diff --git a/.env.example b/.env.example
index c6447fe..f79d8e6 100644
--- a/.env.example
+++ b/.env.example
@@ -16,3 +16,10 @@ SMTP_FROM=noreply@notice.entroia.com
 # Optional Sign in with Slack.
 SLACK_CLIENT_ID=replace-me
 SLACK_CLIENT_SECRET=replace-me
+
+# Optional shared paper corpus; enable only after PostgreSQL publication.
+# Store the real read-only connection string in the protected server research.env.
+FOUNDRY_RESEARCH_DATABASE_URL=
+FOUNDRY_RESEARCH_SCHEMA=research
+# Deployment migration container; independent of database/schema names.
+FOUNDRY_POSTGRES_CONTAINER=foundry-postgres
diff --git a/.gitea/workflows/check.yml b/.gitea/workflows/check.yml
index f3ad8e9..c666327 100644
--- a/.gitea/workflows/check.yml
+++ b/.gitea/workflows/check.yml
@@ -74,6 +74,7 @@ jobs:
         env:
           PUBLIC_ORIGIN: https://api.example.com
           DATABASE_URL: postgres://foundry:fixture@127.0.0.1:5432/foundry
+          FOUNDRY_RESEARCH_DATABASE_URL: ''
           GOOGLE_CLIENT_ID: fixture
           GOOGLE_CLIENT_SECRET: fixture
           FOUNDRY_PROJECTS: '{"notes":"http://localhost:3000/auth/callback"}'
diff --git a/README.md b/README.md
index 1ea1502..2253a25 100644
--- a/README.md
+++ b/README.md
@@ -10,6 +10,10 @@
 4. 由 Gitea Actions 在 `ces-control` 上使用 Go 1.25 构建并检查。部署工作流仅处理 `main` 上影响后端或部署的改动，将二进制启动在服务器 Docker 私网，由 Traefik 提供 HTTPS。通过 `GET /healthz` 检查进程，通过 `GET /readyz` 检查认证库。
 5. 代理避免记录 Google 回调的查询参数及认证请求头。设置入口限流，并定时删除三张临时表（`oauth_attempts`、`login_grants`、`project_sessions`）中 `expires_at <= now()` 的记录。
 
+## 共享论文库（Thesisly）
+
+可选只读接入见[共享论文 API](docs/research-api.md)。Thesisly 使用同一后端的 `/projects/{project}/research/papers` 列表、搜索和详情；只有配置服务器上的 `FOUNDRY_RESEARCH_DATABASE_URL` 时才启用。连接串保存在受保护的 `research.env`，不放入前端或 Actions Variables。当前仍等待 Paper Atlas 全量 PostgreSQL 迁移完成及真实数据验收，尚未启用线上论文接口。
+
 ## Actions 部署配置
 
 在仓库的 Actions Secrets 配置 `GOOGLE_CLIENT_ID` 和 `GOOGLE_CLIENT_SECRET`，在 Actions Variables 配置 `FOUNDRY_PROJECTS`。变量值为 JSON 对象，例如 `{"prompt":"http://localhost:17321/auth/callback"}`。部署读取服务器上预先配置的 `/opt/endoleap-infra/foundry/database.env`，OAuth 值由 Actions 在启动容器时注入。
diff --git a/cmd/foundry/main.go b/cmd/foundry/main.go
index 5b3e2a2..32d4196 100644
--- a/cmd/foundry/main.go
+++ b/cmd/foundry/main.go
@@ -13,6 +13,7 @@ import (
 	"gitea.entroia.com/Zayden-Jung/foundry-core/internal/auth"
 	"gitea.entroia.com/Zayden-Jung/foundry-core/internal/codex"
 	"gitea.entroia.com/Zayden-Jung/foundry-core/internal/platform"
+	"gitea.entroia.com/Zayden-Jung/foundry-core/internal/research"
 	"github.com/jackc/pgx/v5"
 	"github.com/jackc/pgx/v5/stdlib"
 )
@@ -61,6 +62,33 @@ func main() {
 		log.Fatal(err)
 	}
 	mux := http.NewServeMux()
+	var papers research.Repository
+	if researchDSN := os.Getenv("FOUNDRY_RESEARCH_DATABASE_URL"); researchDSN != "" {
+		researchConfig, err := pgx.ParseConfig(researchDSN)
+		if err != nil {
+			log.Fatal("Invalid research database configuration")
+		}
+		researchConfig.RuntimeParams["default_transaction_read_only"] = "on"
+		researchConfig.RuntimeParams["application_name"] = "foundry-research"
+		researchDB := stdlib.OpenDB(*researchConfig)
+		defer researchDB.Close()
+		researchDB.SetMaxOpenConns(2)
+		researchDB.SetMaxIdleConns(2)
+		researchDB.SetConnMaxIdleTime(5 * time.Minute)
+		researchDB.SetConnMaxLifetime(30 * time.Minute)
+		repository, err := research.NewPostgresRepository(researchDB, env("FOUNDRY_RESEARCH_SCHEMA", "research"))
+		if err != nil {
+			log.Fatal("Invalid research database configuration")
+		}
+		ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+		err = repository.Check(ctx)
+		cancel()
+		if err != nil {
+			log.Fatal("Research database is unavailable or not published")
+		}
+		papers = repository
+	}
+	research.Register(mux, papers, callbacks)
 	if err := codex.Register(mux, callbacks, os.Getenv("FOUNDRY_CODEX_SOCKET"), os.Getenv("FOUNDRY_CODEX_GRANTS")); err != nil {
 		log.Fatal(err)
 	}
diff --git a/deploy/check-schema.sh b/deploy/check-schema.sh
index ab87d48..941baa7 100644
--- a/deploy/check-schema.sh
+++ b/deploy/check-schema.sh
@@ -67,12 +67,14 @@ END $$;
 ROLLBACK;
 SQL
 printf 'Migration preserved existing logins and enforced project isolation.\n'
-for package in platform auth; do
+psql < internal/research/testdata/corpus.sql
+for package in platform auth research; do
   CGO_ENABLED=0 go test -mod=readonly -c -o "bin/$package-tests" "./internal/$package"
   docker run --rm --network "container:$container" --user 1000:1000 \
     --read-only --cap-drop ALL --security-opt no-new-privileges \
     --tmpfs /tmp:rw,nosuid,nodev,noexec,size=16m,mode=1777 \
     --mount "type=bind,source=$(pwd)/bin/$package-tests,target=/opt/integration-tests,readonly" \
     -e PLATFORM_TEST_DATABASE_URL=postgres://foundry_fixture@127.0.0.1:5432/postgres?sslmode=disable \
+    -e RESEARCH_TEST_DATABASE_URL=postgres://foundry_research_fixture@127.0.0.1:5432/postgres?sslmode=disable \
     --entrypoint /opt/integration-tests postgres:17-alpine -test.v
 done
diff --git a/deploy/migrate.py b/deploy/migrate.py
index 7cc46ff..d1ff616 100644
--- a/deploy/migrate.py
+++ b/deploy/migrate.py
@@ -2,12 +2,13 @@
 """Apply checked migrations atomically from the trusted deployment runner."""
 import argparse
 import hashlib
+import os
 from pathlib import Path
 import re
 import subprocess
 
 parser = argparse.ArgumentParser()
-parser.add_argument("--container", default="endoleap-dev-postgres-1")
+parser.add_argument("--container", default=os.environ.get("FOUNDRY_POSTGRES_CONTAINER", "foundry-postgres"))
 parser.add_argument("--database", default="foundry_auth")
 parser.add_argument("--role", default="foundry")
 args = parser.parse_args()
diff --git a/deploy/release.sh b/deploy/release.sh
index af98e99..b75658c 100644
--- a/deploy/release.sh
+++ b/deploy/release.sh
@@ -53,6 +53,10 @@ mail_env=()
 if [[ -s "$foundry/mail.env" ]]; then
   mail_env=(--env-file "$foundry/mail.env")
 fi
+research_env=()
+if [[ -s "$foundry/research.env" ]]; then
+  research_env=(--env-file "$foundry/research.env")
+fi
 slack_env=()
 if [[ -n "${SLACK_CLIENT_ID:-}" || -n "${SLACK_CLIENT_SECRET:-}" ]]; then
   for name in SLACK_CLIENT_ID SLACK_CLIENT_SECRET; do
@@ -118,7 +122,7 @@ if [[ -d "$codex_dir" ]]; then
   )
 fi
 
-docker run -d --name "$candidate" "${codex_mounts[@]}" "${mail_env[@]}" "${slack_env[@]}" --network "$network" \
+docker run -d --name "$candidate" "${codex_mounts[@]}" "${mail_env[@]}" "${slack_env[@]}" "${research_env[@]}" --network "$network" \
   --env-file "$foundry/database.env" -e HOST -e PORT -e PUBLIC_ORIGIN \
   -e GOOGLE_CLIENT_ID -e GOOGLE_CLIENT_SECRET -e FOUNDRY_PROJECTS -e FOUNDRY_ONBOARDING \
   --user 1000:1000 --cap-drop ALL --security-opt no-new-privileges \
diff --git a/deploy/research/query-indexes.sql b/deploy/research/query-indexes.sql
new file mode 100644
index 0000000..314cf06
--- /dev/null
+++ b/deploy/research/query-indexes.sql
@@ -0,0 +1,14 @@
+-- Manual operator step against the published Paper Atlas schema, outside a
+-- transaction. Core's SELECT-only runtime role must never execute this file.
+-- These indexes support the actual author search and exact category filter;
+-- the migration already owns the title/abstract English GIN index.
+SET lock_timeout = '2s';
+SET statement_timeout = '20min';
+SET maintenance_work_mem = '32MB';
+SET max_parallel_maintenance_workers = 0;
+CREATE INDEX CONCURRENTLY IF NOT EXISTS paper_tags_tag_paper
+ ON research.paper_tags(tag_id,paper_id);
+CREATE INDEX CONCURRENTLY IF NOT EXISTS papers_authors_search_gin
+ ON research.papers USING gin(to_tsvector('simple'::regconfig,authors_json));
+-- After an interrupted concurrent build, inspect pg_index.indisvalid before
+-- reuse. IF NOT EXISTS alone does not repair an invalid index.
diff --git a/docs/research-api.md b/docs/research-api.md
new file mode 100644
index 0000000..08217ed
--- /dev/null
+++ b/docs/research-api.md
@@ -0,0 +1,145 @@
+# Shared research API
+
+This work belongs to Thesisly 研迹 and uses Foundry Core rather than a separate
+application backend. The corpus is public; account sessions and future private
+collections remain project scoped.
+
+## Current implementation boundary
+
+`internal/research` implements the HTTP contract and a PostgreSQL repository
+based on the fixed Paper Atlas `a3b677b` schema (TEXT/BIGINT columns, native
+English title/abstract search index). `cmd/foundry` enables it only when
+`FOUNDRY_RESEARCH_DATABASE_URL` is configured. The Paper Atlas migration is now published: 21 tables passed its independent
+count/content-hash acceptance. A dedicated `foundry_research_reader` role has
+SELECT on the ten tables consumed here, schema USAGE, and read-only defaults.
+Its connection is stored only in the protected server file below.
+The existing SQLite implementation was withdrawn. Real-corpus integration has
+been checked through a loopback Core and the local frontend. Formal production
+release still requires the user to merge and choose to deploy; this branch is
+not deployed by these checks. Very broad queries remain subject to the budget
+and performance limitation documented below.
+Do not add `FOUNDRY_RESEARCH_DATABASE`.
+
+The server-side deployment may load a protected
+`/opt/endoleap-infra/foundry/research.env` containing
+`FOUNDRY_RESEARCH_DATABASE_URL` and `FOUNDRY_RESEARCH_SCHEMA` (default `research`).
+Keep the connection string out of Git, Actions variables, browser configuration
+and command arguments. Use a dedicated role with schema USAGE and table SELECT
+only. Core uses a separate two-connection pool, read-only connection defaults and
+read-only transactions. Each research transaction disables JIT locally: the
+measured count plan spent about 2.85 seconds compiling JIT functions, and its
+end-to-end execution exceeded eight seconds. This does not alter shared
+PostgreSQL settings or replace exact counts with estimates. Startup checks the published schema/SELECT permissions;
+an explicitly enabled but invalid connection fails candidate readiness.
+
+Deployment migration defaults now follow the shared `foundry-postgres`
+container name. `FOUNDRY_POSTGRES_CONTAINER` or the existing `--container` option
+can select a different container without changing database/table names.
+
+## Routes
+
+- `GET /projects/{project}/research/papers`
+- `GET /projects/{project}/research/papers/{id}`
+
+Only configured Foundry projects are accepted. Reading the public corpus does
+not require a session; these routes do not expose private data or accept writes.
+
+List parameters are `q` (200 Unicode characters, 20 words), `searchIn`
+(`text` or `author`), `topic` (an exact arXiv category), `sort` (`recent` or
+`title`), `page` (1–1000), and `pageSize` (1–50, default 8). Unknown, malformed
+or repeated parameters return 400. `recent` means descending stable paper ID,
+not publication date. Search terms must become bound query parameters, never
+raw SQL or user-supplied full-text syntax.
+
+Title/abstract search uses the migration's English vector with
+`plainto_tsquery`; author search uses simple-language terms over the stored
+author JSON. Both treat input as terms, never full-text operators. Initial full-corpus measurements showed eight-second timeouts for text,
+author and topic counts; the unfiltered count took 7.25 seconds. Execution plans
+confirmed sequential scans for authors and category joins.
+`deploy/research/query-indexes.sql` supplies operator-run concurrent indexes for
+those two missing access paths. It is separate from Core startup and auth
+migrations; do not grant DDL to the runtime reader. Both indexes completed on the confirmed ECS (category 35.4 seconds, author
+436.47 seconds); measured end-to-end checks are documented below.
+
+List responses contain `items`, `total`, `topics`, `page`, and `pageSize`.
+Each paper preserves original title/abstract, authors, source dates, topics and
+source links. Publication decisions need source evidence; absent decisions stay
+unknown. No invented repositories, metrics or generated research conclusions.
+
+The repository counts and reads each list in one read-only snapshot.
+Paging first materializes matching IDs, then joins only that page to the wide
+paper rows. Filtered searches first materialize matching IDs without an
+ORDER BY/LIMIT, then sort/page those IDs, preventing sparse terms from choosing
+a backward heap scan with repeated text-vector computation. Activity sampling showed that selecting full rows before LIMIT
+caused sparse category queries to read large portions of the paper heap. The
+HTTP layer permits two in-flight queries with an eight-second context budget.
+Busy requests return 429 with `Retry-After: 1`; deadlines return 504; storage
+errors return a generic 503 without database details. Missing papers return 404.
+
+## Validation
+
+`go test -race -count=1 ./internal/research` covers routing, query limits,
+Unicode, anonymous reads, rejected mutations, project lookup, errors, saturation
+and timeout slot release. Run compilation/tests only on the user-confirmed ECS
+runner. These transport tests do not prove a database connection, full-corpus
+query performance or production availability.
+
+The PostgreSQL repository batches source/tag/identifier/submission queries in
+the same read-only repeatable-read transaction. Author tuples are displayed as
+given name, family name and suffix; an optional fourth affiliation field
+is excluded from the display name (observed in real arXiv records); null abstracts/dates remain null. Publication
+status comes from the latest observed decision, using decision date and ID as
+tie-breakers, and stays unknown when no decision exists. Unit checks cover
+author shapes, bound user values and schema qualification. The real Core detail
+and homepage HTTP routes have returned 200.
+The frontend has also rendered the actual 3,526,999-paper corpus through a
+loopback SSH tunnel: eight rows, matching detail title/source, desktop and
+mobile layout, no page exceptions or horizontal overflow in that check.
+Author search returned 146 records in 0.336 seconds. After paging/author fixes, category `cs.AR` returned 9,457 records in 4.906
+seconds, `CartoonGAN` returned three in 0.332 seconds, and the multi-term search
+`attention is all you need` returned 8,653 in 6.325 seconds. These were sequential
+manual HTTP checks, not a load test. Very broad text searches can still exceed
+the eight-second budget; clients retain search controls so users can narrow
+the query or retry. This known limit is not an estimated count or sample
+fallback. This is not a
+production deployment or an authentication acceptance result.
+
+The existing `deploy/check-schema.sh` also loads the synthetic fixture in
+`internal/research/testdata/corpus.sql` inside its disposable PostgreSQL
+container (`--network none`, no host port). The research test process receives a
+separate SELECT-only fixture role via `RESEARCH_TEST_DATABASE_URL`. These
+integration checks execute real SQL for filtering, stable paging, nullable
+content, source/tag/identifier joins, latest decision tie-breakers, permission
+denial and SQL-backed HTTP routes. They never connect to the shared migration
+instance and do not establish full-corpus performance or live-data correctness.
+
+## Text storage compatibility (repair 4e62710)
+
+The Paper Atlas repair uses `nul-soh-v1`: NUL is stored as SOH/SOH, and a
+literal SOH as SOH/STX, preserving binary text ordering. Core reads the schema
+comment `paper-atlas:text-codec=nul-soh-v1` through `pg_catalog` at startup,
+restores SQL text before JSON parsing/API responses, and encodes bound search
+terms. An absent/unrelated schema comment keeps legacy text unchanged; an
+unsupported marked codec or malformed encoded text fails explicitly.
+
+Schema comments are public metadata, so consumers need no privileges on the
+private migration controller. The research role still needs only schema USAGE
+and table SELECT. The codec adaptation is on the review branch, pushed with [skip ci] under the
+user's no-CI instruction; the published schema marker has been verified with
+the runtime reader.
+All research tests passed manually against disposable PostgreSQL on the
+confirmed ECS, including the encoded SQL fixture for schema detection, raw TEXT restoration,
+JSON author/metadata decoding, search and the HTTP round trip. Passing isolated
+fixtures does not establish full-corpus API performance.
+
+
+## Identity handoff
+
+The existing production Core received a configuration-only reload using its
+unchanged release binary: the original projects were retained and thesisly was
+registered. Both Google and Slack identity records were observed in this
+project. The user independently confirmed each provider's real authorization,
+reload/new-tab persistence and logout behavior. Browser control was unavailable,
+so these browser results are attributed to user acceptance rather than an
+automated real-account test. This does not deploy the research work branch;
+mail delivery was not exercised.
diff --git a/internal/research/http.go b/internal/research/http.go
new file mode 100644
index 0000000..9f86ac1
--- /dev/null
+++ b/internal/research/http.go
@@ -0,0 +1,210 @@
+// Package research serves the shared, public paper corpus. Account data and
+// private collections do not belong to these anonymous read-only endpoints.
+package research
+
+import (
+	"context"
+	"database/sql"
+	"encoding/json"
+	"errors"
+	"net/http"
+	"net/url"
+	"regexp"
+	"strconv"
+	"strings"
+	"time"
+	"unicode/utf8"
+)
+
+type Source struct {
+	Label      string `json:"label"`
+	URL        string `json:"url"`
+	ExternalID string `json:"externalId"`
+}
+
+type Publication struct {
+	Venue       string `json:"venue"`
+	Year        int    `json:"year"`
+	Track       string `json:"track"`
+	Status      string `json:"status"`
+	EvidenceURL string `json:"evidenceUrl,omitempty"`
+}
+
+type Paper struct {
+	ID           string        `json:"id"`
+	Title        string        `json:"title"`
+	Abstract     *string       `json:"abstract"`
+	Authors      []string      `json:"authors"`
+	PublishedAt  *string       `json:"publishedAt"`
+	Topics       []string      `json:"topics"`
+	Sources      []Source      `json:"sources"`
+	Publications []Publication `json:"publications,omitempty"`
+	DOI          *string       `json:"doi,omitempty"`
+	Journal      *string       `json:"journal,omitempty"`
+	Comments     *string       `json:"comments,omitempty"`
+}
+
+type Filters struct {
+	Query    string
+	SearchIn string
+	Topic    string
+	Sort     string
+	Page     int
+	PageSize int
+}
+
+type Page struct {
+	Items    []Paper  `json:"items"`
+	Total    int64    `json:"total"`
+	Topics   []string `json:"topics"`
+	Page     int      `json:"page"`
+	PageSize int      `json:"pageSize"`
+}
+
+// Repository implementations must use a read-only transaction for each List,
+// so its count, topics and page describe the same database snapshot.
+type Repository interface {
+	List(context.Context, Filters) (Page, error)
+	Get(context.Context, int64) (Paper, error)
+}
+
+type service struct {
+	repository Repository
+	projects   map[string]string
+	slots      chan struct{}
+	timeout    time.Duration
+}
+
+// Register adds only GET endpoints; no credentials or database errors are sent
+// to anonymous callers. A nil repository leaves the existing server unchanged.
+func Register(mux *http.ServeMux, repository Repository, projects map[string]string) {
+	if repository == nil {
+		return
+	}
+	s := &service{repository: repository, projects: projects, slots: make(chan struct{}, 2), timeout: 8 * time.Second}
+	mux.HandleFunc("GET /projects/{project}/research/papers", s.serve)
+	mux.HandleFunc("GET /projects/{project}/research/papers/{id}", s.serve)
+}
+
+func reply(w http.ResponseWriter, status int, value any) {
+	w.Header().Set("Content-Type", "application/json")
+	w.Header().Set("Cache-Control", "no-store")
+	w.Header().Set("X-Content-Type-Options", "nosniff")
+	w.WriteHeader(status)
+	_ = json.NewEncoder(w).Encode(value)
+}
+
+func reject(w http.ResponseWriter, status int, code string) {
+	reply(w, status, map[string]string{"error": code})
+}
+
+var topicPattern = regexp.MustCompile(`^[A-Za-z][A-Za-z0-9.-]{0,79}$`)
+var positiveInteger = regexp.MustCompile(`^[1-9][0-9]*$`)
+
+func parseFilters(raw string) (Filters, error) {
+	f := Filters{SearchIn: "text", Sort: "recent", Page: 1, PageSize: 8}
+	values, err := url.ParseQuery(raw)
+	if err != nil {
+		return f, err
+	}
+	invalid := errors.New("invalid query")
+	for key, value := range values {
+		if len(value) != 1 {
+			return f, invalid
+		}
+		switch key {
+		case "q":
+			if !utf8.ValidString(value[0]) || utf8.RuneCountInString(value[0]) > 200 || len(strings.Fields(value[0])) > 20 || strings.ContainsRune(value[0], 0) {
+				return f, invalid
+			}
+			f.Query = strings.TrimSpace(value[0])
+		case "searchIn":
+			if value[0] != "text" && value[0] != "author" {
+				return f, invalid
+			}
+			f.SearchIn = value[0]
+		case "topic":
+			if value[0] != "" && !topicPattern.MatchString(value[0]) {
+				return f, invalid
+			}
+			f.Topic = value[0]
+		case "sort":
+			if value[0] != "recent" && value[0] != "title" {
+				return f, invalid
+			}
+			f.Sort = value[0]
+		case "page", "pageSize":
+			if !positiveInteger.MatchString(value[0]) {
+				return f, invalid
+			}
+			n, err := strconv.Atoi(value[0])
+			limit := 1000
+			if key == "pageSize" {
+				limit = 50
+			}
+			if err != nil || n > limit {
+				return f, invalid
+			}
+			if key == "page" {
+				f.Page = n
+			} else {
+				f.PageSize = n
+			}
+		default:
+			return f, invalid
+		}
+	}
+	return f, nil
+}
+
+func (s *service) serve(w http.ResponseWriter, r *http.Request) {
+	if _, exists := s.projects[r.PathValue("project")]; !exists {
+		reject(w, 404, "not_found")
+		return
+	}
+	var filters Filters
+	var id int64
+	var err error
+	if value := r.PathValue("id"); value != "" {
+		if !positiveInteger.MatchString(value) || r.URL.RawQuery != "" {
+			reject(w, 400, "invalid_request")
+			return
+		}
+		id, err = strconv.ParseInt(value, 10, 64)
+	} else {
+		filters, err = parseFilters(r.URL.RawQuery)
+	}
+	if err != nil {
+		reject(w, 400, "invalid_request")
+		return
+	}
+	select {
+	case s.slots <- struct{}{}:
+		defer func() { <-s.slots }()
+	default:
+		w.Header().Set("Retry-After", "1")
+		reject(w, 429, "research_busy")
+		return
+	}
+	ctx, cancel := context.WithTimeout(r.Context(), s.timeout)
+	defer cancel()
+	var result any
+	if id != 0 {
+		result, err = s.repository.Get(ctx, id)
+	} else {
+		result, err = s.repository.List(ctx, filters)
+	}
+	if err == nil {
+		reply(w, 200, result)
+		return
+	}
+	var state interface{ SQLState() string }
+	switch {
+	case errors.Is(err, sql.ErrNoRows):
+		reject(w, 404, "not_found")
+	case errors.Is(ctx.Err(), context.DeadlineExceeded), errors.Is(err, context.DeadlineExceeded), errors.As(err, &state) && state.SQLState() == "57014":
+		reject(w, 504, "research_timeout")
+	default:
+		reject(w, 503, "research_unavailable")
+	}
+}
diff --git a/internal/research/http_test.go b/internal/research/http_test.go
new file mode 100644
index 0000000..b9267fc
--- /dev/null
+++ b/internal/research/http_test.go
@@ -0,0 +1,145 @@
+package research
+
+import (
+	"context"
+	"database/sql"
+	"encoding/json"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"testing"
+	"time"
+)
+
+type fixtureRepository struct {
+	filters Filters
+	id      int64
+	err     error
+	wait    bool
+	calls   int
+}
+
+func (f *fixtureRepository) List(ctx context.Context, filters Filters) (Page, error) {
+	f.calls++
+	f.filters = filters
+	if f.wait {
+		<-ctx.Done()
+		return Page{}, ctx.Err()
+	}
+	return Page{Items: []Paper{}, Topics: []string{}, Page: filters.Page, PageSize: filters.PageSize}, f.err
+}
+
+func (f *fixtureRepository) Get(ctx context.Context, id int64) (Paper, error) {
+	f.calls++
+	f.id = id
+	return Paper{ID: "42", Title: "Original title", Authors: []string{}, Topics: []string{}, Sources: []Source{}}, f.err
+}
+
+func request(handler http.Handler, method, path string) *httptest.ResponseRecorder {
+	w := httptest.NewRecorder()
+	handler.ServeHTTP(w, httptest.NewRequest(method, path, nil))
+	return w
+}
+
+func TestAnonymousReadRoutes(t *testing.T) {
+	f := &fixtureRepository{}
+	mux := http.NewServeMux()
+	Register(mux, f, map[string]string{"thesisly": "http://localhost:17322/auth/callback"})
+	w := request(mux, "GET", "/projects/thesisly/research/papers?q=graph&searchIn=author&sort=title&topic=cs.LG&page=2&pageSize=10")
+	if w.Code != 200 || f.filters != (Filters{Query: "graph", SearchIn: "author", Topic: "cs.LG", Sort: "title", Page: 2, PageSize: 10}) {
+		t.Fatalf("list: %d %+v", w.Code, f.filters)
+	}
+	var page Page
+	if err := json.Unmarshal(w.Body.Bytes(), &page); err != nil || page.Items == nil || page.Topics == nil {
+		t.Fatalf("invalid page: %s", w.Body.String())
+	}
+	if w.Header().Get("Cache-Control") != "no-store" || w.Header().Get("Content-Type") != "application/json" {
+		t.Fatal("missing response headers")
+	}
+	w = request(mux, "GET", "/projects/thesisly/research/papers/42")
+	if w.Code != 200 || f.id != 42 || !strings.Contains(w.Body.String(), `"abstract":null`) {
+		t.Fatalf("detail: %d %s", w.Code, w.Body.String())
+	}
+	if got := request(mux, "POST", "/projects/thesisly/research/papers").Code; got != 405 {
+		t.Fatalf("mutation status %d", got)
+	}
+	if got := request(mux, "GET", "/projects/other/research/papers").Code; got != 404 {
+		t.Fatalf("unknown project status %d", got)
+	}
+}
+
+func TestInvalidQueriesNeverReachRepository(t *testing.T) {
+	queries := []string{"unknown=1", "q=x&q=y", "q=%ff", "q=%00", "q=%zz", "page=0", "page=1001", "page=-1", "page=1.5", "page=01", "pageSize=51", "pageSize=0", "pageSize=9999999999999999999999", "searchIn=sql", "sort=published", "topic=arxiv:cs.LG", "q=" + url.QueryEscape(strings.Repeat("字", 201)), "q=" + url.QueryEscape(strings.Repeat("word ", 21))}
+	for _, query := range queries {
+		t.Run(query, func(t *testing.T) {
+			f := &fixtureRepository{}
+			mux := http.NewServeMux()
+			Register(mux, f, map[string]string{"thesisly": "unused"})
+			w := request(mux, "GET", "/projects/thesisly/research/papers?"+query)
+			if w.Code != 400 || f.calls != 0 {
+				t.Fatalf("status %d, calls %d", w.Code, f.calls)
+			}
+		})
+	}
+	for _, path := range []string{"0", "-1", "1x", "9223372036854775808", "1?extra=1"} {
+		mux := http.NewServeMux()
+		f := &fixtureRepository{}
+		Register(mux, f, map[string]string{"thesisly": "unused"})
+		if w := request(mux, "GET", "/projects/thesisly/research/papers/"+path); w.Code != 400 || f.calls != 0 {
+			t.Fatalf("invalid id %q: %d", path, w.Code)
+		}
+	}
+}
+
+func TestQueryDefaultsAndUnicode(t *testing.T) {
+	f, err := parseFilters("q=" + url.QueryEscape(" 论文 Graph "))
+	if err != nil || f.Query != "论文 Graph" || f.Page != 1 || f.PageSize != 8 || f.SearchIn != "text" || f.Sort != "recent" {
+		t.Fatalf("%+v %v", f, err)
+	}
+}
+
+type databaseFailure string
+
+func (failure databaseFailure) Error() string    { return "private database error" }
+func (failure databaseFailure) SQLState() string { return string(failure) }
+
+func TestStorageErrorsAreRedacted(t *testing.T) {
+	for _, test := range []struct {
+		err    error
+		status int
+		code   string
+	}{
+		{sql.ErrNoRows, 404, "not_found"},
+		{context.DeadlineExceeded, 504, "research_timeout"},
+		{errors.Join(databaseFailure("57014"), errors.New("private statement")), 504, "research_timeout"},
+		{databaseFailure("42501"), 503, "research_unavailable"},
+		{errors.New("postgres://private-user:private-password@db/corpus"), 503, "research_unavailable"},
+	} {
+		mux := http.NewServeMux()
+		Register(mux, &fixtureRepository{err: test.err}, map[string]string{"thesisly": "unused"})
+		w := request(mux, "GET", "/projects/thesisly/research/papers/42")
+		if w.Code != test.status || !strings.Contains(w.Body.String(), test.code) || strings.Contains(w.Body.String(), "private") {
+			t.Fatalf("%d %s", w.Code, w.Body.String())
+		}
+	}
+}
+
+func TestConcurrencyAndTimeoutReleaseSlots(t *testing.T) {
+	s := &service{repository: &fixtureRepository{wait: true}, projects: map[string]string{"thesisly": "unused"}, slots: make(chan struct{}, 2), timeout: time.Millisecond}
+	mux := http.NewServeMux()
+	mux.HandleFunc("GET /projects/{project}/research/papers", s.serve)
+	s.slots <- struct{}{}
+	s.slots <- struct{}{}
+	w := request(mux, "GET", "/projects/thesisly/research/papers")
+	if w.Code != 429 || w.Header().Get("Retry-After") != "1" {
+		t.Fatalf("busy %d", w.Code)
+	}
+	<-s.slots
+	<-s.slots
+	w = request(mux, "GET", "/projects/thesisly/research/papers")
+	if w.Code != 504 || len(s.slots) != 0 {
+		t.Fatalf("timeout %d, slots %d", w.Code, len(s.slots))
+	}
+}
diff --git a/internal/research/postgres.go b/internal/research/postgres.go
new file mode 100644
index 0000000..7198dc2
--- /dev/null
+++ b/internal/research/postgres.go
@@ -0,0 +1,389 @@
+package research
+
+import (
+	"context"
+	"database/sql"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"regexp"
+	"strconv"
+	"strings"
+)
+
+// The fixed Paper Atlas migration a3b677b preserves business columns as TEXT
+// and BIGINT and creates the English title/abstract GIN expression below.
+// No schema creation, indexing or other writes belong to this repository.
+type PostgresRepository struct {
+	db      *sql.DB
+	prefix  string
+	schema  string
+	encoded bool
+}
+
+var schemaName = regexp.MustCompile(`^[a-z][a-z0-9_]{0,35}$`)
+
+func NewPostgresRepository(db *sql.DB, schema string) (*PostgresRepository, error) {
+	if db == nil || !schemaName.MatchString(schema) {
+		return nil, errors.New("invalid research database configuration")
+	}
+	return &PostgresRepository{db: db, prefix: `"` + schema + `".`, schema: schema}, nil
+}
+
+func (r *PostgresRepository) query(value string) string {
+	return strings.ReplaceAll(value, "{schema}", r.prefix)
+}
+
+func (r *PostgresRepository) begin(ctx context.Context) (*sql.Tx, error) {
+	tx, err := r.db.BeginTx(ctx, &sql.TxOptions{Isolation: sql.LevelRepeatableRead, ReadOnly: true})
+	if err != nil {
+		return nil, err
+	}
+	_, err = tx.ExecContext(ctx, `SELECT set_config('statement_timeout','8000',true),
+		set_config('lock_timeout','1000',true), set_config('work_mem','2MB',true),
+		set_config('max_parallel_workers_per_gather','0',true), set_config('jit','off',true)`)
+	if err != nil {
+		_ = tx.Rollback()
+		return nil, err
+	}
+	return tx, nil
+}
+
+// Check resolves required columns and SELECT permissions without scanning the
+// corpus. The migration publishes the complete schema atomically.
+func (r *PostgresRepository) Check(ctx context.Context) error {
+	tx, err := r.begin(ctx)
+	if err != nil {
+		return err
+	}
+	defer tx.Rollback()
+	var annotation sql.NullString
+	err = tx.QueryRowContext(ctx, `SELECT pg_catalog.obj_description(oid,'pg_namespace')
+		FROM pg_catalog.pg_namespace WHERE nspname=$1`, r.schema).Scan(&annotation)
+	if err != nil {
+		return err
+	}
+	encoded, err := encodedSchema(annotation)
+	if err != nil {
+		return err
+	}
+	rows, err := tx.QueryContext(ctx, r.query(`SELECT p.id,p.title,p.abstract,p.authors_json,p.first_published,p.title_key,
+		r.external_id,r.url,r.metadata_json,so.name,i.scheme,i.value,t.namespace,t.value,
+		s.id,e.year,e.track,v.acronym,d.status,d.evidence_url,d.observed_at,d.decision_date
+		FROM {schema}papers p LEFT JOIN {schema}source_records r ON r.paper_id=p.id
+		LEFT JOIN {schema}sources so ON so.id=r.source_id LEFT JOIN {schema}identifiers i ON i.paper_id=p.id
+		LEFT JOIN {schema}paper_tags pt ON pt.paper_id=p.id LEFT JOIN {schema}tags t ON t.id=pt.tag_id
+		LEFT JOIN {schema}submissions s ON s.paper_id=p.id LEFT JOIN {schema}editions e ON e.id=s.edition_id
+		LEFT JOIN {schema}venues v ON v.id=e.venue_id LEFT JOIN {schema}decisions d ON d.submission_id=s.id WHERE false`))
+	if err != nil {
+		return err
+	}
+	if err = rows.Close(); err != nil {
+		return err
+	}
+	if err = tx.Commit(); err != nil {
+		return err
+	}
+	r.encoded = encoded
+	return nil
+}
+
+func whereClause(filters Filters) (string, []any) {
+	clauses, values := []string{}, []any{}
+	if filters.Query != "" {
+		values = append(values, filters.Query)
+		if filters.SearchIn == "author" {
+			// JSON stores author strings or [family,given,suffix] tuples. Plain
+			// full-text terms search names regardless of tuple display order.
+			clauses = append(clauses, `to_tsvector('simple'::regconfig,p.authors_json) @@ plainto_tsquery('simple'::regconfig,$1)`)
+		} else {
+			clauses = append(clauses, `to_tsvector('english'::regconfig,coalesce(p.title,'') || ' ' || coalesce(p.abstract,'')) @@ plainto_tsquery('english'::regconfig,$1)`)
+		}
+	}
+	if filters.Topic != "" {
+		values = append(values, filters.Topic)
+		clauses = append(clauses, fmt.Sprintf(`EXISTS(SELECT 1 FROM {schema}paper_tags pt
+			JOIN {schema}tags t ON t.id=pt.tag_id
+			WHERE pt.paper_id=p.id AND t.namespace='arxiv:category' AND t.value=$%d)`, len(values)))
+	}
+	if len(clauses) == 0 {
+		return "", values
+	}
+	return " WHERE " + strings.Join(clauses, " AND "), values
+}
+
+type scanner interface{ Scan(...any) error }
+
+func (r *PostgresRepository) scanPaper(row scanner) (Paper, error) {
+	p := Paper{Authors: []string{}, Topics: []string{}, Sources: []Source{}}
+	var id int64
+	var abstract, published sql.NullString
+	var authors string
+	if err := row.Scan(&id, r.text(&p.Title), r.text(&abstract), r.text(&authors), r.text(&published)); err != nil {
+		return p, err
+	}
+	p.ID = strconv.FormatInt(id, 10)
+	if abstract.Valid {
+		p.Abstract = &abstract.String
+	}
+	if published.Valid {
+		p.PublishedAt = &published.String
+	}
+	var err error
+	p.Authors, err = decodeAuthors(authors)
+	return p, err
+}
+
+func decodeAuthors(raw string) ([]string, error) {
+	var entries []json.RawMessage
+	if err := json.Unmarshal([]byte(raw), &entries); err != nil {
+		return nil, errors.New("invalid stored authors")
+	}
+	names := []string{}
+	for _, entry := range entries {
+		var name string
+		if json.Unmarshal(entry, &name) == nil {
+			if name = strings.TrimSpace(name); name != "" {
+				names = append(names, name)
+			}
+			continue
+		}
+		var tuple []string
+		if json.Unmarshal(entry, &tuple) != nil || len(tuple) == 0 || len(tuple) > 4 {
+			return nil, errors.New("unsupported stored author")
+		}
+		// arXiv may append an affiliation after family, given and suffix.
+		// It is source metadata, not part of the display name.
+		if len(tuple) == 4 {
+			tuple = tuple[:3]
+		}
+		if len(tuple) >= 2 {
+			tuple[0], tuple[1] = tuple[1], tuple[0]
+		}
+		parts := []string{}
+		for _, part := range tuple {
+			if part = strings.TrimSpace(part); part != "" {
+				parts = append(parts, part)
+			}
+		}
+		if name = strings.Join(parts, " "); name != "" {
+			names = append(names, name)
+		}
+	}
+	return names, nil
+}
+
+func (r *PostgresRepository) List(ctx context.Context, filters Filters) (Page, error) {
+	result := Page{Items: []Paper{}, Topics: []string{}, Page: filters.Page, PageSize: filters.PageSize}
+	tx, err := r.begin(ctx)
+	if err != nil {
+		return result, err
+	}
+	defer tx.Rollback()
+	if r.encoded {
+		filters.Query = encodeText(filters.Query)
+		filters.Topic = encodeText(filters.Topic)
+	}
+	where, args := whereClause(filters)
+	if err = tx.QueryRowContext(ctx, r.query(`SELECT count(*) FROM {schema}papers p`+where), args...).Scan(&result.Total); err != nil {
+		return result, err
+	}
+	order := "p.id DESC"
+	if filters.Sort == "title" {
+		order = "p.title_key,p.id"
+	}
+	args = append(args, filters.PageSize, (filters.Page-1)*filters.PageSize)
+	// Select IDs before loading potentially TOASTed titles/abstracts. With
+	// sparse categories, selecting full rows here can make the planner walk
+	// hundreds of thousands of wide heap rows before finding eight matches.
+	candidateQuery := `SELECT p.id FROM {schema}papers p`
+	candidateOrder := order
+	prefix := ""
+	if where != "" {
+		columns := "p.id"
+		if filters.Sort == "title" {
+			columns += ",p.title_key"
+		}
+		// Apply search before LIMIT/ORDER BY so a sparse term cannot turn
+		// into a backward heap scan that recomputes vectors for millions.
+		prefix = fmt.Sprintf(`matches AS MATERIALIZED (SELECT %s FROM {schema}papers p%s), `, columns, where)
+		candidateQuery = "SELECT id FROM matches"
+		candidateOrder = strings.ReplaceAll(order, "p.", "")
+	}
+	query := fmt.Sprintf(`WITH %spage_ids AS MATERIALIZED (
+		%s ORDER BY %s LIMIT $%d OFFSET $%d
+		) SELECT p.id,p.title,p.abstract,p.authors_json,p.first_published
+		FROM page_ids JOIN {schema}papers p ON p.id=page_ids.id ORDER BY %s`, prefix, candidateQuery, candidateOrder, len(args)-1, len(args), order)
+	rows, err := tx.QueryContext(ctx, r.query(query), args...)
+	if err != nil {
+		return result, err
+	}
+	for rows.Next() {
+		p, err := r.scanPaper(rows)
+		if err != nil {
+			_ = rows.Close()
+			return result, err
+		}
+		result.Items = append(result.Items, p)
+	}
+	err = rows.Err()
+	_ = rows.Close()
+	if err != nil {
+		return result, err
+	}
+	if err = r.enrich(ctx, tx, result.Items); err != nil {
+		return result, err
+	}
+	rows, err = tx.QueryContext(ctx, r.query(`SELECT value FROM {schema}tags WHERE namespace='arxiv:category' ORDER BY value`))
+	if err != nil {
+		return result, err
+	}
+	for rows.Next() {
+		var topic string
+		if err = rows.Scan(r.text(&topic)); err != nil {
+			_ = rows.Close()
+			return result, err
+		}
+		result.Topics = append(result.Topics, topic)
+	}
+	err = rows.Err()
+	_ = rows.Close()
+	if err != nil {
+		return result, err
+	}
+	return result, tx.Commit()
+}
+
+func (r *PostgresRepository) Get(ctx context.Context, id int64) (Paper, error) {
+	tx, err := r.begin(ctx)
+	if err != nil {
+		return Paper{}, err
+	}
+	defer tx.Rollback()
+	p, err := r.scanPaper(tx.QueryRowContext(ctx, r.query(`SELECT id,title,abstract,authors_json,first_published FROM {schema}papers WHERE id=$1`), id))
+	if err != nil {
+		return p, err
+	}
+	papers := []Paper{p}
+	if err = r.enrich(ctx, tx, papers); err != nil {
+		return p, err
+	}
+	return papers[0], tx.Commit()
+}
+
+// Batch metadata queries avoid per-paper round trips and share the count/page
+// snapshot. User values are bound; only the validated schema is interpolated.
+func (r *PostgresRepository) enrich(ctx context.Context, tx *sql.Tx, papers []Paper) error {
+	if len(papers) == 0 {
+		return nil
+	}
+	ids := make([]int64, len(papers))
+	byID := map[int64]*Paper{}
+	for i := range papers {
+		id, err := strconv.ParseInt(papers[i].ID, 10, 64)
+		if err != nil {
+			return err
+		}
+		ids[i] = id
+		byID[id] = &papers[i]
+	}
+	rows, err := tx.QueryContext(ctx, r.query(`SELECT DISTINCT pt.paper_id,t.value
+		FROM {schema}paper_tags pt JOIN {schema}tags t ON t.id=pt.tag_id
+		WHERE pt.paper_id=ANY($1) AND t.namespace='arxiv:category' ORDER BY pt.paper_id,t.value`), ids)
+	if err != nil {
+		return err
+	}
+	for rows.Next() {
+		var id int64
+		var topic string
+		if err = rows.Scan(&id, r.text(&topic)); err != nil {
+			_ = rows.Close()
+			return err
+		}
+		byID[id].Topics = append(byID[id].Topics, topic)
+	}
+	err = rows.Err()
+	_ = rows.Close()
+	if err != nil {
+		return err
+	}
+	rows, err = tx.QueryContext(ctx, r.query(`SELECT r.paper_id,s.name,r.external_id,r.url,r.metadata_json
+		FROM {schema}source_records r JOIN {schema}sources s ON s.id=r.source_id
+		WHERE r.paper_id=ANY($1) ORDER BY r.paper_id,r.id`), ids)
+	if err != nil {
+		return err
+	}
+	for rows.Next() {
+		var id int64
+		var source Source
+		var metadata string
+		if err = rows.Scan(&id, r.text(&source.Label), r.text(&source.ExternalID), r.text(&source.URL), r.text(&metadata)); err != nil {
+			_ = rows.Close()
+			return err
+		}
+		p := byID[id]
+		switch source.Label {
+		case "arxiv":
+			source.Label = "arXiv"
+		case "openreview":
+			source.Label = "OpenReview"
+		case "pmlr":
+			source.Label = "PMLR"
+		}
+		p.Sources = append(p.Sources, source)
+		var fields map[string]json.RawMessage
+		if json.Unmarshal([]byte(metadata), &fields) == nil {
+			for key, target := range map[string]**string{"doi": &p.DOI, "journal-ref": &p.Journal, "comments": &p.Comments} {
+				var value string
+				if *target == nil && json.Unmarshal(fields[key], &value) == nil && strings.TrimSpace(value) != "" {
+					*target = &value
+				}
+			}
+		}
+	}
+	err = rows.Err()
+	_ = rows.Close()
+	if err != nil {
+		return err
+	}
+	rows, err = tx.QueryContext(ctx, r.query(`SELECT paper_id,value FROM {schema}identifiers WHERE paper_id=ANY($1) AND scheme='doi' ORDER BY paper_id,value`), ids)
+	if err != nil {
+		return err
+	}
+	for rows.Next() {
+		var id int64
+		var value string
+		if err = rows.Scan(&id, r.text(&value)); err != nil {
+			_ = rows.Close()
+			return err
+		}
+		if byID[id].DOI == nil && strings.TrimSpace(value) != "" {
+			byID[id].DOI = &value
+		}
+	}
+	err = rows.Err()
+	_ = rows.Close()
+	if err != nil {
+		return err
+	}
+	rows, err = tx.QueryContext(ctx, r.query(`SELECT s.paper_id,v.acronym,e.year,e.track,coalesce(d.status,'unknown'),coalesce(d.evidence_url,'')
+		FROM {schema}submissions s JOIN {schema}editions e ON e.id=s.edition_id JOIN {schema}venues v ON v.id=e.venue_id
+		LEFT JOIN LATERAL (SELECT status,evidence_url FROM {schema}decisions WHERE submission_id=s.id
+			ORDER BY observed_at DESC,coalesce(decision_date,observed_at) DESC,id DESC LIMIT 1) d ON true
+		WHERE s.paper_id=ANY($1) ORDER BY s.paper_id,e.year DESC,v.acronym,e.track,s.id`), ids)
+	if err != nil {
+		return err
+	}
+	for rows.Next() {
+		var id int64
+		var publication Publication
+		if err = rows.Scan(&id, r.text(&publication.Venue), &publication.Year, r.text(&publication.Track), r.text(&publication.Status), r.text(&publication.EvidenceURL)); err != nil {
+			_ = rows.Close()
+			return err
+		}
+		byID[id].Publications = append(byID[id].Publications, publication)
+	}
+	err = rows.Err()
+	_ = rows.Close()
+	return err
+}
diff --git a/internal/research/postgres_integration_test.go b/internal/research/postgres_integration_test.go
new file mode 100644
index 0000000..65c999b
--- /dev/null
+++ b/internal/research/postgres_integration_test.go
@@ -0,0 +1,186 @@
+package research
+
+import (
+	"context"
+	"database/sql"
+	"encoding/json"
+	"errors"
+	"net/http"
+	"os"
+	"reflect"
+	"testing"
+	"time"
+
+	_ "github.com/jackc/pgx/v5/stdlib"
+)
+
+func integrationRepository(t *testing.T) (*PostgresRepository, context.Context) {
+	t.Helper()
+	dsn := os.Getenv("RESEARCH_TEST_DATABASE_URL")
+	if dsn == "" {
+		t.Skip("requires the disposable research fixture database")
+	}
+	db, err := sql.Open("pgx", dsn)
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+	db.SetMaxOpenConns(2)
+	repository, err := NewPostgresRepository(db, "research_fixture")
+	if err != nil {
+		t.Fatal(err)
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 20*time.Second)
+	t.Cleanup(cancel)
+	return repository, ctx
+}
+
+func paperIDs(papers []Paper) []string {
+	ids := []string{}
+	for _, paper := range papers {
+		ids = append(ids, paper.ID)
+	}
+	return ids
+}
+
+func TestPostgresPageSearchAndFilters(t *testing.T) {
+	r, ctx := integrationRepository(t)
+	if err := r.Check(ctx); err != nil {
+		t.Fatal(err)
+	}
+	for _, test := range []struct {
+		filters Filters
+		total   int64
+		ids     []string
+	}{
+		{Filters{Page: 1, PageSize: 2}, 5, []string{"5", "4"}},
+		{Filters{Page: 2, PageSize: 2}, 5, []string{"3", "2"}},
+		{Filters{Page: 3, PageSize: 2}, 5, []string{"1"}},
+		{Filters{Page: 1, PageSize: 2, Sort: "title"}, 5, []string{"2", "5"}},
+		{Filters{Page: 1, PageSize: 8, Query: "graph"}, 2, []string{"3", "1"}},
+		{Filters{Page: 1, PageSize: 8, Query: "graph OR zebra"}, 0, []string{}},
+		{Filters{Page: 1, PageSize: 8, Query: "Ada Smith", SearchIn: "author"}, 1, []string{"1"}},
+		{Filters{Page: 1, PageSize: 8, Topic: "cs.LG"}, 1, []string{"1"}},
+		{Filters{Page: 1, PageSize: 8, Topic: "cs.DC", Query: "graph"}, 1, []string{"3"}},
+		{Filters{Page: 1, PageSize: 8, Query: `x'); DROP TABLE papers;--`}, 0, []string{}},
+	} {
+		page, err := r.List(ctx, test.filters)
+		if err != nil {
+			t.Fatalf("%+v: %v", test.filters, err)
+		}
+		if page.Total != test.total || !reflect.DeepEqual(paperIDs(page.Items), test.ids) {
+			t.Fatalf("%+v: total=%d ids=%v", test.filters, page.Total, paperIDs(page.Items))
+		}
+		if !reflect.DeepEqual(page.Topics, []string{"cs.AI", "cs.DC", "cs.LG"}) {
+			t.Fatalf("mixed tag namespaces: %v", page.Topics)
+		}
+		if page.Page != test.filters.Page || page.PageSize != test.filters.PageSize {
+			t.Fatal("lost paging metadata")
+		}
+	}
+}
+
+func TestPostgresMetadataPreservesEvidence(t *testing.T) {
+	r, ctx := integrationRepository(t)
+	p, err := r.Get(ctx, 1)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if p.Title != "Graph & learning <original>" || p.Abstract == nil || *p.Abstract != "Neural graph models preserve original abstracts." || !reflect.DeepEqual(p.Authors, []string{"Ada Smith", "The Research Group"}) {
+		t.Fatalf("original content changed: %+v", p)
+	}
+	if !reflect.DeepEqual(p.Topics, []string{"cs.AI", "cs.LG"}) || len(p.Sources) != 1 || p.Sources[0].ExternalID != "fixture-1" || p.Sources[0].Label != "arXiv" {
+		t.Fatalf("incorrect source/tag joins: %+v", p)
+	}
+	if p.DOI == nil || *p.DOI != "10.fixture/canonical" || p.Journal == nil || *p.Journal != "Journal Fixture" || p.Comments == nil || *p.Comments != "Original comment" {
+		t.Fatalf("metadata missing: %+v", p)
+	}
+	if len(p.Publications) != 1 || p.Publications[0].Status != "withdrawn" || p.Publications[0].EvidenceURL != "https://example.test/decision/3" {
+		t.Fatalf("latest decision ordering changed: %+v", p.Publications)
+	}
+	p, err = r.Get(ctx, 3)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !reflect.DeepEqual(p.Authors, []string{"Carla Network"}) || len(p.Sources) != 2 || p.DOI == nil || *p.DOI != "10.fixture/other" || len(p.Publications) != 1 || p.Publications[0].Status != "unknown" || p.Publications[0].EvidenceURL != "" {
+		t.Fatalf("unknown decision/source mapping: %+v", p)
+	}
+	p, err = r.Get(ctx, 2)
+	if err != nil || p.Abstract != nil || p.PublishedAt != nil || p.Sources == nil || p.Topics == nil {
+		t.Fatalf("nullable data changed: %+v %v", p, err)
+	}
+	if _, err = r.Get(ctx, 99999); !errors.Is(err, sql.ErrNoRows) {
+		t.Fatalf("missing paper: %v", err)
+	}
+}
+
+func TestPostgresCorpusIsReadOnlyAndHTTPWorks(t *testing.T) {
+	r, ctx := integrationRepository(t)
+	tx, err := r.begin(ctx)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var readOnly, isolation, jit string
+	if err = tx.QueryRowContext(ctx, "SHOW transaction_read_only").Scan(&readOnly); err != nil {
+		t.Fatal(err)
+	}
+	if err = tx.QueryRowContext(ctx, "SHOW transaction_isolation").Scan(&isolation); err != nil {
+		t.Fatal(err)
+	}
+	if err = tx.QueryRowContext(ctx, "SHOW jit").Scan(&jit); err != nil {
+		t.Fatal(err)
+	}
+	_ = tx.Rollback()
+	if readOnly != "on" || isolation != "repeatable read" || jit != "off" {
+		t.Fatalf("unexpected read transaction settings: %s %s jit=%s", readOnly, isolation, jit)
+	}
+	_, err = r.db.ExecContext(ctx, `UPDATE research_fixture.papers SET title='unexpected write' WHERE id=1`)
+	var state interface{ SQLState() string }
+	if !errors.As(err, &state) || state.SQLState() != "42501" {
+		t.Fatalf("expected database permission denial, got %v", err)
+	}
+	mux := http.NewServeMux()
+	Register(mux, r, map[string]string{"thesisly": "http://localhost:17322/auth/callback"})
+	w := request(mux, "GET", "/projects/thesisly/research/papers?q=graph")
+	var page Page
+	if w.Code != 200 || json.Unmarshal(w.Body.Bytes(), &page) != nil || page.Total != 2 {
+		t.Fatalf("SQL-backed HTTP list: %d %s", w.Code, w.Body.String())
+	}
+	w = request(mux, "GET", "/projects/thesisly/research/papers/1")
+	var paper Paper
+	if w.Code != 200 || json.Unmarshal(w.Body.Bytes(), &paper) != nil || paper.ID != "1" {
+		t.Fatalf("SQL-backed HTTP detail: %d %s", w.Code, w.Body.String())
+	}
+	if w = request(mux, "GET", "/projects/thesisly/research/papers/99999"); w.Code != 404 {
+		t.Fatalf("missing HTTP status %d", w.Code)
+	}
+}
+
+func TestPostgresPublishedTextCodec(t *testing.T) {
+	legacy, ctx := integrationRepository(t)
+	r, err := NewPostgresRepository(legacy.db, "research_encoded_fixture")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err = r.Check(ctx); err != nil {
+		t.Fatal(err)
+	}
+	p, err := r.Get(ctx, 1)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if p.Title != "Graph\x00original\x01" || p.Abstract == nil || *p.Abstract != "Text\x00tail" || !reflect.DeepEqual(p.Authors, []string{"Ada\x00Smith", "Group\x01Name"}) || p.Comments == nil || *p.Comments != "Original\x00comment" {
+		t.Fatalf("published text was not restored: %+v", p)
+	}
+	mux := http.NewServeMux()
+	Register(mux, r, map[string]string{"thesisly": "http://localhost:17322/auth/callback"})
+	w := request(mux, "GET", "/projects/thesisly/research/papers/1")
+	var decoded Paper
+	if w.Code != 200 || json.Unmarshal(w.Body.Bytes(), &decoded) != nil || decoded.Title != p.Title || decoded.Abstract == nil || *decoded.Abstract != *p.Abstract {
+		t.Fatalf("HTTP lost original control characters: %d %s", w.Code, w.Body.String())
+	}
+	page, err := r.List(ctx, Filters{Page: 1, PageSize: 8, Query: "graph"})
+	if err != nil || page.Total != 2 || len(page.Items) != 2 || page.Items[1].Title != p.Title {
+		t.Fatalf("encoded corpus search: %+v %v", page, err)
+	}
+}
diff --git a/internal/research/postgres_test.go b/internal/research/postgres_test.go
new file mode 100644
index 0000000..e1bffb8
--- /dev/null
+++ b/internal/research/postgres_test.go
@@ -0,0 +1,78 @@
+package research
+
+import (
+	"database/sql"
+	"encoding/json"
+	"reflect"
+	"strings"
+	"testing"
+)
+
+func TestStoredAuthorShapes(t *testing.T) {
+	for _, test := range []struct {
+		raw  string
+		want []string
+	}{
+		{`[]`, []string{}},
+		{`[["Buhrman","Harry","","CWI"],["Family","Given","Jr.","University"]]`, []string{"Harry Buhrman", "Given Family Jr."}},
+		{`["The BABAR Collaboration","B. Aubert"]`, []string{"The BABAR Collaboration", "B. Aubert"}},
+		{`[["Balázs","C.",""],["Berger","E. L.",""],["Smith","Ada","Jr."]]`, []string{"C. Balázs", "E. L. Berger", "Ada Smith Jr."}},
+		{`[" 张三 ",["Only family"],["","Only given",""],""]`, []string{"张三", "Only family", "Only given"}},
+	} {
+		got, err := decodeAuthors(test.raw)
+		if err != nil || !reflect.DeepEqual(got, test.want) {
+			t.Fatalf("%s: %#v, %v", test.raw, got, err)
+		}
+		encoded, _ := json.Marshal(got)
+		if string(encoded) == "null" {
+			t.Fatal("missing authors must remain an array")
+		}
+	}
+	for _, raw := range []string{`broken`, `{"unexpected":"object"}`, `[42]`, `[["a","b","c","d","e"]]`} {
+		if _, err := decodeAuthors(raw); err == nil {
+			t.Fatalf("accepted unsupported authors %s", raw)
+		}
+	}
+}
+
+func TestQueryValuesRemainBound(t *testing.T) {
+	query := `graph'); DROP TABLE papers; -- OR author:*`
+	where, args := whereClause(Filters{Query: query, Topic: "cs.LG", SearchIn: "text"})
+	if strings.Contains(where, query) || !reflect.DeepEqual(args, []any{query, "cs.LG"}) {
+		t.Fatalf("unbound user input: %s %#v", where, args)
+	}
+	if !strings.Contains(where, "plainto_tsquery('english'::regconfig,$1)") || !strings.Contains(where, "t.value=$2") {
+		t.Fatalf("unexpected search contract: %s", where)
+	}
+	where, args = whereClause(Filters{Query: "Ada Smith", SearchIn: "author"})
+	if !strings.Contains(where, "p.authors_json") || strings.Contains(where, "p.abstract") || !reflect.DeepEqual(args, []any{"Ada Smith"}) {
+		t.Fatalf("author search includes unrelated text: %s", where)
+	}
+	where, args = whereClause(Filters{Topic: "cs.AI"})
+	if !strings.Contains(where, "t.value=$1") || !reflect.DeepEqual(args, []any{"cs.AI"}) {
+		t.Fatalf("topic-only binding: %s %#v", where, args)
+	}
+	where, args = whereClause(Filters{})
+	if where != "" || len(args) != 0 {
+		t.Fatal("empty filters changed the corpus")
+	}
+}
+
+func TestSchemaCannotEscapeQualification(t *testing.T) {
+	for _, name := range []string{"", `research";DROP SCHEMA public;--`, "a.b", "Public", strings.Repeat("a", 37)} {
+		if _, err := NewPostgresRepository(&sql.DB{}, name); err == nil {
+			t.Fatalf("accepted schema %q", name)
+		}
+	}
+	if _, err := NewPostgresRepository(nil, "research"); err == nil {
+		t.Fatal("accepted nil database")
+	}
+	r, err := NewPostgresRepository(&sql.DB{}, "research_v2")
+	if err != nil {
+		t.Fatal(err)
+	}
+	got := r.query(`SELECT p.id FROM {schema}papers p JOIN {schema}source_records s ON s.paper_id=p.id`)
+	if got != `SELECT p.id FROM "research_v2".papers p JOIN "research_v2".source_records s ON s.paper_id=p.id` {
+		t.Fatalf("unqualified corpus tables: %s", got)
+	}
+}
diff --git a/internal/research/testdata/corpus.sql b/internal/research/testdata/corpus.sql
new file mode 100644
index 0000000..3bbd5ac
--- /dev/null
+++ b/internal/research/testdata/corpus.sql
@@ -0,0 +1,81 @@
+-- Synthetic fixture for the disposable, network-isolated CI database only.
+-- Required columns/types follow Paper Atlas's fixed a3b677b migration.
+CREATE SCHEMA research_fixture;
+CREATE TABLE research_fixture.papers (
+ id bigint PRIMARY KEY, title text NOT NULL, abstract text, title_key text COLLATE "C" NOT NULL,
+ authors_json text NOT NULL, first_published text
+);
+CREATE INDEX papers_title_key ON research_fixture.papers(title_key);
+CREATE INDEX papers_search_gin ON research_fixture.papers USING gin
+ (to_tsvector('english'::regconfig,coalesce(title,'') || ' ' || coalesce(abstract,'')));
+CREATE TABLE research_fixture.sources (id bigint PRIMARY KEY,name text NOT NULL);
+CREATE TABLE research_fixture.source_records (
+ id bigint PRIMARY KEY,source_id bigint REFERENCES research_fixture.sources(id),
+ paper_id bigint REFERENCES research_fixture.papers(id),external_id text,url text,metadata_json text
+);
+CREATE TABLE research_fixture.identifiers (scheme text,value text,paper_id bigint REFERENCES research_fixture.papers(id),PRIMARY KEY(scheme,value));
+CREATE TABLE research_fixture.tags (id bigint PRIMARY KEY,namespace text,value text,UNIQUE(namespace,value));
+CREATE TABLE research_fixture.paper_tags (
+ paper_id bigint REFERENCES research_fixture.papers(id),tag_id bigint REFERENCES research_fixture.tags(id),
+ evidence_key text,PRIMARY KEY(paper_id,tag_id,evidence_key)
+);
+CREATE TABLE research_fixture.venues (id bigint PRIMARY KEY,acronym text);
+CREATE TABLE research_fixture.editions (id bigint PRIMARY KEY,venue_id bigint REFERENCES research_fixture.venues(id),year bigint,track text);
+CREATE TABLE research_fixture.submissions (
+ id bigint PRIMARY KEY,paper_id bigint REFERENCES research_fixture.papers(id),edition_id bigint REFERENCES research_fixture.editions(id)
+);
+CREATE TABLE research_fixture.decisions (
+ id bigint PRIMARY KEY,submission_id bigint REFERENCES research_fixture.submissions(id),
+ status text,evidence_url text,observed_at text,decision_date text
+);
+CREATE INDEX decisions_submission_order ON research_fixture.decisions(submission_id,observed_at DESC,id DESC);
+
+INSERT INTO research_fixture.papers VALUES
+ (1,'Graph & learning <original>','Neural graph models preserve original abstracts.','graph learning','[["Smith","Ada",""],"The Research Group"]','2024-01-02'),
+ (2,'Alpha topology',NULL,'alpha topology','["Bruno Park"]',NULL),
+ (3,'Graph systems','Graph systems and data.','graph systems','[["Network","Carla","","Fixture University"]]','2023-01-01'),
+ (4,'Zebra methods','Evidence and uncertainty.','zebra methods','[]','2022-01-01'),
+ (5,'Alpha Topology','Another independent record.','alpha topology','["Dana Other"]','2025-01-01');
+INSERT INTO research_fixture.sources VALUES(10,'arxiv'),(20,'pmlr');
+INSERT INTO research_fixture.source_records VALUES
+ (101,10,1,'fixture-1','https://example.test/papers/1','{"journal-ref":"Journal Fixture","comments":"Original comment"}'),
+ (202,20,3,'fixture-3','https://example.test/papers/3','{"doi":"10.fixture/other"}'),
+ (203,10,3,'fixture-3-arxiv','https://example.test/papers/3/arxiv','{}');
+INSERT INTO research_fixture.identifiers VALUES('doi','10.fixture/canonical',1);
+INSERT INTO research_fixture.tags VALUES(1,'arxiv:category','cs.LG'),(2,'arxiv:category','cs.AI'),(3,'arxiv:category','cs.DC'),(4,'different:namespace','cs.LG');
+INSERT INTO research_fixture.paper_tags VALUES(1,1,'first'),(1,1,'second'),(1,2,'first'),(3,3,'first'),(2,4,'first');
+INSERT INTO research_fixture.venues VALUES(1,'ICLR'),(2,'ICML');
+INSERT INTO research_fixture.editions VALUES(7,1,2024,'main'),(8,2,2023,'main');
+INSERT INTO research_fixture.submissions VALUES(11,1,7),(12,3,8);
+INSERT INTO research_fixture.decisions VALUES
+ (1,11,'accepted','https://example.test/decision/1','2024-02-01','2024-01-30'),
+ (2,11,'rejected','https://example.test/decision/2','2024-02-01','2024-01-31'),
+ (3,11,'withdrawn','https://example.test/decision/3','2024-02-01','2024-01-31'),
+ (4,11,'accepted','https://example.test/decision/4','2024-01-01','2099-01-01');
+
+CREATE ROLE foundry_research_fixture LOGIN;
+REVOKE ALL ON SCHEMA research_fixture FROM PUBLIC;
+GRANT USAGE ON SCHEMA research_fixture TO foundry_research_fixture;
+GRANT SELECT ON ALL TABLES IN SCHEMA research_fixture TO foundry_research_fixture;
+
+-- Separate encoded schema mirrors the published Paper Atlas nul-soh-v1 format.
+CREATE SCHEMA research_encoded_fixture;
+COMMENT ON SCHEMA research_encoded_fixture IS 'paper-atlas:text-codec=nul-soh-v1';
+DO $$
+DECLARE item text;
+BEGIN
+ FOREACH item IN ARRAY ARRAY['papers','sources','source_records','identifiers','tags','paper_tags','venues','editions','submissions','decisions'] LOOP
+  EXECUTE format('CREATE TABLE research_encoded_fixture.%I (LIKE research_fixture.%I INCLUDING ALL)',item,item);
+  EXECUTE format('INSERT INTO research_encoded_fixture.%I SELECT * FROM research_fixture.%I',item,item);
+ END LOOP;
+END $$;
+UPDATE research_encoded_fixture.papers
+SET title='Graph'||chr(1)||chr(1)||'original'||chr(1)||chr(2),
+    abstract='Text'||chr(1)||chr(1)||'tail',
+    authors_json='["Ada\u0000Smith","Group\u0001Name"]'
+WHERE id=1;
+UPDATE research_encoded_fixture.source_records
+SET metadata_json='{"comments":"Original\u0000comment"}' WHERE id=101;
+REVOKE ALL ON SCHEMA research_encoded_fixture FROM PUBLIC;
+GRANT USAGE ON SCHEMA research_encoded_fixture TO foundry_research_fixture;
+GRANT SELECT ON ALL TABLES IN SCHEMA research_encoded_fixture TO foundry_research_fixture;
diff --git a/internal/research/text.go b/internal/research/text.go
new file mode 100644
index 0000000..abebbbb
--- /dev/null
+++ b/internal/research/text.go
@@ -0,0 +1,88 @@
+package research
+
+import (
+	"database/sql"
+	"errors"
+	"strings"
+)
+
+// Matches Paper Atlas repair 4e62710's nul-soh-v1 protocol: NUL -> SOH/SOH and literal
+// SOH -> SOH/STX. All other bytes retain their original binary ordering.
+const textCodec = "nul-soh-v1"
+
+func encodedSchema(annotation sql.NullString) (bool, error) {
+	const prefix = "paper-atlas:text-codec="
+	if !annotation.Valid || !strings.HasPrefix(annotation.String, prefix) {
+		return false, nil
+	}
+	if strings.TrimPrefix(annotation.String, prefix) != textCodec {
+		return false, errors.New("unsupported research text codec")
+	}
+	return true, nil
+}
+
+func encodeText(value string) string {
+	return strings.ReplaceAll(strings.ReplaceAll(value, "\x01", "\x01\x02"), "\x00", "\x01\x01")
+}
+
+func decodeText(value string) (string, error) {
+	if !strings.ContainsRune(value, '\x01') {
+		return value, nil
+	}
+	var result strings.Builder
+	for {
+		position := strings.IndexByte(value, 1)
+		if position < 0 {
+			result.WriteString(value)
+			return result.String(), nil
+		}
+		result.WriteString(value[:position])
+		if position+1 >= len(value) {
+			return "", errors.New("invalid research text escape")
+		}
+		switch value[position+1] {
+		case 1:
+			result.WriteByte(0)
+		case 2:
+			result.WriteByte(1)
+		default:
+			return "", errors.New("invalid research text escape")
+		}
+		value = value[position+2:]
+	}
+}
+
+type textColumn struct {
+	encoded bool
+	target  any
+}
+
+func (column textColumn) Scan(source any) error {
+	var value sql.NullString
+	if err := value.Scan(source); err != nil {
+		return err
+	}
+	if value.Valid && column.encoded {
+		var err error
+		value.String, err = decodeText(value.String)
+		if err != nil {
+			return err
+		}
+	}
+	switch target := column.target.(type) {
+	case *string:
+		if !value.Valid {
+			return errors.New("unexpected null research text")
+		}
+		*target = value.String
+	case *sql.NullString:
+		*target = value
+	default:
+		return errors.New("unsupported research text destination")
+	}
+	return nil
+}
+
+func (r *PostgresRepository) text(target any) sql.Scanner {
+	return textColumn{encoded: r.encoded, target: target}
+}
diff --git a/internal/research/text_test.go b/internal/research/text_test.go
new file mode 100644
index 0000000..a972ef2
--- /dev/null
+++ b/internal/research/text_test.go
@@ -0,0 +1,69 @@
+package research
+
+import (
+	"database/sql"
+	"strings"
+	"testing"
+)
+
+func TestCorpusTextCodecRoundTripAndOrder(t *testing.T) {
+	values := []string{"", "\x00", "\x00\x01", "\x01", "\x01\x02", "\x02", "A", "A\x00", "A\x01", "Ada 张三 Ω", "Z\x00\x01\x02尾"}
+	for _, value := range values {
+		encoded := encodeText(value)
+		decoded, err := decodeText(encoded)
+		if err != nil || decoded != value || strings.ContainsRune(encoded, 0) {
+			t.Fatalf("round trip failed for %q", value)
+		}
+		for _, other := range values {
+			if strings.Compare(value, other) != strings.Compare(encoded, encodeText(other)) {
+				t.Fatalf("binary ordering changed for %q %q", value, other)
+			}
+		}
+	}
+	for _, invalid := range []string{"\x01", "prefix\x01x", "\x01\x00"} {
+		if _, err := decodeText(invalid); err == nil {
+			t.Fatalf("accepted malformed escape %q", invalid)
+		}
+	}
+}
+
+func TestCorpusTextColumnsRespectCodecAndNulls(t *testing.T) {
+	var required string
+	if err := (textColumn{target: &required}).Scan("\x01\x01"); err != nil || required != "\x01\x01" {
+		t.Fatal("legacy text was decoded")
+	}
+	if err := (textColumn{encoded: true, target: &required}).Scan([]byte("before\x01\x01after\x01\x02")); err != nil || required != "before\x00after\x01" {
+		t.Fatalf("encoded text did not decode: %q %v", required, err)
+	}
+	var nullable sql.NullString
+	if err := (textColumn{encoded: true, target: &nullable}).Scan(nil); err != nil || nullable.Valid {
+		t.Fatal("null changed into empty text")
+	}
+	if err := (textColumn{encoded: true, target: &nullable}).Scan("\x01\x01"); err != nil || !nullable.Valid || nullable.String != "\x00" {
+		t.Fatal("nullable text was not restored")
+	}
+	if err := (textColumn{encoded: true, target: &required}).Scan(nil); err == nil {
+		t.Fatal("accepted null for required text")
+	}
+	if err := (textColumn{encoded: true, target: &required}).Scan("\x01x"); err == nil {
+		t.Fatal("accepted corrupt encoded text")
+	}
+}
+
+func TestCorpusSchemaCodecMarker(t *testing.T) {
+	for _, annotation := range []sql.NullString{{}, {String: "unrelated schema description", Valid: true}} {
+		encoded, err := encodedSchema(annotation)
+		if encoded || err != nil {
+			t.Fatal("legacy schema changed encoding")
+		}
+	}
+	encoded, err := encodedSchema(sql.NullString{String: "paper-atlas:text-codec=nul-soh-v1", Valid: true})
+	if !encoded || err != nil {
+		t.Fatal("published codec was not recognized")
+	}
+	for _, name := range []string{"unknown", "", "legacy"} {
+		if _, err := encodedSchema(sql.NullString{String: "paper-atlas:text-codec=" + name, Valid: true}); err == nil {
+			t.Fatal("unknown codec was accepted")
+		}
+	}
+}