name, warehouseDemand.Role, swapQuery, 1, false, "", swapResp)
	swapEv.ExcludeIDs = placedIDs
	swapIDs := make([]string, 0, len(swapResp.Results))
	for _, r := range swapResp.Results {
		swapIDs = append(swapIDs, r.ID)
	}
	swapJacc := jaccardStrings(placedIDs, swapIDs)
	swapEv.Note = fmt.Sprintf("Jaccard(orig, swap) = %.3f (lower = better; 0 = fully replaced)", swapJacc)
	output.Events = append(output.Events, swapEv)

	// ── Phase 3: merge — alpha + beta combined under alice ──────
	log.Printf("[stress] phase 3: merge (alpha + beta combined, alice handles)")
	mergedDemand := append(append([]Demand{}, contracts[0].Demand...), contracts[1].Demand...)
	for _, d := range mergedDemand {
		mergedC := &Contract{Name: contracts[0].Name + "+" + contracts[1].Name, Location: contracts[0].Location + " + " + contracts[1].Location, Shift: "shared"}
		q := buildQuery(mergedC, d, 1)
		resp := must(matrixSearch(hc, *gateway, q, corpora, *k, true, coords[0].PlaybookCorpus))
		ev := captureEvent("merge", 24, "alice", mergedC.Name, d.Role, q, 1, true, coords[0].PlaybookCorpus, resp)
		output.Events = append(output.Events, ev)
	}

	// ── Phase 4: handover — bob takes alpha contract, USING ─────
	// alice's playbook namespace. Tests whether Alice's recordings
	// surface in Bob's results when Bob runs Alice's contract.
	log.Printf("[stress] phase 4: handover (bob takes alpha, using alice's playbook)")
	aliceRecordedAnswers := map[string]string{} // role → recorded answer id
	for _, ev := range output.Events {
		if ev.Phase == "baseline" && ev.Coordinator == "alice" && len(ev.TopK) > 0 {
			aliceRecordedAnswers[ev.Role] = ev.TopK[0].ID
		}
	}
	handoverHitsTop1 := 0
	handoverHitsTopK := 0
	handoverRun := 0
	for _, d := range contracts[0].Demand {
		q := buildQuery(&contracts[0], d, 1)
		resp := must(matrixSearch(hc, *gateway, q, corpora, *k, true, coords[0].PlaybookCorpus))
		ev := captureEvent("handover", 30, "bob", contracts[0].Name, d.Role, q, 1, true, coords[0].PlaybookCorpus, resp)
		output.Events = append(output.Events, ev)
		handoverRun++
		recordedID, ok := aliceRecordedAnswers[d.Role]
		if !ok {
			continue
		}
		if len(ev.TopK) > 0 && ev.TopK[0].ID == recordedID {
			handoverHitsTop1++
			handoverHitsTopK++
		} else {
			for _, r := range ev.TopK {
				if r.ID == recordedID {
					handoverHitsTopK++
					break
				}
			}
		}
	}
	output.Learning.HandoverQueriesRun = handoverRun
	output.Learning.RecordedAnswersTop1Count = handoverHitsTop1
	output.Learning.RecordedAnswersTopKCount = handoverHitsTopK
	if handoverRun > 0 {
		output.Learning.HandoverHitRate = float64(handoverHitsTop1) / float64(handoverRun)
	}

	// ── Phase 4b: paraphrase handover ───────────────────────────
	// Bob runs PARAPHRASED versions of Alice's queries against
	// Alice's playbook. The verbatim handover above is the trivial
	// case (identical queries → identical retrieval → playbook
	// boost). The paraphrase handover is the real test: did Alice's
	// institutional memory survive the wording change Bob would
	// naturally introduce?
	if *withParaphraseHandover {
		log.Printf("[stress] phase 4b: paraphrase handover (bob runs paraphrased versions of alice's queries)")
		pHandoverRun := 0
		pTop1 := 0
		pTopK := 0
		for _, d := range contracts[0].Demand {
			origQuery := buildQuery(&contracts[0], d, 1)
			paraphrase, err := generateParaphrase(hc, *ollama, *judgeModel, origQuery)
			if err != nil {
				log.Printf("  paraphrase gen failed for %s: %v", d.Role, err)
				continue
			}
			resp, err := matrixSearch(hc, *gateway, paraphrase, corpora, *k, true, coords[0].PlaybookCorpus)
			if err != nil {
				log.Printf("  paraphrase search failed for %s: %v", d.Role, err)
				continue
			}
			ev := captureEvent("handover-paraphrase", 36, "bob", contracts[0].Name, d.Role, paraphrase, 1, true, coords[0].PlaybookCorpus, resp)
			ev.Note = "paraphrase of: " + origQuery
			output.Events = append(output.Events, ev)
			pHandoverRun++
			recordedID, ok := aliceRecordedAnswers[d.Role]
			if !ok {
				continue
			}
			if len(ev.TopK) > 0 && ev.TopK[0].ID == recordedID {
				pTop1++
				pTopK++
			} else {
				for _, r := range ev.TopK {
					if r.ID == recordedID {
						pTopK++
						break
					}
				}
			}
		}
		output.Learning.ParaphraseHandoverRun = pHandoverRun
		output.Learning.ParaphraseTop1Count = pTop1
		output.Learning.ParaphraseTopKCount = pTopK
		if pHandoverRun > 0 {
			output.Learning.ParaphraseHandoverHitRate = float64(pTop1) / float64(pHandoverRun)
		}
	}

	// ── Phase 5: split — surge re-distributed across 3 coords ──
	log.Printf("[stress] phase 5: split (alpha surge spread across all 3 coords)")
	for i, d := range contracts[0].Demand {
		coord := coords[i%len(coords)]
		c := &contracts[0]
		q := buildQuery(c, d, 2)
		resp := must(matrixSearch(hc, *gateway, q, corpora, *k, true, coord.PlaybookCorpus))
		ev := captureEvent("split", 42, coord.Name, c.Name+"-share-"+coord.Name, d.Role, q, 2, true, coord.PlaybookCorpus, resp)
		output.Events = append(output.Events, ev)
	}

	// ── Phase 6: non-determinism check ─────────────────────────
	// Reissue each baseline query once and compare top-K Jaccard.
	log.Printf("[stress] phase 6: non-determinism (reissue baselines, measure Jaccard)")
	jaccards := []float64{}
	for _, ev := range output.Events {
		if ev.Phase != "baseline" {
			continue
		}
		resp := must(matrixSearch(hc, *gateway, ev.Query, corpora, *k, false, "")) // playbook OFF for reissue to isolate retrieval stability
		reissue := captureEvent("reissue", 48, ev.Coordinator, ev.Contract, ev.Role, ev.Query, 1, false, "", resp)
		output.Events = append(output.Events, reissue)
		// Compare against ev.TopK (also playbook-on baseline). Note:
		// this conflates retrieval stability with playbook stability.
		// We capture both ev (playbook on) and a fresh retrieval (off);
		// real determinism = retrieval-only top-K comparison.
		freshRetrievalResp := must(matrixSearch(hc, *gateway, ev.Query, corpora, *k, false, ""))
		freshRetrievalEv := captureEvent("reissue-retrieval-only", 48, ev.Coordinator, ev.Contract, ev.Role, ev.Query, 1, false, "", freshRetrievalResp)
		j := jaccardTopK(reissue.TopK, freshRetrievalEv.TopK)
		jaccards = append(jaccards, j)
	}
	output.Determinism.NumReissuedPairs = len(jaccards)
	output.Determinism.MeanJaccard = mean(jaccards)

	// ── Phase 7: diversity analysis ─────────────────────────────
	log.Printf("[stress] phase 7: diversity analysis")
	output.Diversity = computeDiversity(output.Events)

	// ── write ───────────────────────────────────────────────────
	if err := os.MkdirAll(filepath.Dir(*out), 0o755); err != nil {
		log.Fatalf("mkdir: %v", err)
	}
	bs, _ := json.MarshalIndent(output, "", "  ")
	if err := os.WriteFile(*out, bs, 0o644); err != nil {
		log.Fatalf("write %s: %v", *out, err)
	}

	log.Printf("[stress] DONE — events=%d", len(output.Events))
	log.Printf("[stress]   diversity: same-role-across-contracts mean Jaccard = %.3f (n=%d)",
		output.Diversity.SameRoleAcrossContractsMeanJaccard, output.Diversity.NumPairsSameRoleAcrossContracts)
	log.Printf("[stress]              different-roles-same-contract mean Jaccard = %.3f (n=%d)",
		output.Diversity.DifferentRolesSameContractMeanJaccard, output.Diversity.NumPairsDifferentRolesSameContract)
	log.Printf("[stress]   determinism: mean Jaccard on reissue = %.3f (n=%d)",
		output.Determinism.MeanJaccard, output.Determinism.NumReissuedPairs)
	log.Printf("[stress]   learning verbatim: handover hit rate (top-1) = %d/%d = %.0f%%",
		output.Learning.RecordedAnswersTop1Count, output.Learning.HandoverQueriesRun,
		output.Learning.HandoverHitRate*100)
	if output.Learning.ParaphraseHandoverRun > 0 {
		log.Printf("[stress]   learning paraphrase: handover hit rate (top-1) = %d/%d = %.0f%% (top-K = %d/%d)",
			output.Learning.ParaphraseTop1Count, output.Learning.ParaphraseHandoverRun,
			output.Learning.ParaphraseHandoverHitRate*100,
			output.Learning.ParaphraseTopKCount, output.Learning.ParaphraseHandoverRun)
	}
	log.Printf("[stress]   results → %s", *out)
}

// generateParaphrase asks the judge model to rephrase a staffing query
// while preserving intent — same prompt template as
// scripts/playbook_lift/main.go, kept here as a copy to avoid a shared
// internal package for two scripts. If callers ever need a third
// paraphraser, lift this into internal/paraphrase/.
func generateParaphrase(hc *http.Client, ollamaURL, model, query string) (string, error) {
	system := `You rephrase staffing queries while preserving intent.
Output JSON only: {"paraphrase": "<rephrased query>"}.
Rules:
- Keep the same role, certifications, geography, and constraints.
- Vary the wording (synonyms, reordered clauses, different sentence shape).
- Do NOT add or remove requirements.
- Do NOT explain — just emit the JSON.`
	body, _ := json.Marshal(map[string]any{
		"model":  model,
		"stream": false,
		"format": "json",
		"messages": []map[string]string{
			{"role": "system", "content": system},
			{"role": "user", "content": query},
		},
		"options": map[string]any{"temperature": 0.5},
	})
	req, _ := http.NewRequest("POST", ollamaURL+"/api/chat", bytes.NewReader(body))
	req.Header.Set("Content-Type", "application/json")
	resp, err := hc.Do(req)
	if err != nil {
		return "", err
	}
	defer resp.Body.Close()
	if resp.StatusCode/100 != 2 {
		return "", fmt.Errorf("ollama chat: HTTP %d", resp.StatusCode)
	}
	rb, _ := io.ReadAll(resp.Body)
	var ollamaResp struct {
		Message struct {
			Content string `json:"content"`
		} `json:"message"`
	}
	if err := json.Unmarshal(rb, &ollamaResp); err != nil {
		return "", err
	}
	var out struct {
		Paraphrase string `json:"paraphrase"`
	}
	if err := json.Unmarshal([]byte(ollamaResp.Message.Content), &out); err != nil {
		return "", fmt.Errorf("decode paraphrase: %w (content=%q)", err, ollamaResp.Message.Content)
	}
	if strings.TrimSpace(out.Paraphrase) == "" {
		return "", fmt.Errorf("empty paraphrase (content=%q)", ollamaResp.Message.Content)
	}
	return out.Paraphrase, nil
}

// ── helpers ──────────────────────────────────────────────────────

func loadContracts(dir string) ([]Contract, error) {
	files, err := filepath.Glob(filepath.Join(dir, "contract_*.json"))
	if err != nil {
		return nil, err
	}
	if len(files) == 0 {
		return nil, fmt.Errorf("no contract_*.json files in %s", dir)
	}
	var out []Contract
	for _, f := range files {
		bs, err := os.ReadFile(f)
		if err != nil {
			return nil, err
		}
		var c Contract
		if err := json.Unmarshal(bs, &c); err != nil {
			return nil, fmt.Errorf("%s: %w", f, err)
		}
		out = append(out, c)
	}
	return out, nil
}

func buildQuery(c *Contract, d Demand, surge int) string {
	var b strings.Builder
	if surge > 1 {
		b.WriteString(fmt.Sprintf("URGENT: need %d ", d.Count*surge))
	} else {
		b.WriteString(fmt.Sprintf("Need %d ", d.Count))
	}
	b.WriteString(d.Role)
	if c.Location != "" {
		b.WriteString(" for ")
		b.WriteString(c.Location)
	}
	if c.Shift != "" {
		b.WriteString(", ")
		b.WriteString(c.Shift)
		b.WriteString(" shift")
	}
	if len(d.Certs) > 0 {
		b.WriteString(", certifications: ")
		b.WriteString(strings.Join(d.Certs, ", "))
	}
	if len(d.Skills) > 0 {
		b.WriteString(", skills: ")
		b.WriteString(strings.Join(d.Skills, ", "))
	}
	return b.String()
}

func captureEvent(phase string, hour int, coord, contract, role, query string, surge int, usePlaybook bool, pbCorpus string, resp *matrixResp) Event {
	topK := make([]ResultRef, 0, len(resp.Results))
	for i, r := range resp.Results {
		topK = append(topK, ResultRef{Rank: i, ID: r.ID, Corpus: r.Corpus, Distance: r.Distance})
	}
	return Event{
		Phase:             phase,
		Hour:              hour,
		Coordinator:       coord,
		Contract:          contract,
		Role:              role,
		Query:             query,
		SurgeMultiplier:   surge,
		UsePlaybook:       usePlaybook,
		PlaybookCorpus:    pbCorpus,
		TopK:              topK,
		PerCorpusCounts:   resp.PerCorpusCounts,
		PlaybookBoosted:   resp.PlaybookBoosted,
		PlaybookInjected:  resp.PlaybookInjected,
		TimestampUnixNano: time.Now().UnixNano(),
	}
}

func computeDiversity(events []Event) Diversity {
	// Filter to baseline events for clean apples-to-apples.
	type key struct{ contract, role string }
	byKey := map[key][]string{}
	for _, ev := range events {
		if ev.Phase != "baseline" {
			continue
		}
		k := key{ev.Contract, ev.Role}
		ids := make([]string, len(ev.TopK))
		for i, r := range ev.TopK {
			ids[i] = r.ID
		}
		byKey[k] = ids
	}

	// Same role across contracts: same `role`, different `contract`.
	rolesSeen := map[string][][]string{}
	contractsSeen := map[string][]struct {
		role string
		ids  []string
	}{}
	for k, ids := range byKey {
		rolesSeen[k.role] = append(rolesSeen[k.role], ids)
		contractsSeen[k.contract] = append(contractsSeen[k.contract], struct {
			role string
			ids  []string
		}{k.role, ids})
	}

	var (
		sameRoleJacc       []float64
		diffRolesJacc      []float64
	)
	// Same-role-across-contracts: each role's idsSet pair-wise.
	for _, idsList := range rolesSeen {
		for i := 0; i < len(idsList); i++ {
			for j := i + 1; j < len(idsList); j++ {
				sameRoleJacc = append(sameRoleJacc, jaccardStrings(idsList[i], idsList[j]))
			}
		}
	}
	// Different-roles-same-contract.
	for _, items := range contractsSeen {
		for i := 0; i < len(items); i++ {
			for j := i + 1; j < len(items); j++ {
				if items[i].role == items[j].role {
					continue
				}
				diffRolesJacc = append(diffRolesJacc, jaccardStrings(items[i].ids, items[j].ids))
			}
		}
	}

	return Diversity{
		SameRoleAcrossContractsMeanJaccard:    mean(sameRoleJacc),
		DifferentRolesSameContractMeanJaccard: mean(diffRolesJacc),
		NumPairsSameRoleAcrossContracts:       len(sameRoleJacc),
		NumPairsDifferentRolesSameContract:    len(diffRolesJacc),
	}
}

func jaccardTopK(a, b []ResultRef) float64 {
	aIDs := make([]string, len(a))
	bIDs := make([]string, len(b))
	for i, r := range a {
		aIDs[i] = r.ID
	}
	for i, r := range b {
		bIDs[i] = r.ID
	}
	return jaccardStrings(aIDs, bIDs)
}

func jaccardStrings(a, b []string) float64 {
	if len(a) == 0 && len(b) == 0 {
		return 1.0
	}
	setA := map[string]bool{}
	for _, x := range a {
		setA[x] = true
	}
	intersect := 0
	for _, x := range b {
		if setA[x] {
			intersect++
		}
	}
	union := len(setA)
	for _, x := range b {
		if !setA[x] {
			union++
		}
	}
	if union == 0 {
		return 0
	}
	return float64(intersect) / float64(union)
}

func mean(xs []float64) float64 {
	if len(xs) == 0 {
		return 0
	}
	s := 0.0
	for _, x := range xs {
		s += x
	}
	return s / float64(len(xs))
}

// ── HTTP helpers ─────────────────────────────────────────────────

func matrixSearch(hc *http.Client, gw, query string, corpora []string, k int, usePlaybook bool, playbookCorpus string, excludeIDs ...string) (*matrixResp, error) {
	body, _ := json.Marshal(matrixSearchReq{
		QueryText:      query,
		Corpora:        corpora,
		K:              k,
		UsePlaybook:    usePlaybook,
		PlaybookCorpus: playbookCorpus,
		ExcludeIDs:     excludeIDs,
	})
	req, _ := http.NewRequest("POST", gw+"/v1/matrix/search", bytes.NewReader(body))
	req.Header.Set("Content-Type", "application/json")
	resp, err := hc.Do(req)
	if err != nil {
		return nil, err
	}
	defer resp.Body.Close()
	if resp.StatusCode/100 != 2 {
		rb, _ := io.ReadAll(resp.Body)
		return nil, fmt.Errorf("matrix.search %d: %s", resp.StatusCode, string(rb))
	}
	var out matrixResp
	if err := json.NewDecoder(resp.Body).Decode(&out); err != nil {
		return nil, err
	}
	return &out, nil
}

// ingestFreshWorker embeds + adds a single fresh worker to the
// vectord 'workers' index. Two HTTP hops via the gateway: /v1/embed
// to get the vector, /v1/vectors/workers/add to insert. Used by the
// new-resume-injection phase to test mid-run absorption of fresh
// candidates without restart.
func ingestFreshWorker(hc *http.Client, gw, id, text string, metadata map[string]any) error {
	embedBs, _ := json.Marshal(map[string]any{
		"texts": []string{text},
		"model": "nomic-embed-text-v2-moe",
	})
	req, _ := http.NewRequest("POST", gw+"/v1/embed", bytes.NewReader(embedBs))
	req.Header.Set("Content-Type", "application/json")
	resp, err := hc.Do(req)
	if err != nil {
		return fmt.Errorf("embed: %w", err)
	}
	defer resp.Body.Close()
	if resp.StatusCode/100 != 2 {
		rb, _ := io.ReadAll(resp.Body)
		return fmt.Errorf("embed %d: %s", resp.StatusCode, string(rb))
	}
	var er struct {
		Vectors [][]float32 `json:"vectors"`
	}
	if err := json.NewDecoder(resp.Body).Decode(&er); err != nil {
		return fmt.Errorf("decode embed: %w", err)
	}
	if len(er.Vectors) == 0 || len(er.Vectors[0]) == 0 {
		return fmt.Errorf("embed returned no vectors")
	}

	metaBs, _ := json.Marshal(metadata)
	addBs, _ := json.Marshal(map[string]any{
		"items": []map[string]any{
			{"id": id, "vector": er.Vectors[0], "metadata": json.RawMessage(metaBs)},
		},
	})
	req2, _ := http.NewRequest("POST", gw+"/v1/vectors/index/workers/add", bytes.NewReader(addBs))
	req2.Header.Set("Content-Type", "application/json")
	resp2, err := hc.Do(req2)
	if err != nil {
		return fmt.Errorf("vectord add: %w", err)
	}
	defer resp2.Body.Close()
	if resp2.StatusCode/100 != 2 {
		rb, _ := io.ReadAll(resp2.Body)
		return fmt.Errorf("vectord add %d: %s", resp2.StatusCode, string(rb))
	}
	return nil
}

// parsedDemand is the LLM-extracted structure from an inbox message
// body — what a real coordinator would type into a search form.
// Empty fields are honest: the body didn't say.
type parsedDemand struct {
	Role     string   `json:"role"`
	Count    int      `json:"count"`
	Location string   `json:"location"`
	Certs    []string `json:"certs"`
	Skills   []string `json:"skills"`
	Shift    string   `json:"shift"`
}

// AsQuery composes a matrix.search query string from the parsed
// fields. Mirrors buildQuery's shape so search-time semantics match
// what the contract-driven phases produce. Empty fields are skipped
// rather than emitted as "" markers.
func (p parsedDemand) AsQuery() string {
	var b strings.Builder
	if p.Count > 0 {
		fmt.Fprintf(&b, "Need %d ", p.Count)
	} else {
		b.WriteString("Need ")
	}
	b.WriteString(p.Role)
	if p.Location != "" {
		b.WriteString(" for ")
		b.WriteString(p.Location)
	}
	if p.Shift != "" {
		b.WriteString(", ")
		b.WriteString(p.Shift)
		b.WriteString(" shift")
	}
	if len(p.Certs) > 0 {
		b.WriteString(", certifications: ")
		b.WriteString(strings.Join(p.Certs, ", "))
	}
	if len(p.Skills) > 0 {
		b.WriteString(", skills: ")
		b.WriteString(strings.Join(p.Skills, ", "))
	}
	return b.String()
}

// judgeInboxResult rates the top retrieval against the ORIGINAL inbox
// body. Returns 1-5 (5 = perfect match, 1 = irrelevant); 0 on any
// error. Real product driver: a tight-distance result on a
// LLM-parsed query may still be wrong-domain (parser dropped a
// critical constraint, or the corpus genuinely has no match). The
// rating gives coordinators an honest "this is close in vector
// space but doesn't actually fit your ask" signal.
func judgeInboxResult(hc *http.Client, ollamaURL, model, inboxBody string, top matrixResult) int {
	system := `You rate retrieval results for a staffing co-pilot.
Rate the result 1-5 against the original inbox request:
  5 = perfect match (this person/role IS what was asked for)
  4 = strong match (right field, right level, minor mismatches)
  3 = adjacent match (related field or partial overlap)
  2 = weak/tangential match
  1 = irrelevant
Output JSON only: {"rating": N, "reason": "<one sentence>"}.`
	user := fmt.Sprintf("Original inbox request:\n%s\n\nResult corpus: %s\nResult ID: %s\nResult metadata:\n%s",
		inboxBody, top.Corpus, top.ID, string(top.Metadata))
	body, _ := json.Marshal(map[string]any{
		"model":  model,
		"stream": false,
		"format": "json",
		"messages": []map[string]string{
			{"role": "system", "content": system},
			{"role": "user", "content": user},
		},
		"options": map[string]any{"temperature": 0},
	})
	req, _ := http.NewRequest("POST", ollamaURL+"/api/chat", bytes.NewReader(body))
	req.Header.Set("Content-Type", "application/json")
	resp, err := hc.Do(req)
	if err != nil {
		return 0
	}
	defer resp.Body.Close()
	if resp.StatusCode/100 != 2 {
		return 0
	}
	rb, _ := io.ReadAll(resp.Body)
	var ollamaResp struct {
		Message struct {
			Content string `json:"content"`
		} `json:"message"`
	}
	if err := json.Unmarshal(rb, &ollamaResp); err != nil {
		return 0
	}
	var v struct {
		Rating int `json:"rating"`
	}
	if err := json.Unmarshal([]byte(ollamaResp.Message.Content), &v); err != nil {
		return 0
	}
	if v.Rating < 1 || v.Rating > 5 {
		return 0
	}
	return v.Rating
}

// parseInboxDemand asks the judge model to extract structured fields
// from an inbox body. Same Ollama+JSON-format pattern as the
// generateParaphrase function. Real production would have a dedicated
// small model fine-tuned on staffing-language inbox parsing; here we
// use the same model that judges relevance. temperature=0 for
// deterministic extraction.
func parseInboxDemand(hc *http.Client, ollamaURL, model, inboxBody string) (*parsedDemand, error) {
	system := `You parse staffing requests from email/SMS bodies. Extract structured fields.
Output JSON ONLY, this exact shape: {"role": "...", "count": N, "location": "...", "certs": [...], "skills": [...], "shift": "..."}.

Rules:
- role: the job role being requested (lowercase, e.g. "warehouse worker", "forklift operator")
- count: number of workers needed (integer; if "a few" or unspecified, use 1)
- location: city + state if both mentioned (e.g. "Cleveland, OH"); city only if state missing
- certs: certification list as named in the body (e.g. ["OSHA-30", "forklift cert"])
- skills: skill list as named in the body (e.g. ["pallet jack", "spanish"])
- shift: "day"|"swing"|"night" if mentioned, else ""
- If a field isn't in the body, use empty string or empty array (never null)
- Do NOT explain — emit the JSON only.`
	body, _ := json.Marshal(map[string]any{
		"model":  model,
		"stream": false,
		"format": "json",
		"messages": []map[string]string{
			{"role": "system", "content": system},
			{"role": "user", "content": inboxBody},
		},
		"options": map[string]any{"temperature": 0},
	})
	req, _ := http.NewRequest("POST", ollamaURL+"/api/chat", bytes.NewReader(body))
	req.Header.Set("Content-Type", "application/json")
	resp, err := hc.Do(req)
	if err != nil {
		return nil, err
	}
	defer resp.Body.Close()
	if resp.StatusCode/100 != 2 {
		return nil, fmt.Errorf("ollama chat: HTTP %d", resp.StatusCode)
	}
	rb, _ := io.ReadAll(resp.Body)
	var ollamaResp struct {
		Message struct {
			Content string `json:"content"`
		} `json:"message"`
	}
	if err := json.Unmarshal(rb, &ollamaResp); err != nil {
		return nil, err
	}
	var out parsedDemand
	if err := json.Unmarshal([]byte(ollamaResp.Message.Content), &out); err != nil {
		return nil, fmt.Errorf("decode parsed demand: %w (content=%q)", err, ollamaResp.Message.Content)
	}
	if strings.TrimSpace(out.Role) == "" {
		return nil, fmt.Errorf("parsed demand has empty role (content=%q)", ollamaResp.Message.Content)
	}
	return &out, nil
}

// postInbox sends an inbox message (email or SMS) to observerd via
// the gateway. observerd records it as an ObservedOp with
// Source=SourceInbox; downstream actions (search, ingest, etc.) are
// the caller's concern.
func postInbox(hc *http.Client, gw, msgType, sender, subject, body, priority, tag string) error {
	bodyJSON, _ := json.Marshal(map[string]any{
		"type":     msgType,
		"sender":   sender,
		"subject":  subject,
		"body":     body,
		"priority": priority,
		"tag":      tag,
	})
	req, _ := http.NewRequest("POST", gw+"/v1/observer/inbox", bytes.NewReader(bodyJSON))
	req.Header.Set("Content-Type", "application/json")
	resp, err := hc.Do(req)
	if err != nil {
		return err
	}
	defer resp.Body.Close()
	if resp.StatusCode/100 != 2 {
		rb, _ := io.ReadAll(resp.Body)
		return fmt.Errorf("inbox %d: %s", resp.StatusCode, string(rb))
	}
	return nil
}

// coordByName looks up a coordinator by name. Used by inbox-triggered
// searches that route based on the email's tagged coordinator.
func coordByName(coords []Coordinator, name string) Coordinator {
	for _, c := range coords {
		if c.Name == name {
			return c
		}
	}
	return coords[0]
}

func playbookRecord(hc *http.Client, gw, query, answerID, answerCorpus string, score float64, corpus string) error {
	body, _ := json.Marshal(map[string]any{
		"query_text":    query,
		"answer_id":     answerID,
		"answer_corpus": answerCorpus,
		"score":         score,
		"tags":          []string{"multi-coord-stress"},
		"corpus":        corpus,
	})
	req, _ := http.NewRequest("POST", gw+"/v1/matrix/playbooks/record", bytes.NewReader(body))
	req.Header.Set("Content-Type", "application/json")
	resp, err := hc.Do(req)
	if err != nil {
		return err
	}
	defer resp.Body.Close()
	if resp.StatusCode/100 != 2 {
		rb, _ := io.ReadAll(resp.Body)
		return fmt.Errorf("playbook record %d: %s", resp.StatusCode, string(rb))
	}
	return nil
}

func must[T any](v T, err error) T {
	if err != nil {
		log.Fatalf("[stress] %v", err)
	}
	return v
}