01 — CAP-RAW

Capture every printed text token exactly once while preserving the provider response as evidence.

ACTIVE Built and verified

4 tool files copied and hash-verified against the original.

Provider ML Parallel capture 0 archived step 01 of 26

Where this output sits

Source evidence›Candidate evidence›Human decision›Audit acceptance›Release›Production

Source evidence, candidate evidence, human decision, audit acceptance, release and production are six separate things. This step produces the highlighted one and nothing further.

Scope

Sends each frozen page image to a pinned Google Document AI processor through a Cloudflare Worker, freezes the raw provider response bytes before any parsing, then extracts tokens, lines, ordinals, polygons and confidences without altering a single character.

This role is allowed to

Recognize printed text and geometry
Report uncertainty

How it works

fed by 00 INGEST

Inputs

Immutable original page pixelsPinned processor/version/region configuration

Provider ML Pinned Google Document AI OCR.

Send the frozen page image to the pinned Document AI processor
Freeze the raw provider response bytes before anything parses them
Parse tokens, lines, ordinals, polygons and confidences
Bind every parsed item back to the raw response by hash

Must pass

Raw response bytes frozen before parsing
Token strings preserved byte-for-byte
Every parsed item has raw-response lineage

passes →

Outputs

Unmodified provider responseExact UTF-8 tokens and linesOrdinals, polygons, confidence and request/response hashes

hands off to COMPLETENESS · ROLE/TITLE · TYPE · ORDER · TABLE-ASSIGN · CRITICAL

fails →

HOLD

CAP_CALL_FAILEDCAP_RAW_NOT_FROZENCAP_TOKEN_MUTATEDCAP_MISSING

Nothing continues on a failed gate. Uncertainty becomes an explicit HOLD, and no later step may read an unanswered item as an accepted one.

Non-scope — what this role does NOT own

Layout meaning
Reading order
Article boundaries
Spelling correction
Arabic normalisation

Explicitly forbidden

Normalize Arabic
Correct spelling or digits
Merge with MAP meaning
Drop low-confidence tokens

Dependencies and position

Starts: After INGEST; parallel with MAP-GEO.
Previous step: 00 INGEST / FREEZE
Next step: 02 MAP-GEO / MAP-M
Hands off to: COMPLETENESS, ROLE/TITLE, TYPE, ORDER, TABLE-ASSIGN, CRITICAL

Exact inputs

Immutable original page pixels
Pinned processor/version/region configuration

Exact outputs

Unmodified provider response
Exact UTF-8 tokens and lines
Ordinals, polygons, confidence and request/response hashes

Performer and AI/ML boundary

Performer: Pinned Google Document AI OCR.
Class: Provider ML
AI boundary: OCR recognition is allowed; generative rewriting is not.

Training information

Not trained by this project. Google Document AI processor `pretrained-ocr-v2.1-2024-08-07`, pinned by version. Its training data, accuracy characteristics and failure modes are the provider's and are not disclosed to us.

Deterministic validation and acceptance gates

Raw response bytes frozen before parsing
Token strings preserved byte-for-byte
Every parsed item has raw-response lineage

HOLD and failure behaviour

Failure codes this step may emit:

CAP_CALL_FAILED
CAP_RAW_NOT_FROZEN
CAP_TOKEN_MUTATED
CAP_MISSING

Uncertainty becomes an explicit HOLD. Omission never converts uncertainty into acceptance, and no downstream step may treat an unanswered item as an accepted one.

Downstream handoff

COMPLETENESS, ROLE/TITLE, TYPE, ORDER, TABLE-ASSIGN, CRITICAL

Active tool

Status: Active tool
Run / inspect: python tools/acquire_all_docai_cap_v33.py — calls the Cloudflare Worker marsoom-rev6-phase1-cf-trial-003, which calls Google Document AI
Input: frozen page images + issue manifest
Output: raw Document AI response JSON + request_manifest.json per page

Dependencies

Python 3.12 (NOT installed on this machine)
Cloudflare Worker marsoom-rev6-phase1-cf-trial-003
Google Document AI processor pretrained-ocr-v2.1-2024-08-07
R2 frozen-raw storage

Why this is the active version

These four tools are the CAP acquisition path actually used to produce the sealed issue-4547 capture inside the SUCCESS_001 trial (1,416 verified files, 68 pages). The v33 suffix is the highest CAP tool generation present anywhere. The remote-raw runner is the one that freezes provider bytes before parsing, which is the V8 gate.

Copied files — source receipt

File	Original SHA-256	Bytes	Copy
`run_cloudflare_docai_cap_remote_raw_v33.py`	d1f62a8932af7764…	9,115	verified
`acquire_all_docai_cap_v33.py`	1fd80249fcc53ded…	6,950	verified
`recover_docai_cap_from_frozen_r2_v33.py`	161beca28d886ff7…	4,679	verified
`dehydrate_verified_r2_raw_v33.py`	5a5a0c4b40fae594…	3,676	verified

4 file(s), all hash-verified against the original. Full detail in tool/SOURCE_RECEIPT.json. Copy-only: the historical source is never modified.

Archived versions

No superseded versions recorded for this role.

Known limitations

Arabic OCR error modes are inherited wholesale and cannot be corrected at this step by design.
Direct-pixel auditing found cases where CAP text does not match what the page actually prints — 25 of 160 audited candidates in the 2026-09-04 audit were CAP/pixel mismatches.
No independent witness currently checks CAP, because MAP-GEO has no active tool.
Recall against visible ink is unmeasured: the completeness ledger that would measure it is only partly implemented.

Security and privacy

Original page images are transmitted to Google Document AI via a Cloudflare Worker. Provider credentials live in the worker's secret store, never in this repository. The frozen raw responses are evidence and are not published.

Cost behaviour

Metered by Google per page. Calls are cached and frozen in R2 so a page is never paid for twice. No call may be made without an explicit cost reservation and owner authorisation.

Copying a tool into this repository does not authorise running it, retraining it, calling a model, or processing a new issue. No paid call may be made without the owner's explicit authorisation and a hard cost cap.

Provenance

Registry: registry/roles.json (schema marsoom.e2t.roles.v1)
Derivation: 28 legacy roles - VER - HUMAN + REV = 27 active steps
Legacy role id: CAP
Legacy source SHA-256: 6f5d5d7ed8869e45307424c2193f95ffab84abc44a6a188f5c7f98c2a48ec64d
Generated: 2026-09-04T22:13:12.433Z

Feedback and decisions

Feedback is recorded per source and never merged into an invented consensus. Where sources disagree, both positions stand and the owner decides.

Source	Events
Naser / owner Final authority. Overrides every other source.	none recorded
Codex / orchestrator Architecture and sequencing.	none recorded
Builder / designer Implementation reality and constraints.	none recorded
Independent reviewer Adversarial review of claims.	none recorded
Auditor Evidence verification against artifacts.	none recorded
Human REV reviewer Page-level truth from the review site.	none recorded

Recorded events

No feedback events recorded yet. The ledger exists and is append-only: feedback/FEEDBACK_LEDGER.jsonl.

Editing feedback is not possible: a change is a new event whose supersedes names the one it replaces, and the original stays exactly as written.

Generated from registry/roles.json by tools/build-reports.mjs. Do not hand-edit — edit the registry and rebuild.
Original page pixels are the visual authority. CAP owns the exact captured text.