O017/D120 Authentic-Delivery Wrapper
Traceable Mathematical Work
1 Authentic delivery and assessment materials
The nine units of Traceable Mathematical Work remain the sole narrative spine for O017/D120. This document does not replace or abridge them, nor does it add a tenth unit. It supplies the delivery contract needed to follow the book’s local exercises and synthetic cases with evidence of authentic work. All five assessment families below are required and cannot compensate for one another.
The accompanying calibration package uses real mathematical sources, but all learners, evaluators, seminars, review cycles, and transactions within that package are synthetic. The package demonstrates the form of good evidence; it is not evidence that a learner or community ever carried out those events, and it never awards credit.
The canonical machine-readable definitions are in o017-delivery-definitions.json, the submission schema is in o017-delivery.schema.json, and complete examples are in the calibration package.
These requirements govern learner assessment. They are not human-dependent conditions on preparing, validating, or publishing this teaching edition.
1.1 No-contact and no-transmission rules
This wrapper, its renderers and checkers, and local evaluators may prepare, check, and retain packages locally. They must not send messages, open issues, upload files, complete forms, register seminars, or contact outside people or communities without separate human authorization specific to one named transaction. Mathematical suitability, prepared status, or a passing local assessment does not authorize transmission.
A transaction authorization must specify at least the target, artifact, version to be sent, channel, responsible human sender, personal-data boundary, and validity period. A missing field means authorization_present = false. Authorization must not be inherited from another task or inferred from a general desire to contribute.
External records may assert only events supported by evidence. A local draft is prepared, not submitted. A system receipt can support acknowledged status, but not accepted. The statuses accepted, merged, and published each require a different event witness. If the external state cannot be determined, use unknown while retaining the last proven state.
1.2 Evaluator identity and privacy
An assessment package uses opaque evaluator IDs, such as O017-WRAP-EVAL-7F3K9Q, rather than names, email addresses, account names, or student numbers. Generate IDs randomly; do not derive them from hashes of names or email addresses. Where genuinely necessary, a trusted administrator keeps the mapping from IDs to real identities separately in encrypted, access-controlled storage. That mapping does not belong in the portfolio, repository, or public package.
For a two-person cycle, the administrator creates this record:
evaluator_distinctness_attestation
It states only that the two IDs represent two different humans and that conflicts of interest have been checked. The attestation does not reveal identities. Agents, models, or one person changing roles may be used for practice, but they do not satisfy the independence gate.
2 Five required assessment families
2.1 A01 — Real-source search
The learner starts with a mathematical question, not a supplied list of sources. They conduct a search without pre-seeded candidates, then freeze at least one real paper or monograph. The package must contain:
- the question and topic boundary before searching;
- queries, services used, and retrieval times;
- every candidate admitted to the comparison, not just the winner;
- assessment of primary authority, version/edition, locators, access rights, and relationships among sources;
- a stopping rule written before the final conclusion;
- a persistent identifier or authoritative URL, access date, and byte witness or version witness where available; and
- uncertainties and reasons for excluding candidates.
A pass is not determined by the number of results. The gate is traceability from question to query, query to candidates, and candidates to the exact frozen source appropriate to the claim that will be used.
2.2 A02 — Real seminar or supervised equivalent
The main pathway is to attend or deliver a real mathematics seminar. Minimum evidence comprises the event identity, date and time zone, learner’s role, timed or segmented notes, dependency map, question log, privacy-minimizing evidence of attendance or delivery, corrected final notes, and an explicit distinction between what was heard, what was inferred, and what was verified only after the event.
If live audio/video is inaccessible because of disability, bandwidth, time zone, safety, or the absence of an available event, a supervised equivalent may use a recording and official transcript/notes from a real mathematics seminar or lecture. A human supervisor must assign the segments, check the dependency map and question log, and attest that the learner did the work. This pathway is assessed with the same criteria and recorded as follows:
supervised_equivalent
It is not claimed as live attendance.
An access alternative must not depend on audio alone. Provide structured text, section or time locators, formulas in text/MathML/LaTeX, descriptions of visual information, and a way to submit questions without speaking live.
2.3 A03 — Two-person author–referee cycle
Two different humans take the author and referee roles. The package preserves:
- the frozen initial manuscript and its hash;
- the referee’s scope and conflict-of-interest statement;
- a report mapping every comment to a locator and evidence;
- the author’s response to each comment;
- a disposition table using
accepted,partially_accepted,rejected_with_reason, or unresolved; - the revision, before-and-after comparison, and new hash; and
- a list of issues still open.
One person playing both roles, two accounts belonging to the same person, or automated review without a human referee constitute practice, not evidence of a passing assessment. Style scores cannot compensate for failures of mathematical validity, independence, or revision traceability.
2.4 A04 — Real-source erratum dossier
The learner freezes a real open source at a particular edition or representation, identifies the defect locator, separates the statement’s status from a conjecture about its cause, supplies a minimal reproducer, assesses severity and confidence separately, offers the smallest possible correction, and audits downstream consequences. A second human performs an independent mathematical check and marks the parts checked, method, result, and limitations.
An already published correction can still support an audit if the learner freezes the older version, reproduces the diagnosis independently, and then uses the official correction as a comparison. The dossier must not claim a new discovery. Without a second human checker, the highest status is ready_for_review.
2.5 A05 — Community transaction or weaker local pathway
The community_transaction pathway requires specific authorization before transmission. The learner sends one concise, nonduplicative, reviewable contribution that follows the target’s rules. Passing does not require acceptance; evidence of authorized submission is sufficient, provided later status remains truthful. The ledger uses prepared, submitted, acknowledged, accepted, merged, published, rejected, withdrawn, or unknown and does not erase history.
If external action is unsafe, unavailable, or unauthorized, an independent human may assess the real-source package locally. This result is labelled externally_reviewed_local. It satisfies the weaker A05 branch, but it is not evidence of submission or community participation. The final credential must retain that label. The calibration package demonstrates this pathway and stops at prepared; no contact or transmission occurs.
3 Noncompensable gates
All nine gates below must be pass. A high numerical score on other gates cannot substitute for them.
| ID | Gate | Minimum passing condition |
|---|---|---|
O017-WRAP-G01 |
Source and version | Real, authoritative source with locators and a frozen version/representation. |
O017-WRAP-G02 |
Search provenance | Complete queries, times, candidates, stopping rule, selection, and uncertainties. |
O017-WRAP-G03 |
Seminar context | Real event or supervised equivalent with evidence and the correct pathway label. |
O017-WRAP-G04 |
Human independence | Two opaque IDs represent two different humans; attestation and conflict-of-interest check are available. |
O017-WRAP-G05 |
Mathematical validity | Claims, reproducer/proof, diagnosis, correction, and limitations are checked against the frozen version. |
O017-WRAP-G06 |
Critique–revision chain | Initial manuscript, report, response, dispositions, revision, and before-and-after evidence are traceable. |
O017-WRAP-G07 |
Status and authorization | No external status without evidence; no transmission without specific authorization. |
O017-WRAP-G08 |
Privacy and access | Personal data are minimized and equivalent access alternatives are available and actually tested. |
O017-WRAP-G09 |
Evidence manifest | Every artifact has an ID, locator, byte count, hash, rights statement, verification status, and retention owner. |
Any fail, not_evaluated, or unknown prevents an assessment pass. not_applicable is valid only for criteria that do not apply under the branch definition; it cannot be used for any of the nine gates above.
4 Results and credentials
Each of A01–A05 is assessed separately. The overall decision can only be:
complete_community_transaction: A01–A05 pass and A05 has evidence ofsubmittedor a later event;complete_externally_reviewed_local: A01–A04 pass and the local A05 branch passes with an independent human reviewer;incomplete: at least one family or noncompensable gate has not passed; orcalibration_only: a demonstration package that must not award credit.
Acceptance, merging, or publication of a contribution is not required for passing and does not automatically add to the score. A documented rejection can remain consistent with a pass when package quality, authorization, and truthful status reporting have passed. Conversely, a contribution that happens to be accepted does not compensate for failures of rights, privacy, or independence.
5 Using the machine-readable package
- Copy the example’s structure without copying its calibration claims as learner data.
- Create new stable package, submission, attempt, evaluator, and evidence IDs.
- Freeze all artifacts; record their byte counts and SHA-256 in the manifest.
- Validate JSON against the schema and CSV as UTF-8 with exact headers.
- Have evaluators determine every criterion and gate; do not compute averages before all noncompensable gates pass.
- Issue a credential only after the A05 branch label and external status match the evidence.
- Keep the event ledger append-only. A correction adds a new event; it does not overwrite history.
The calibration package provides one complete example for each family. All person references are synthetic IDs, all events are simulations, every credit_eligible value is false, and every transmission_performed value is false.
6 Primary sources for calibration
Calibration uses the following official open sources without claiming endorsement from their publishers or instructors:
- OpenStax, Calculus Volume 1, §5.3, official publisher page and CC BY-NC-SA 4.0 license: https://openstax.org/books/calculus-volume-1/pages/5-3-the-fundamental-theorem-of-calculus;
- MIT OpenCourseWare 18.100A, Lecture 22, Fall 2020, official page and CC BY-NC-SA 4.0 license: official Lecture 22 page;
- G. H. Hardy, “Addenda and Corrigenda”, Proceedings of the London Mathematical Society (1928), version-of-record DOI https://doi.org/10.1112/plms/s2-28.1.553-t.
The calibration text paraphrases metadata and carries out its own mathematical calculations; it does not copy source layouts, images, videos, or prose. When a component’s redistribution rights are not sufficiently clear, the package retains only its citation, bibliographic facts, locators, and original analysis.