refactor(results): remove public trace artifact surface

[christso]

Summary

Bead: av-9ly

AgentV result bundles no longer advertise or persist trace.json as a public result artifact. New eval, export, combine, validation, metrics, and docs paths now use the transcript, metrics, result, answer, grading, timing, and external_trace surfaces as the supported inspection contract, while the dashboard still resolves transcripts from transcript_path and renders per-case results normally.

The remaining trace-envelope code is kept where it still projects transcript/result read models or handles legacy read-only inputs. Old trace_path data is rejected from new/public indexes and stripped during combine so copied legacy bundles do not re-publish the removed surface.

Verification

bun test apps/cli/test/commands/eval/artifact-writer.test.ts apps/cli/test/commands/results/combine.test.ts apps/cli/test/commands/results/export.test.ts apps/cli/test/commands/results/validate.test.ts apps/cli/test/commands/results/remote-auto-export.test.ts apps/cli/src/commands/results/serve-file-tree.test.ts packages/core/test/evaluation/results-repo.test.ts
# 166 pass, 0 fail

bun --filter @agentv/core build
bun --filter @agentv/sdk build
bun --filter agentv typecheck

bunx biome check .agents/conventions.md AGENTS.md ROADMAP.md apps/cli/src/commands/eval/artifact-writer.ts apps/cli/src/commands/results/combine-run.ts apps/cli/src/commands/results/projection-bundle.ts apps/cli/src/commands/results/serve-file-tree.test.ts apps/cli/src/commands/results/serve.ts apps/cli/src/commands/results/validate.ts apps/cli/test/commands/eval/artifact-writer.test.ts apps/cli/test/commands/results/combine.test.ts apps/cli/test/commands/results/export.test.ts apps/cli/test/commands/results/remote-auto-export.test.ts apps/cli/test/commands/results/validate.test.ts apps/web/src/content/docs/docs/evaluation/running-evals.mdx apps/web/src/content/docs/docs/tools/import.mdx apps/web/src/content/docs/docs/tools/prepare.mdx apps/web/src/content/docs/docs/tools/results.mdx docs/adr/0003-keep-opik-export-as-post-run-adapter-over-agentv-result-bundles.md docs/adr/0008-normalized-transcript-artifact-contract.md packages/core/src/evaluation/metrics.ts packages/core/src/evaluation/result-artifact-contract.ts packages/core/src/evaluation/run-artifacts.ts packages/core/test/evaluation/results-repo.test.ts
# Checked 15 files. No fixes applied.

cd apps/dashboard && bun run build

bun apps/cli/src/cli.ts eval run /tmp/agentv-av-9ly-dogfood/dataset.eval.yaml --targets /tmp/agentv-av-9ly-dogfood/targets.yaml --target azure-live --grader-target azure-live --workers 1 --agent-timeout 90 --experiment av-9ly-dogfood --threshold 0
bun apps/cli/src/cli.ts results validate .agentv/results/av-9ly-dogfood/2026-06-26T08-49-03-415Z

Dashboard UAT used:

bun apps/cli/src/cli.ts dashboard --dir . --single --port 3127

Inspected with agent-browser at http://localhost:3127: run list, run detail, checks view, transcript tab, and file tree. The API returned metrics_path, transcript_path, and transcript_raw_path, with no trace fields; transcript rendering loaded from transcript.jsonl.

Dogfood

• Local OpenAI-compatible endpoint check returned 502 with Provided authentication token is expired. Please try signing in again.
• pi-cli via the local endpoint completed at .agentv/results/av-9ly-dogfood/pi-cli-local-2026-06-26T08-46-50-011Z but scored 0%; the bundle had transcript/metrics surfaces and no trace.json.
• codex-sdk via the local endpoint failed on /v1/responses with 401 Unauthorized, token_expired.
• copilot-sdk via the local endpoint failed with provider authentication HTTP 401.
• Supplemental live Azure provider plus live LLM grader passed 100% at .agentv/results/av-9ly-dogfood/2026-06-26T08-49-03-415Z; results validate passed and file/grep checks found no trace.json, trace_path, or artifact_pointers.trace.

Evidence

Private evidence is on EntityProcess/agentv-private branch evidence/av-9ly-remove-public-trace-artifact, commit 01dca48, path av-9ly/remove-public-trace-artifact/.

Key artifact: screenshots/dashboard-transcript-tab.png shows the dashboard transcript view populated from the new supported transcript surface.

Risks

• Legacy trace-session serving is retained as read-only compatibility for old materialized inputs, but new bundles do not write trace_path or artifact_pointers.trace.
• prepared_attempt.trace_path remains intentionally because it describes prepared trace input metadata, not public result artifact discovery.

---

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	8b15647
Status:	✅  Deploy successful!
Preview URL:	https://86d84167.agentv.pages.dev
Branch Preview URL:	https://refactor-av-9ly-remove-publi.agentv.pages.dev

View logs

[christso]

Review passed on head 8b15647e7acd19b43239efc3f518a84dce668c7d.

What I verified:

• Reviewed origin/main...HEAD for the public result artifact contract: new generated/indexed/exported bundles no longer expose trace.json, trace_path, or artifact_pointers.trace as public surfaces.
• Confirmed transcript and Dashboard paths still resolve from transcript_path; the prior stale trace-session missing-state expectation now checks for transcript.jsonl guidance instead of trace.json.
• Confirmed metrics.json remains useful without source_artifacts.trace_path; it keeps trace identity metadata plus transcript_path, grading_path, and timing_path.
• Rechecked combine/export/remote publishing. The earlier remote-publish risk is fixed: sidecar publishing is transcript-only, unsupported pointer families are stripped from the published index, and trace.json is skipped from the published results tree.
• Ran a targeted legacy fixture locally: a source run containing both artifact_pointers.trace and artifact_pointers.transcript publishes only the transcript pointer, writes no trace.json to the results branch, and writes only transcript.jsonl to agentv/artifacts/v1.
• Inspected private evidence branch EntityProcess/agentv-private:evidence/av-9ly-remove-public-trace-artifact at 01dca48045c8076eacb0398d06685335a363eee6; the canonical live bundle and provider-blocker bundles contain transcript/raw transcript/metrics/result/grading/timing artifacts and no trace.json, trace_path, or artifact_pointers.trace matches. Dashboard screenshots are present.

Validation run:

• bun --filter @agentv/core build
• bun test apps/cli/test/commands/eval/artifact-writer.test.ts apps/cli/test/commands/results/combine.test.ts apps/cli/test/commands/results/export.test.ts apps/cli/test/commands/results/validate.test.ts apps/cli/test/commands/results/remote-auto-export.test.ts apps/cli/src/commands/results/serve-file-tree.test.ts packages/core/test/evaluation/results-repo.test.ts -> 166 pass, 0 fail.
• gh pr checks 1526 -> Build, Typecheck, Lint, Test, Check Links, Validate Marketplace, Validate Evals, and Cloudflare Pages all pass.

Residual risk: legacy trace-session reader support intentionally remains isolated for old artifacts, so any future cleanup should avoid re-promoting that path into the public result artifact contract.