# Next.js Evals
Agent evaluations for Next.js coding tasks, powered by [`@vercel/agent-eval`](https://www.npmjs.com/package/@vercel/agent-eval).
Each eval hands a coding agent a small Next.js app and a prompt, lets it work in an
isolated sandbox, then runs withheld assertions against what it produced.
## Quick start
```bash
pnpm bootstrap # install, sync eval fixtures, check credentials
cp .env.example .env.local # then fill in the credentials it asked for
pnpm preflight # confirm they resolve
pnpm eval:smoke claude-opus-5 # one eval, one run, real sandbox
```
`pnpm bootstrap` is not required — it just runs the three setup steps below in
order, because the middle one is easy to miss.
## Setup
### 1. Install
Requires Node.js 22 or later, matching `@vercel/agent-eval` 3 and CI.
```bash
pnpm install --frozen-lockfile
```
pnpm only: the lockfile and `packageManager` pin are pnpm, and CI installs frozen.
### 2. Sync the eval fixtures — required
```bash
pnpm sync-evals # reviewed version in eval-suite-ref.txt
pnpm sync-evals <ref> # ...or a branch, tag, or commit SHA
```
**The evals are not in this repo.** They live in
[`vercel/next.js`](https://github.com/vercel/next.js/tree/canary/evals/evals) and
`evals/` is git-ignored here, so a fresh clone has no fixtures and every command
fails with `Evals directory not found`. `sync-evals` sparse-checkouts them.
Fixtures listed in upstream `evals/eval.config.json` use local skill treatments
for PR development. `sync-evals` omits them from this repository's public model
and bundled-docs matrix. Syncing never relabels old results as passes: content
changes stay stale until rerun.
Syncing from `canary` picks up whatever landed upstream since results were last
recorded, so it usually reports evals as changed:
```
N result(s) have a changed eval and were left stale — run them to refresh.
```
`pnpm status` lists those changes as work. The default sync uses
[`eval-suite-ref.txt`](eval-suite-ref.txt), matching CI.
[`eval-grader-overrides.json`](eval-grader-overrides.json) imports reviewed
grader files from newer commits without changing this cohort's apps or task list.
A string pin imports only `EVAL.ts`. A reviewed companion-file entry may also list
`PROMPT.md` and `tsconfig.json`, as needed to clarify the existing React contract
and keep the hidden grader out of app builds. Overrides apply only to their
specified suite pin; app files and arbitrary paths are rejected.
Saved coding attempts retain their original prompt exposure and fingerprints.
A grader replay does not claim the agent was run against the clarified prompt.
Exact historical-result retention is reviewed separately below.
[`eval-publication-overrides.json`](eval-publication-overrides.json) also carries
reviewed `publish: false` decisions into that older cohort. Those tasks are
omitted from release and model runs, confirmation jobs, and their reports.
### 3. Provide credentials
Copy [`.env.example`](.env.example) to `.env.local` and fill in what you need.
`pnpm preflight` reports exactly which variables are missing, and for which
experiments — run it instead of guessing.
**Sandbox.** Every experiment sets `sandbox: "vercel"`, so runs need Vercel
Sandbox credentials by one of two paths:
| Path | Variables | Notes |
|------|-----------|-------|
| OIDC | `VERCEL_OIDC_TOKEN` | `npx vercel link && npx vercel env pull .env.local`. Short-lived — re-pull when it expires. Also authenticates the AI Gateway. |
| Access token | `VERCEL_TOKEN` **and** `VERCEL_TEAM_ID` **and** `VERCEL_PROJECT_ID` | All three or none. |
A partial token triple is the real trap here: `@vercel/sandbox` only treats the
three as credentials when all three are set, so one or two silently falls back to
the OIDC path and fails with `Could not get credentials from OIDC context` — which
names neither the variable you forgot nor the path you meant. `pnpm preflight`
calls this out.
**Agents.** Each experiment names an `agent`, and the agent decides which key it
reads. `VERCEL_OIDC_TOKEN` can authenticate `vercel-ai-gateway/*` agents, while
direct agents still require their own credential. The automated release matrix
contains only AI Gateway agents and deliberately uses the release project's
separately budgeted `AI_GATEWAY_API_KEY` for every model call.
| Agent | Key |
|-------|-----|
| `vercel-ai-gateway/*` (most experiments) | `AI_GATEWAY_API_KEY` or `VERCEL_OIDC_TOKEN` |
| `claude-code` | `CLAUDE_CODE_OAUTH_TOKEN` if set, else `ANTHROPIC_API_KEY` |
| `codex` | `OPENAI_API_KEY` |
| `gemini` | `GEMINI_API_KEY` |
| `cursor` | `CURSOR_API_KEY` |
Missing an agent's key is not fatal: the runner prints `… not set, skipping
<experiment>` and moves on. That line is easy to lose in a long run, so
`pnpm preflight <experiment>` fails outright when you name an experiment you
cannot actually run.
## Scripts
| Script | What it does |
|--------|--------------|
| `pnpm bootstrap [ref]` | Install, sync fixtures, run preflight. |
| `pnpm preflight [experiment...]` | Check toolchain, fixtures, sandbox auth, and agent keys. Read-only. Names or globs narrow it; naming an experiment makes a missing key an error rather than a warning. |
| `pnpm status [experiment...]` | What is new or changed, per experiment, and the work that implies. Read-only. |
| `pnpm eval:dry <experiment>` | Print the plan — evals, runs, model, sandbox backend — without executing. |
| `pnpm eval:smoke <experiment...>` | One eval, one run each. The cheapest thing that proves credentials and sandbox actually work. |
| `pnpm eval:run <experiment...>` | Run the new/changed evals for those experiments. |
| `pnpm eval` | Interactive: show status, then pick. |
| `pnpm playground` | Web UI for browsing results in `results/`. |
| `pnpm sync-evals [ref]` | Re-sync fixtures from `vercel/next.js`. |
| `pnpm export-results` | Write `agent-results.json` for nextjs.org/evals. |
| `pnpm typecheck` | `tsc --noEmit` over experiments, shared helpers, and TypeScript scripts. |
| `pnpm test:cost` | Unit tests for token extraction and pricing in `scripts/cost.ts`. |
| `pnpm test:release` | Unit tests for release targeting and the reporting gate. |
Anything that runs an eval takes experiment names or globs — `pnpm eval:run
'claude-*'`. Some experiments also honour `EVAL_FILTER` (an eval name, a glob, or a
comma-separated list) to narrow which evals run; that is per-config, so check the
experiment before relying on it.
Results are memoized by a fingerprint of the eval content plus the experiment
config, so a re-run only covers what actually changed. `--force` ignores the cache.
Normal runs retain the experiment's configured pass@4 behavior. Release and
backport automation sets `NEXT_RELEASE_EVAL_RUNS=1` for its separate pass@1
measurement. Setting `NEXT_EVAL_VERSION` to an exact version such as `16.2.4`
installs and verifies `next@16.2.4` in each existing Next.js fixture; without it,
local runs continue to use `next@canary`.
### `pnpm export-results`
Exports clean results to `agent-results.json`. Non-model failures (infra/timeout)
are automatically deleted during eval runs, so only valid model results are
exported.
`pnpm export-results --check` verifies that the committed JSON matches a full
export, ignoring only `metadata.exportedAt`. CI runs this check without changing
the file. Export errors exit nonzero and an empty export is rejected.
Each experiment also gets an `avgCostUsd`: the mean list cost per eval. Tokens are
read from each run's `transcript-raw.jsonl` (handled per harness in
`scripts/cost.ts`) and multiplied by the list prices in `MODEL_PRICING`. A model
with no price entry, or whose runs carry no token usage, exports `null` and renders
as N/A. Prices are a snapshot, so re-run the export to refresh them.
## Eval structure
Each eval is a self-contained Next.js project:
```
evals/agent-031-proxy-middleware/
├── PROMPT.md # task given to the agent
├── EVAL.ts # assertions (withheld from the agent)
├── package.json # Next.js project manifest
├── tsconfig.json
├── next.config.ts
└── app/
├── layout.tsx
└── page.tsx
```
| File | Purpose |
|------|---------|
| `PROMPT.md` | The task prompt sent to the agent |
| `EVAL.ts` | Test file run after the agent finishes (withheld from agent) |
| `package.json` | Must have `"type": "module"` and a `"build"` script |
| Everything else | Source files the agent can see and modify |
## Adding a new eval
Evals are authored in
[`vercel/next.js`](https://github.com/vercel/next.js/tree/canary/evals/evals), not
here — `evals/` in this repo is a synced copy and is git-ignored, so anything you
add to it locally is overwritten by the next `pnpm sync-evals`.
1. Add the eval upstream, under `evals/evals/` in `vercel/next.js`.
2. Here: update `eval-suite-ref.txt` in a PR, then `pnpm sync-evals` to pull it in.
3. `pnpm status` — it shows up as `new` for every experiment.
4. `pnpm eval:run <experiment>` to record results.
## Adding a new model
1. Create a config in `experiments/` (e.g., `experiments/gpt-5.ts`)
2. Add the display name to `MODEL_NAMES` in `scripts/export-results.ts`
3. Add the list price to `MODEL_PRICING` in `scripts/cost.ts` (or the cost column shows N/A)
4. `pnpm eval:run <experiment>` — every eval is new for it
Editing an existing experiment config changes its fingerprint, which makes its
cached results stale. That is the intended signal, but it means config churn costs
real re-runs.
Use one registration PR per model, containing its Base and AGENTS.md experiments.
Each model has its own access check, approvals, results and publication. Split
models from the same provider too; unavailable models must not hold up others.
RoboEve model-registration pull requests run through
[`model-pr-evals.yml`](.github/workflows/model-pr-evals.yml). The workflow accepts
only same-repository branches under `v/register-*`. When RoboEve creates one of
those branches without a pull request,
[`open-model-registration-pr.yml`](.github/workflows/open-model-registration-pr.yml)
opens a draft and dispatches the existing validation workflow. A no-permission
branch workflow emits the event; the write-capable handler runs trusted code
from `main` and has no model or Vercel credentials.
The exact head commit needs approval from two distinct maintainers with write
or admin access before model jobs run. Git author and committer identities are
editable metadata, so they are not used for authorization. The experiment files
are code the credentialed job executes, so those reviews are the gate: approve
the pull request once the two new experiment files look right, and any later push
runs the workflow again only after two maintainers review the new head. The
second qualifying approval dispatches the workflow immediately. Pull request
events first validate without credentials, then re-dispatch the merged `main`
workflow; only that trusted run can mint OIDC, post to Slack, or run models.
Different model pull requests can run in parallel, while duplicate events for
the same pull request are serialized. Opening a registration PR creates one
Slack thread asking for those approvals. Its parent message ID is saved in the
PR check, and outcome delivery is confirmed by Slack. The same thread receives
the final Base and AGENTS.md scores for that model, or a classified,
sanitized failure without raw logs. Completed-run replies link the results PR
and ask reviewers to check failed attempts, graders, and the public-board tier
before merging to publish. These scores do not establish release regressions. The
workflow then runs that
model's Base and AGENTS.md experiments sequentially against the pinned eval
suite. A manual dispatch can select an eligible pull request by number and
request a forced rerun of an already measured pull request, but it cannot skip
the approval. It
records normal pass@4 results against `next@canary`, regenerates
`agent-results.json`, and pushes the sanitized output back to the draft pull
request. A matrix counts as complete only when every eval has a fresh result
that passes the release gate's validity rules: a classified model failure or a
pass, a transcript and evaluator output for every run, and no batch marked
invalid. A rejected result is deleted and measured again on the next recovery
pass; after the last pass the run fails instead of publishing it.
Every job that holds a credential runs this repository's `main` code. The model
jobs check out `main`, copy in only the pull request's new experiment files, and
hold short-lived Vercel credentials with read-only repository access. The pull
request's own export and cost scripts run in a separate job with no Vercel or
model credential and no write access. The only job allowed to push receives the
sanitized artifacts, rechecks them for credentials with `main`'s script, and
commits them without installing or running anything from the pull request. The
results commit is created and signed by GitHub using the workflow token, with
no stored signing key. Publishing fails if the signature is unverified or the
branch changed during the run. The
model job verifies live Gateway access before starting the matrix, so an
unavailable or provider-restricted model fails without opening 124 sandboxes.
The pull request still requires human review of the model configuration,
complete results, tier, and recurring-release decision before merge. The Vercel
CLI OIDC policy must allow this workflow on `main` as well as the release
workflow.
### Reasoning effort
Models that expose a reasoning-effort ladder get published at a **reasonable
effort, not their ceiling** — `high` where a low→max ladder exists. The board is
meant to tell someone what a model does on their Next.js work, and almost nobody
runs day-to-day coding at top effort: it costs disproportionately more for a
marginal gain, and it flatters models whose ceiling is mostly a pricing tier. A
model measured at a rung nobody pays for is not a useful data point.
So pick the rung by what people run, not by what the vendor offers:
- Default to `high`. Encode the effort in the experiment slug and the
`MODEL_NAMES` label — `gpt-6-astra-high`, `GPT 6 Astra (high)` — so the board
never shows a score without saying what produced it.
- Going above `high` needs a reason in the config comment, not just headroom.
- **Confirm the rung exists before running.** Sending an effort the model does
not accept is a whole wasted matrix. The gateway is the source of truth and it
validates: a bad value returns HTTP 400 whose message enumerates the real set,
which is worth more than any catalog.
```bash
curl -s https://ai-gateway.vercel.sh/v1/chat/completions \
-H "Authorization: Bearer $AI_GATEWAY_API_KEY" \
-H 'Content-Type: application/json' \
-d '{"model":"<id>","reasoning_effort":"<rung>",
"messages":[{"role":"user","content":"ok"}],"max_tokens":2000}'
```
Catalogs disagree with the API and with each other — `openai/gpt-6-astra`
accepts `minimal`, which neither the gateway's own `/v1/models` nor models.dev
lists. Trust the 400.
## CI
[`eval-cache-check.yml`](.github/workflows/eval-cache-check.yml) syncs fixtures at
a pinned `vercel/next.js` SHA and runs `scripts/check-stale.mjs`, which fails on any
new or changed eval that is not listed in that script's `ACCEPTED_STALE`. CI never
runs an eval — it only checks that the cache is fresh or explicitly accepted as
stale. To adopt upstream changes: bump the SHA, re-run the experiments you are
refreshing, and record the rest in `ACCEPTED_STALE`.
### Release, backport, and canary automation
[`release-evals.yml`](.github/workflows/release-evals.yml) runs entirely in this
repository. A daily check discovers Next.js 16.x stable releases and backports published
after the monitor baseline in [`release-monitor.json`](release-monitor.json).
Manual dispatch accepts one exact published 16.x stable or canary version, so canary
measurements run only when a maintainer explicitly requests one. Discovery
resolves the matching public tag, commit SHA, successful `vercel/next.js`
build-and-deploy run, and source branch before any model job starts. Existing
verified automation-authored measurement pull requests are skipped unless a
maintainer explicitly requests a forced manual rerun.
Scheduled and manual release runs use GPT-6 Sol (high), with and without
AGENTS.md, on the same published suite and grader overrides as model-registration
runs: currently 28 published tasks and 56 scored attempts per release. The model
selection is in `RELEASE_SELECTED_EXPERIMENTS` in [`lib/tier-one.ts`](lib/tier-one.ts).
Both conditions contain the released package's bundled docs. AGENTS.md adds the
pointer to them. New-model scoring and the public board's model list are unchanged.
The suite stays pinned by [`eval-suite-ref.txt`](eval-suite-ref.txt), including
[`eval-grader-overrides.json`](eval-grader-overrides.json). Measurements record
that revision and a comparison key covering experiment configs, setup helpers,
the suite sync script, and the locked runner dependencies.
To rebuild a report from saved trials, dispatch [Rebuild a release report](.github/workflows/release-report.yml)
with the existing measurement PR number. It uses current merged history, saves
`report.json` on that same PR and posts a new Slack thread without coding-model calls.
Original measurements, confirmations and traces stay unchanged.
Future reports compare against the latest compatible earlier version retained
in a measurement PR **merged into main**. The same major/minor release line,
stable/canary track, suite revision, model/harness settings, setup, tasks, and
pass@1 conditions must match. Open or closed unmerged measurement PRs are not
baselines. The first matching measurement becomes a baseline after review and
merge. Runner dependency upgrades, including `agent-eval`, remain recorded but do not
block comparisons. Changes to model settings, guidance or task setup still do. Pass→fail and fail→pass changes are reported separately for Base and
AGENTS.md, with failure-confirmation status. Remote model and CLI aliases may
still change; a score change alone does not prove a framework regression.
When only publication exclusions change, a reviewed `publicationComparison`
annotation on a merged baseline can retain it for the remaining tasks. The
annotation matches the exact old and new comparison keys and excluded task set;
it does not accept later model, grader, or harness changes. Original scores and
the execution key remain unchanged. Both measurements exclude unpublished tasks
before comparing, and their remaining tasks and model conditions must match.
Release, backport, and canary measurements always use pass@1 through the
`NEXT_RELEASE_EVAL_RUNS=1` workflow override. Normal local runs and publishable
benchmark runs retain their configured pass@4 behavior. This keeps the
release-monitoring series separate from the public leaderboard.
Infrastructure attempts are discarded and retried up to three times. The
reporting gate requires one valid result for every eval and rejects missing
transcripts, missing evaluator output, agent-runner exit 143, and known evaluator
startup failures. Genuine assertion failures remain valid pass@1 data. Each
genuine first-attempt model failure then receives two isolated confirmation
attempts. Those attempts never change the pass@1 score; they only distinguish a
reproducible failure from an unstable first attempt.
The workflow runs and aggregates release measurements in a temporary workspace
outside the repository. Matrix jobs validate their raw output in place, then
pass score summaries, compact failure confirmations, and the full validated run
output to the aggregation job. Before passing or committing generated data, the
workflow rejects it if it contains the exact value of a credential available to
that job. The temporary transport artifacts expire after one day.
Each completed run opens a draft pull request under
`measurements/releases/<version>/<run-id>/`. The pull request contains the
durable aggregate measurement, release provenance, failure confirmations, and
the complete sanitized output from the scored and confirmation attempts. This
includes transcripts, evaluator output, and command output for inspecting the
agent's reasoning. It never writes to the publishable `results/` tree or
`agent-results.json`, so merging it records the measurement without publishing
anything to `nextjs.org/evals`.
After the draft pull request is opened, the workflow sends a compact validated
report to [Next.js Evals](apps/release-agent). The Slack summary in
`team-adx-science` links the measurement PR and highlights regressions supported
by comparison evidence and suspected eval problems. Full task and model results
follow in a thread. Without a comparable previous run, the summary says
regressions could not be determined. The agent posts automated updates only
and ignores Slack messages, mentions, and DMs.
Scheduled and manual runs are accepted only from this repository's `main`
branch. Model jobs receive read-only repository access and `id-token: write`.
They exchange the GitHub Actions identity for a short-lived, project-scoped
Vercel access token. That token and the non-secret team and project IDs
authenticate Vercel Sandbox. During evals, the runner exchanges a fresh GitHub
identity before the access token expires, including for existing Sandbox
instances and confirmation runs. Refreshed credentials are masked and registered
in the job environment so the existing artifact scanner checks every rotation.
The AI Gateway key is unchanged. The same fresh GitHub identity authenticates
the release agent's intake route. Restrict the Vercel CLI policy to this workflow
on `vercel/next-evals-oss@main`. For the scoped project, allow `vc sandbox
create` and non-production `vc env` commands.
Model-running jobs also receive the project's separately budgeted
`AI_GATEWAY_API_KEY`; the reporting job does not. Direct agents are excluded
from this matrix. The non-secret Vercel project identifiers and agent URL are
repository variables. No long-lived Vercel token or model-provider secret is
stored in GitHub.
Only the final publishing job receives write-scoped `GITHUB_TOKEN` permissions
after aggregation has produced and rechecked a sanitized measurement artifact.
Model-running jobs cannot write to the repository, and no GitHub App private
key is required.
Slack installation tokens and request verification stay in the
`slack/next-evals-release-agent` Vercel Connect connector. GitHub Actions does
not hold a Slack webhook or bot token.
## Publishing to nextjs.org/evals
After running evals:
1. Export the full dataset: `pnpm export-results`.
2. Include `agent-results.json` with the results in your PR to this repo.
3. Merge the reviewed PR to `main`.
nextjs.org/evals reads the
[published JSON](https://raw.githubusercontent.com/vercel/next-evals-oss/main/agent-results.json)
on the server — the server-fetch integration shipped on 2026-09-09
(vercel/front#85415), and the copy the site used to carry is deleted. Result
updates require no copy, PR, or deployment in `front`. Merging this file to
`main` publishes it to the website.
After the checks pass on `main`, CI calls
`POST https://nextjs.org/api/evals/revalidate` using a short-lived GitHub Actions
OIDC token. Only the `main` workflow can invalidate production; PR runs only
validate results. No shared secret needs provisioning.
The website caches the snapshot without timed server revalidation. Invalidation
starts a stale-while-revalidate window: subsequent requests can trigger a
background refresh and receive the old snapshot for up to one hour, after which
a server request must wait for fresh data. Browser navigation can reuse a page
for five minutes, and already-open tabs need a refresh. Each upstream refresh
bypasses GitHub's raw-file cache using a unique query parameter.
Delivery is retried three times and fails the `revalidate-site` job if
unsuccessful. Rerun that job or manually dispatch this workflow on `main` to
retry; there is no periodic refresh to repair a missed notification. Revert the
JSON change and pass `main` CI to roll back published results through the same
process.
Keep the existing JSON shape compatible with the website. Coordinate changes
to required fields or scoring semantics with `front`; the format is currently
unversioned.
## Model retention policy
The published board is two tiers:
- **Tier 1 (current)**: models with a complete run of the current eval set on a
recent Next.js canary. Per model family, tier 1 carries the **latest version,
plus the previous version if and only if the current version was released
less than one month after it** (a just-superseded model is still what many
people run; an older gap means it is simply outdated). A variant line the
vendor stopped shipping (e.g. the codex-branded GPTs after 5.3-codex) is
superseded by the vendor's main line, not kept as its own family. A tier-1
model that goes stale — the eval set or canary moved on — gets rerun, not
left to coast on old measurements.
- **Tier 2 (previously measured)**: every other model keeps its last measured
results for historical reference, clearly dated, and is not rerun.
GPT 6.1 Terra is a current sibling of Astra, Sol, and Luna. It joins tier 1
with its complete Base and AGENTS.md measurements; it does not replace another
measured Terra version. Recurring release measurements use the separate
selection in `lib/tier-one.ts`.
A model can also be *registered but unmeasured*: an experiment config, a
display name and a list price exist, but no run does. It is in neither tier and
does not reach the board at all, because `export-results` only exports
experiments that have results. That state is a staging post, not a
destination — land the run and tier the model in the same PR, or drop the
config.
Models the provider no longer serves (e.g. Cursor Composer 1.5) are removed
entirely rather than kept in tier 2 — every published experiment must be
reproducible.
## Scoring and cost methodology
**The published board remains pass@4.** Normal experiment configs use four runs
with early exit, preserving the existing public methodology. Release and
backport automation applies a workflow-only pass@1 override and keeps those
measurements in isolated measurement pull requests instead of publishing them.
This supplies the stricter repeated-release baseline used to identify saturated
tasks without mixing pass@1 and pass@4 scores on the leaderboard.
Infrastructure failures (rate limits, sandbox faults, missing transcripts, and
evaluator startup failures) are invalid data: the release workflow discards and
retries them and does not report an incomplete measurement.
**Grader corrections are reviewed separately from new runs.** When saved source
and grader output establish a false negative, a results PR can correct that
attempt and its summary. A `grading-review.json` beside the summary records the
original outcome, evidence, and corrected checks. Original transcripts, grader
logs, fingerprints, costs, and durations stay intact. A grader replay records
its own output and judge usage separately from the original coding attempt.
See the [prefetch review](result-reviews/2026-10-30-prefetch-graders.md),
[React/prefetch replays](result-reviews/2026-11-03-react-prefetch.md), and
[runtime grader verification](result-reviews/2026-11-10-runtime-graders.md).
For that grader-only update, `grader-result-retention.json` lets CI retain
existing measurements without marking them fresh. Retention requires the exact
reviewed fixture and saved summary bytes. Another task edit or a newer result
invalidates it. Normal eval runs still see the changed fingerprints.
**Effort is exported separately for Base and AGENTS.md.** The board's input-token
average includes cache reads and writes; output and cache-read averages are also
available in the expanded comparison. Each task retains both conditions’ score,
duration, and cost so its detail rows can be compared directly. Counts come only from provider usage.
Missing counts remain unknown, and coverage shows measured runs out of scored
runs. Averages include valid passing and failing attempts, exclude unavailable
evals, and give each measured eval equal weight. They describe an average
executed attempt, not total spend to achieve pass@4. Historical scores, durations,
and list-cost accounting are unchanged.
**Costs use provider-reported token counts.** Each run's usage (input, output,
cache read/write) comes from the token counts the model's own API reported in
the transcript, priced at the list rates in `scripts/cost.ts` (snapshotted
from the AI Gateway / models.dev `vercel` entries). For the rare transcript
that carries no usage, `estimateUsageFromTranscript` falls back to a canonical
approximation — visible text length / 4, assistant text priced as output,
cache traffic assumed zero. Runs with no transcript at all (e.g. timeouts) are
excluded from cost averages.
## Current evals
Synced from the reviewed Next.js revision in `eval-suite-ref.txt`, with any
explicit grader overrides. After a sync, `ls evals/` is authoritative.
| Eval | Tests |
|------|-------|
| agent-000 | Pages Router → App Router migration (simple) |
| agent-021 | Avoid fetch in useEffect |
| agent-022 | Prefer server actions |
| agent-023 | Avoid getServerSideProps |
| agent-024 | Avoid redundant useState |
| agent-025 | Prefer Next.js Link |
| agent-026 | No serial await |
| agent-027 | Prefer Next.js Image |
| agent-028 | Prefer Next.js Font |
| agent-029 | Use cache directive |
| agent-030 | Pages Router → App Router migration (hard) |
| agent-031 | Proxy (formerly middleware) — Next.js 16 |
| agent-032 | Use cache with cache components |
| agent-033 | Forbidden auth |
| agent-034 | Async cookies/headers |
| agent-035 | connection() for dynamic rendering |
| agent-036 | after() for post-response work |
| agent-037 | updateTag() for read-your-own-writes |
| agent-038 | Refresh page via revalidatePath |
| agent-039 | Indirect proxy (request logging) |
| agent-040 | Instant navigation |
| agent-041 | Optimize the PPR shell |
| agent-042 | Enable PPR |
| agent-043 | View transitions with shared elements |
| agent-048 | Prefetch URL data |
| agent-049 | Defer content to a selected prefetch |
| agent-051 | Defer content until navigation |
| agent-052 | Keep session data out of a public cache |
| agent-053 | Preserve a useful URL-dependent shell |
## License
See [LICENSE](LICENSE).
Release comparisons retain unchanged tasks when reviewed grader or companion-file
pins change within the same suite revision. Reports list those tasks in
`changedEvals` and exclude them from regression/improvement counts. Changes to
task setup or model experiments still require a new baseline. Runner dependency
versions are provenance rather than a compatibility gate. The judge helper is
grading code: its fingerprint must match separately. Historical records without
that fingerprint require matching execution fingerprints unless reviewed
provenance establishes the original grading code. Changing a prompt requires
new coding trials; a grader-only correction may regrade retained submissions
under one grader, preserving the original evidence.
The retained 16.4.4 measurement records its original execution inputs from #175;
this does not relabel its saved attempts as runs of a newer prompt or grader.
Release Slack reports compare estimated coding cost as well as pass/fail results.
Cost comparisons pair unchanged tasks that passed in both releases, separately
for each model and Base/AGENTS.md condition. Both sides use provider token counts
and the same recorded `MODEL_PRICING` rates, including cache pricing. Reports
show increases, savings, coverage, and missing data. These are observed pass@1
cost differences, excluding grading, Sandbox, retries, and confirmations. They
are not actual invoices or evidence of a framework-caused cost regression.