@firedrill-tools/unstructured 0.1.4

package/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Reload Tech Inc.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/README.md

@@ -0,0 +1,275 @@
+# @firedrill-tools/unstructured
+
+A synthetic **Unstructured** account for [Firedrill](https://firedrill.run) drills. It stands in for the two public
+Unstructured HTTP surfaces so an agent that partitions documents or drives ingestion workflows can be tested locally,
+deterministically and offline:
+
+- the **Partition Endpoint** — `POST /general/v0/general` with a multipart upload, answering the familiar JSON array of
+  elements (`type`, `element_id`, `text`, `metadata`), optionally chunked;
+- the **Workflow Endpoint** subset — source connectors, destination connectors, workflows made of partition / chunk /
+  embed / prompter nodes, and jobs that run a workflow over the files a source holds.
+
+Everything is synthetic. Partitioning is a **deterministic, rule-based partition of UTF-8 text documents** (plain text,
+Markdown, HTML, CSV/TSV, RFC 822 e-mail, XML and pre-partitioned Unstructured JSON). No model, OCR or layout detection
+runs, no connector reaches a real bucket or vector store, and no Unstructured service is ever contacted. Firedrill owns
+the state, the virtual clock and the evidence. Backend only: there is no browser app.
+
+## Install
+
+```sh
+firedrill tool add @firedrill-tools/unstructured --install
+firedrill serve --scenario baseline --no-open
+```
+
+`firedrill tool add` copies the starter rows into a new world (35 rows, virtual time 2026-09-16T09:00:00Z) and grants
+the default actor every operation. Existing worlds receive neither data nor grants: add the 25 `unstructured.<operation>`
+grants listed below and the rows from `starter.json` yourself. Point a client at the world's HTTP origin with the
+`unstructured-api-key` header set to `FIREDRILL_HTTP_TOKEN` (the declared connection recipe `unstructured-api` maps
+`UNSTRUCTURED_API_URL` and `UNSTRUCTURED_API_KEY`). The real product serves the partition and platform APIs from two
+different hosts; here both are routes of the world's single origin.
+
+## Identities
+
+An API key belongs to one organisation workspace and sees every connector, workflow and job of that workspace. The actor
+attribute `unstructuredWorkspaceId` names the workspace row (`workspaces` namespace):
+
+| actor attribute | behaviour |
+|---|---|
+| absent (a fresh `firedrill tool add` actor) | falls back to the seeded default workspace `ws_northgate`; the fresh actor sees all seeded data |
+| `unstructuredWorkspaceId: "ws_northgate"` / `"ws_archive"` | scoped to that workspace; records of the other workspace behave as if they did not exist (404 on get, absent from lists, 422 `Source connector not found` when referenced) |
+| any other value | `401 {"detail": "API key is invalid"}` on every operation |
+
+Framework grants decide whether an operation may be called at all; a missing grant is rendered as
+`403 {"detail": "This API key is not granted this operation in the Firedrill world."}`. There is no read-only key level in
+the real product, so none is modelled.
+
+## Starting data
+
+The fictional **Northgate Research** account (`ws_northgate`) plus a second workspace used only for scoping tests:
+
+- **Sources**: `Policy archive` (s3, key `policy-archive`, 7 files: `handbook.md`, `refund-policy.txt`, `vendors.csv`,
+  `release-notes.html`, `elements.json`, `empty.txt`, `ledger.txt`), `Field notes` (google_drive, 3 files:
+  `2026-09-08-site-visit.txt`, `inbound.eml`, `glossary.md` with non-ASCII text and a 2,400-character paragraph),
+  `Legacy scans` (azure, 2 files that always fail: an encrypted PDF and a PNG), and `Archive dump` in `ws_archive`.
+- **Destinations**: `Parsed output bucket` (s3), `Policy index` (pinecone), and one s3 bucket in `ws_archive`.
+- **Workflows**: `policy-ingest` (custom: partition/fast → chunk_by_title 800 chars → embed, daily), `notes-basic` (auto),
+  `legacy-scan` (custom, **inactive**), and `archive-flow` in `ws_archive`.
+- **Jobs** (statuses derived from virtual time): two `COMPLETED` runs of `policy-ingest` (the download fixtures), one
+  `STOPPED` and one `IN_PROGRESS` run of `notes-basic`, one `FAILED` run of `legacy-scan` (both files failed), and one
+  `SCHEDULED` run of `policy-ingest` created two seconds before virtual now.
+- Two connection checks (`SUCCESS` for the policy archive, `FAILURE` for the legacy scans), `meta/counters` and
+  `meta/limits`.
+
+Connector secrets are stored as the placeholder `********`; no value in the starter data is a real credential, address or
+person. `ledger.txt` (2,800 short blocks) is deliberately large: partitioning it with chunking and `orig_elements`, or
+downloading it from a finished job, exceeds the 900 KB response budget and answers `413`.
+
+## Operations and routes
+
+All routes authenticate with the `unstructured-api-key` header (the framework compares it with the world's HTTP token).
+Platform collection routes take **no trailing slash** (`/api/v1/sources`); the OpenAPI's `/api/v1/sources/` form answers
+the framework's 404 because Firedrill route templates cannot end with `/`. No MCP aliases are declared (see
+*Compatibility*); the canonical `unstructured.<operation>` names are always exposed. The "UNS-MCP name" column is a
+reading aid only.
+
+| operation | UNS-MCP name (reading aid) | HTTP route |
+|---|---|---|
+| `general.partition` | – | `POST /general/v0/general` (multipart; JSON array or `text/csv`) |
+| `sources.list` | `list_sources` | `GET /api/v1/sources?source_type=` |
+| `sources.create` | `create_source_connector` | `POST /api/v1/sources` |
+| `sources.get` | `get_source_info` | `GET /api/v1/sources/{source_id}` |
+| `sources.update` | `update_source_connector` | `PUT /api/v1/sources/{source_id}` |
+| `sources.delete` | `delete_source_connector` | `DELETE /api/v1/sources/{source_id}` → `{}` |
+| `sources.check_connection` | – | `POST /api/v1/sources/{source_id}/connection-check` |
+| `sources.get_connection_check` | – | `GET /api/v1/sources/{source_id}/connection-check` |
+| `destinations.list` | `list_destinations` | `GET /api/v1/destinations?destination_type=` |
+| `destinations.create` | `create_destination_connector` | `POST /api/v1/destinations` |
+| `destinations.get` | `get_destination_info` | `GET /api/v1/destinations/{destination_id}` |
+| `destinations.update` | `update_destination_connector` | `PUT /api/v1/destinations/{destination_id}` |
+| `destinations.delete` | `delete_destination_connector` | `DELETE /api/v1/destinations/{destination_id}` → `{}` |
+| `workflows.list` | `list_workflows` | `GET /api/v1/workflows?source_id=&destination_id=&status=&name=&page=&page_size=&sort_by=&sort_direction=` |
+| `workflows.create` | `create_workflow` | `POST /api/v1/workflows` |
+| `workflows.get` | `get_workflow_info` | `GET /api/v1/workflows/{workflow_id}` |
+| `workflows.update` | `update_workflow` | `PUT /api/v1/workflows/{workflow_id}` |
+| `workflows.delete` | `delete_workflow` | `DELETE /api/v1/workflows/{workflow_id}` → `{}` |
+| `workflows.run` | `run_workflow` | `POST /api/v1/workflows/{workflow_id}/run` → 202 (empty body, or multipart `input_files` parts) |
+| `jobs.list` | `list_jobs` | `GET /api/v1/jobs?workflow_id=&status=&page=&page_size=` |
+| `jobs.get` | `get_job_info` | `GET /api/v1/jobs/{job_id}` |
+| `jobs.cancel` | `cancel_job` | `POST /api/v1/jobs/{job_id}/cancel` → `{ id, status, message }` |
+| `jobs.get_details` | – | `GET /api/v1/jobs/{job_id}/details` |
+| `jobs.get_failed_files` | – | `GET /api/v1/jobs/{job_id}/failed-files` |
+| `jobs.download_output` | – | `GET /api/v1/jobs/{job_id}/download?file_id=&node_id=` |
+
+### Partitioning
+
+`POST /general/v0/general` accepts `files` parts (several are consolidated into one array in part order) and the
+parameters that affect text partitioning and chunking: `content_type`, `strategy` (`auto|fast|hi_res|ocr_only|vlm`,
+validated and otherwise inert), `output_format` (`application/json` or `text/csv`), `encoding` (UTF-8/ASCII names only),
+`languages[]` (recorded in metadata whole; at most 20 entries of at most 20 characters, otherwise a 422), `include_page_breaks`, `starting_page_number`, `unique_element_ids`,
+`xml_keep_tags`, `chunking_strategy` (`basic|by_title|by_page`), `max_characters`, `new_after_n_chars`, `overlap`,
+`overlap_all`, `combine_under_n_chars`, `multipage_sections`, `include_orig_elements`. `split_pdf_*`, `ocr_languages`,
+`include_slide_notes`, `pdf_infer_table_structure`, `skip_infer_table_types`, `extract_image_block_types` and
+`coordinates` are accepted and ignored. A request `files` part sent with `Content-Transfer-Encoding: base64` is decoded before partitioning; transfer encodings *inside* an uploaded e-mail are not (see the e-mail notes below).
+
+The file type comes from the `content_type` field, then the part's `Content-Type`, then the filename extension
+(`.txt .md .html .htm .csv .tsv .json .eml .xml .rst`). Elements carry `filename`, `filetype`, `languages`,
+`page_number` (form feeds and HTML `page-break` styles advance it), `last_modified`, `parent_id` (nearest preceding
+Title of smaller depth), `category_depth` on titles and list items, `text_as_html` on tables, `link_texts`/`link_urls`,
+`emphasized_text_contents`/`emphasized_text_tags`, and for e-mail `sent_from`, `sent_to`, `cc_recipient`,
+`bcc_recipient`, `subject`, `email_message_id`. `element_id` is the first 32 hex characters of SHA-256 over filename,
+page, ordinal and text (a UUID when `unique_element_ids=true`). Chunks are `CompositeElement`, `Table` or `TableChunk`
+elements whose `orig_elements` is base64 of a real gzip stream (stored blocks) of the original elements — any gzip
+reader and the `unstructured` library's `elements_from_base64_gzipped_json` decode it.
+
+Each successful call writes one `partition-log` row and emits `partition.completed`, so drills can assert on a
+stateless endpoint.
+
+### Workflows and jobs
+
+`workflows.run` snapshots the workflow's nodes on the job and decides each file's fate immediately: a file fails when
+its seeded `error` is set or the partitioner rejects it. Job **status is derived from virtual time** at every read:
+`SCHEDULED` for 5 s after creation, `IN_PROGRESS` for 10 s per file plus 1 s per 4 KB, then `COMPLETED`
+(`processing_status` `SUCCESS` or `COMPLETED_WITH_ERRORS`) or `FAILED` when every file failed; `jobs.cancel` on a
+running job makes it `STOPPED`. Nothing happens between calls and no event is scheduled: advancing the world clock is
+what "finishes" a job. `jobs.download_output` recomputes the elements of one successful file with the snapshotted
+partition and chunk node settings (`file_id` and `node_id` of an `output_node_files` entry; the last node stands in for
+the destination). Embed and prompter nodes are validated and inert. Schedules are stored and echoed but never fire.
+
+## Compatibility
+
+`manifest.compatibility` is empty: **no official client has been exercised against this Tool.** The route shapes follow the
+public Unstructured documentation, the Platform OpenAPI document and the open-source partition server
+(`unstructured-api`), so `unstructured-client` (PyPI / npm) configured with `server_url = FIREDRILL_HTTP_URL` and
+`api_key_auth = FIREDRILL_HTTP_TOKEN` should reach `client.general.partition`, `client.sources.*`,
+`client.destinations.*`, `client.workflows.*` and `client.jobs.*` for the implemented subset, but that is untested. No
+MCP aliases are declared: the only published MCP contract with these resources (UNS-MCP) is unmaintained and documents no
+argument shapes, and the hosted Transform MCP server publishes no tool list.
+
+Error bodies use the FastAPI envelopes both endpoints use: `{"detail": "<message>"}` for 400/401/403/404/413/429/500/503
+and `{"detail": [{"loc": [...], "msg": "...", "type": "..."}]}` for 422 field errors (types `missing`, `int_parsing`,
+`bool_parsing`, `enum`, `extra_forbidden`, `value_error`). Known deviations, all framework-level:
+
+- Requests rejected **before any codec runs** keep the Firedrill envelope: a missing or wrong `unstructured-api-key`
+  (401), an unknown path or the trailing-slash collection form (404 `framework.HTTP_ROUTE_NOT_FOUND`), a wrong method on a
+  known path (405 `framework.HTTP_METHOD_NOT_ALLOWED` on the current framework; a 2026-09-16 build answered 404
+  `HTTP_ROUTE_NOT_FOUND` — either way a framework envelope, never a `{detail}`), unparsable or empty JSON on a
+  JSON route (400 `framework.HTTP_BODY_INVALID`), invalid UTF-8 or a body over 1 MiB.
+- A schema-invalid argument (for example a 600-character id) answers **400** with the `detail` array, not 422; the codec
+  builds the array but the framework fixes the status.
+- Binary uploads never reach the Tool: the framework decodes multipart bodies as UTF-8 text first, so a PDF or image
+  answers the framework's 400 `request body must be valid UTF-8`. Text files with a binary-looking name (`scan.pdf`) reach
+  the Tool and answer `400 {"detail": "application/pdf not currently supported"}`.
+- `unstructured-api-version: 0.1.0-firedrill` is a Tool-specific response header on the partition route.
+- Several wire details are assumptions recorded in `specs/unstructured/SPEC.md` until a client run settles them:
+  create routes answer 200 (not 201), run answers 202, deletes answer `{}`, cancel answers `{ id, status, message }`,
+  jobs list newest first, `Retry-After: 1` and the 429 text, the fixed CSV column set, several 422 message texts and the
+  `sources`/`destinations`/`reason` fields on workflow and job objects.
+
+## Events and faults
+
+| id | kind | when |
+|---|---|---|
+| `partition.completed` | event | every successful `general.partition` (`log_id`, filenames, filetypes, strategy, chunking strategy, counts) |
+| `job.created` | event | every successful `workflows.run` — also under `run-response-lost`, because that fault fires after commit |
+| `job.cancelled` | event | every successful `jobs.cancel` |
+| `partition-rate-limited` | fault (before) | `general.partition` → 429 `Rate limit exceeded. Please retry after a short delay.` with `Retry-After: 1`; nothing logged |
+| `partition-overloaded` | fault (before) | `general.partition` → 503 `Server is under heavy load. Please try again later.` (the open server's text) |
+| `run-response-lost` | fault (after commit) | `workflows.run` → 500 `Internal Server Error` although the job **was** created; a naive retry creates a second job, `jobs.list` shows the truth |
+
+Scenarios shipped with the conformance project: `baseline`, `rate-limited`, `overloaded`, `run-response-lost`,
+`tight-limits` (lowers `meta/limits` so every bounded scan fails `500 state exceeds the supported bound` instead of
+truncating) and `small-responses` (lowers `meta/limits.response_bytes` to 520 so every list or page over the budget
+fails `413 RESPONSE_TOO_LARGE` instead of being shortened).
+
+## Limitations
+
+- **Text documents only.** PDF, Office, image, e-book and archive formats are refused (`… not currently supported`)
+  even though the real service supports them. No OCR, layout model, VLM, table-structure inference, coordinates, image
+  extraction or language detection; `strategy`, `languages`, `vlm_*`, `coordinates` and the PDF/OCR parameters are
+  validated or accepted and otherwise inert (`vlm` still requires `vlm_model` and `vlm_model_provider`).
+- **Element classification is heuristic and Tool-specific.** Title/NarrativeText/ListItem/Address/EmailAddress rules
+  approximate the library for these formats but are not it; `element_id`s are deterministic but not the library's hash.
+  E-mail: headers end at the first empty line (CRLF, LF and bare CR line endings may be mixed; a leading mbox `From `
+  envelope line is skipped); `Subject`, `From`, `To`, `Cc` and `Bcc` decode RFC 2047 encoded words in UTF-8, US-ASCII or
+  ISO-8859-1 (other charsets stay literal), and address lists split on commas outside quoted strings, angle brackets and
+  comments; a multipart body is split on its RFC 2046
+  boundary, matched case-sensitively, and the **first text part in document order** (depth first through nested multiparts)
+  is partitioned — as HTML for `text/html`, as plain text otherwise. Attachments and later alternatives are not partitioned,
+  and `Content-Transfer-Encoding` (base64, quoted-printable) is not decoded. Malformed structure answers `422 {"detail":
+  "File is not a valid eml: …"}` naming it: a header line that is not a field before the blank line, a missing, empty,
+  over-70-character or invalid boundary, no delimiter, a delimiter with no line break after it, no close delimiter, more
+  than 1,000 parts, or multiparts nested deeper than 4 levels.
+- **Chunking**: `basic`, `by_title`, `by_page` only; `by_similarity` is refused. `orig_elements` is a valid but
+  uncompressed gzip stream (larger than the real payload). A table row longer than `max_characters` becomes its own run
+  of `TableChunk` pieces (its markup rides with the first piece); nothing is cut. Chunk output is sized per chunk
+  group before any piece of that group is built: a text of L characters splits into 1 + ⌈(L − max)/(max − overlap)⌉
+  pieces holding L + (pieces − 1)·overlap characters (arithmetic), and every piece carries a copy of the group's metadata
+  (the merged `link_urls`/`link_texts`/emphasis arrays, e-mail recipients, `orig_elements`), which is built **once** per
+  group — linear in that group's elements, its gzip included — measured once in JSON bytes and multiplied by the piece
+  count. When the running total of the request (summed across its files: chunk text + pieces × metadata + 90 bytes per
+  chunk, a lower bound of the JSON) passes the response budget the request answers `413 {"detail": "Partition output of
+  at least N bytes exceeds the 921600 byte response limit; upload fewer or smaller files, raise max_characters or lower
+  overlap"}` before a single piece of that group exists — an `overlap` of `max_characters − 1` with `overlap_all=true`
+  (accepted, as the library accepts it) would otherwise multiply the text by `max_characters`, and 5,000 links or 20,000
+  `<b>` marks in one paragraph chunked at a small `max_characters` would otherwise multiply their metadata into
+  gigabytes. The pieces that are built share the group's metadata (arrays by reference, one shallow copy per piece).
+  The work done before each check is bounded by that group's input, never by its piece count. Workflow chunk nodes are sized the same way at run time (against the default
+  921,600 bytes; the failure is recorded as the file's error) and again by `jobs.download_output`. CSV output uses a
+  fixed column set.
+- **Workflow Endpoint subset**: no `POST /api/v1/jobs`, destination connection checks, notifications, templates,
+  soft-deleted or recommender listings, `created_since`/`created_before`, `source_ids`/`destination_ids` fan-out or
+  `skip_preflight` (all answer 422 `… not supported by this Tool` or 404). Connector `config` is checked only for the
+  presence of a small per-type required-key table; secrets (`secret_access_key`, `token`, `password`, `client_secret`,
+  `private_key`, `api_key`, …) are masked as `********` in every response and a masked value on update keeps the stored
+  secret. Connector and workflow `key` values must be unique within a workspace. Hard deletes only.
+- **Jobs are simulated in virtual time** (see above); no page accounting or billing.
+- **Bounds** (a bound fails loudly; nothing is returned shortened): 2,000 connectors, 5,000 workflows and 5,000 jobs per namespace, 200 files per
+  source, 32 files per request, 900 KB of runtime files per run, 900 KB per response (`meta/limits.response_bytes`),
+  100,000 text blocks or elements per document, 100,000 characters per element or chunk (`max_characters`,
+  `new_after_n_chars`, `overlap` and `combine_under_n_chars` accept at most 100,000, else a pydantic-style 422
+  `less_than_equal`), 10,000 rows × 200 cells per CSV/TSV file, HTML table or Markdown pipe table, 256 levels of HTML or XML element
+  nesting, 512 levels of JSON nesting, 64 multipart parts per request, 10,000 header lines, 1,000 MIME parts and 4 multipart nesting
+  levels per e-mail, 20 `languages` entries of at most 20 characters (the partition parameter and a workflow partition
+  node's `settings.languages`; more is a 422 naming `languages`). Blank lines and paragraph continuation lines do not count
+  as text blocks. A partitioner bound is a `422 {"detail": …}` naming it —
+  `File has a table with more than 10000 rows`, `File has a table row with more than 200 cells`, `File is not valid
+  html: elements nested deeper than 256 levels` (`… not a valid xml: …` for XML), `File has more than 10000 rows` /
+  `… 200 columns` for CSV, `File has too many text blocks` / `File produces too many elements` — never a shortened
+  element. Element metadata (`link_texts`/`link_urls`, emphasis, `image_url`, e-mail recipients, `subject`,
+  `email_message_id`, merged chunk arrays) is never cut either: it is bounded only by the 1 MiB file cap and the
+  response budget below. Caller metadata on an `application/json` upload is returned exactly as given when it fits the
+  output schema (at most 100 keys of at most 100 characters, known keys with their type, other values scalars of at
+  most 10,000 characters or arrays of at most 200 scalars) and otherwise fails the whole upload with the 400 schema
+  message — no key, string or array is dropped from a 200. Every partitioner and chunker pass is a forward scan over the input
+  with explicit bounds: the HTML/XML tokenizer is one forward scan whose attribute and raw-text searches stop at their own
+  tag, an entity reference is looked up within the 12 code units after its `&`, the tree builder keeps an explicit stack
+  with per-name open counts, text is accumulated in arrays and joined once, Markdown inline searches are memoised forward
+  scans, the e-mail parser walks lines with an index that must advance on every iteration (a scan that did not advance
+  fails 422), chunk output is sized against the response budget per chunk group before any piece of that group is built
+  (§Chunking above), merged chunk link/emphasis arrays are copied item by item (never spread as call arguments) and fail
+  413 once they pass the budget, and no backtracking regular expression runs over caller text. While elements are built a
+  running lower bound of their encoded size (text, id, filename, metadata strings and array items, e-mail metadata) is
+  checked against the response budget whenever the elements themselves are serialised (unchunked output, or
+  `orig_elements` against three quarters of the budget), so an oversized result answers 413 before megabytes of elements
+  are encoded. **Measured bound, not a proof:** `test/hostile.mjs` (see `specs/unstructured/VERIFICATION.md` §13) runs
+  20,000 generated hostile inputs per seed — every format and the chunker, sizes up to just under the 1 MiB cap — directly
+  through the partitioner with the result JSON-encoded as the route does (every fourth case also decodes a hostile
+  multipart/form-data request envelope through both multipart route decoders, whose parser walks the body forward once), and fails on any case over 250 ms, any thrown
+  error or any undeclared error code; the worst case observed is recorded there (about 100–200 ms, reached by 1 MB inputs
+  that build the 100,000-element bound, where one SHA-256 element id per element dominates). Earlier HTTP measurements on
+  `firedrill serve` are recorded per fix round in VERIFICATION §9–§12. A namespace larger than its bound answers `500 {"detail": "state exceeds the supported bound of N <namespace>
+  rows"}`. Every response — element arrays, CSV, job downloads, the unpaginated `sources`/`destinations` arrays and each
+  `workflows`/`jobs` page — is measured in UTF-8 bytes before it is returned; one that would pass the budget answers
+  `413 {"detail": "… exceeds the 921600 byte response limit; lower page_size"}` (or `filter by source_type …` for the
+  connector lists). A page always holds exactly the requested rows or fails: it is never shortened, because a shortened
+  page would make the rows before the next page boundary unreachable.
+- Generic `/api/v1/…` and `/general/…` paths could collide with another installed Tool that claims the same paths.
+
+## Trademarks
+
+Unstructured is a trademark of its owner. The name is used only to identify the service this Tool simulates in a test
+environment; there is no affiliation with or endorsement by Unstructured Technologies, Inc. No logo is shipped.
+
+## License
+
+Apache-2.0 — see `LICENSE`.

package/firedrill/agent.target.json

@@ -0,0 +1,16 @@
+{
+  "schemaVersion": 1,
+  "target": {
+    "id": "conformance-agent",
+    "kind": "command",
+    "bindings": [
+      "http"
+    ],
+    "executable": "node",
+    "arguments": [
+      "test/conformance.mjs"
+    ],
+    "workingDirectory": ".",
+    "timeoutMs": 120000
+  }
+}

package/firedrill/baseline.scenario.json

@@ -0,0 +1,5 @@
+{
+  "schemaVersion": 1,
+  "id": "baseline",
+  "title": "Northgate Research account as seeded: 3 sources, 2 destinations, 3 workflows, 6 jobs, no faults"
+}

package/firedrill/conformance.suite.json

@@ -0,0 +1,21 @@
+{
+  "schemaVersion": 1,
+  "id": "conformance",
+  "title": "Unstructured Tool conformance: every operation, declared error, event and fault",
+  "drills": [
+    "unstructured-partition",
+    "unstructured-chunking",
+    "unstructured-partition-errors",
+    "unstructured-connectors",
+    "unstructured-workflows-jobs",
+    "unstructured-fresh-actor",
+    "unstructured-archivist",
+    "unstructured-revoked-key",
+    "unstructured-denied",
+    "unstructured-rate-limited",
+    "unstructured-overloaded",
+    "unstructured-run-lost",
+    "unstructured-tight-limits",
+    "unstructured-small-responses"
+  ]
+}

package/firedrill/overloaded.scenario.json

@@ -0,0 +1,11 @@
+{
+  "schemaVersion": 1,
+  "id": "overloaded",
+  "title": "general.partition answers 503 SERVICE_OVERLOADED with the open server's exact text; nothing is logged",
+  "faults": [
+    {
+      "packageId": "unstructured",
+      "faultId": "partition-overloaded"
+    }
+  ]
+}

package/firedrill/rate-limited.scenario.json

@@ -0,0 +1,11 @@
+{
+  "schemaVersion": 1,
+  "id": "rate-limited",
+  "title": "general.partition answers 429 RATE_LIMITED with Retry-After: 1; nothing is logged",
+  "faults": [
+    {
+      "packageId": "unstructured",
+      "faultId": "partition-rate-limited"
+    }
+  ]
+}

package/firedrill/run-response-lost.scenario.json

@@ -0,0 +1,11 @@
+{
+  "schemaVersion": 1,
+  "id": "run-response-lost",
+  "title": "workflows.run commits the job and its event, then answers 500 Internal Server Error",
+  "faults": [
+    {
+      "packageId": "unstructured",
+      "faultId": "run-response-lost"
+    }
+  ]
+}

package/firedrill/small-responses.scenario.json

@@ -0,0 +1,21 @@
+{
+  "schemaVersion": 1,
+  "id": "small-responses",
+  "title": "meta/limits.response_bytes lowered to 520 so every list or page that would pass the response budget fails RESPONSE_TOO_LARGE instead of being shortened",
+  "state": [
+    {
+      "action": "upsert",
+      "packageId": "unstructured",
+      "namespace": "meta",
+      "rowId": "limits",
+      "value": {
+        "connectors": 2000,
+        "workflows": 5000,
+        "jobs": 5000,
+        "files_per_source": 200,
+        "source_file_bytes": 262144,
+        "response_bytes": 520
+      }
+    }
+  ]
+}

package/firedrill/tight-limits.scenario.json

@@ -0,0 +1,21 @@
+{
+  "schemaVersion": 1,
+  "id": "tight-limits",
+  "title": "meta/limits lowered so every bounded scan fails INTERNAL_ERROR instead of truncating",
+  "state": [
+    {
+      "action": "upsert",
+      "packageId": "unstructured",
+      "namespace": "meta",
+      "rowId": "limits",
+      "value": {
+        "connectors": 2,
+        "workflows": 2,
+        "jobs": 2,
+        "files_per_source": 2,
+        "source_file_bytes": 262144,
+        "response_bytes": 921600
+      }
+    }
+  ]
+}