Check lot dates against the approved expiry-date basis

Setup requirements

1. Exclude medicines, food safety decisions and any release recommendation; the exercise checks records only.

2. Owner explicitly selects day/month arithmetic, inclusivity and precedence; a manufacturer printed expiry can override a calculated hypothesis only under that rule.

3. Before any pilot, confirm an actual unresolved need and approve minimized inputs, processing environment, retention and recipients. This design does not authorize execution.

Inputs and access

1. Name

   lot_export.csv

   Source or owner

   Inventory owner

   Scope

   Synthetic lot IDs, recorded manufacturing/receipt/expiry dates and source references.

   Sensitivity

   Synthetic fixture only; actual commercial, personal or location data requires minimized approved handling.

   Permission needed

   Specific owner approval before any real-data pilot. No live account access, credential handling or upload is authorized.

2. Name

   supplier_date_evidence.txt

   Source or owner

   Quality owner

   Scope

   Synthetic manufacturer labels and certificates containing date text.

   Sensitivity

   Synthetic fixture only; actual commercial, personal or location data requires minimized approved handling.

   Permission needed

   Specific owner approval before any real-data pilot. No live account access, credential handling or upload is authorized.

3. Name

   expiry_rules.json

   Source or owner

   Quality owner

   Scope

   Approved rule per item, calendar conventions and evidence precedence.

   Sensitivity

   Synthetic fixture only; actual commercial, personal or location data requires minimized approved handling.

   Permission needed

   Specific owner approval before any real-data pilot. No live account access, credential handling or upload is authorized.

Human approval boundaries

1. Before a pilot, the accountable owner approves exact synthetic/minimized copies, processing environment, retention and allowed recipients.

2. Owner validates identity mappings, authority, status definitions and all unresolved evidence.

3. Any source mutation, account access, upload, external communication, commercial commitment, payment or changed permission requires separate appropriate authorization.

The problem

Receipt-based lot dates may conflict with an owner-approved manufacturing-date basis.

Suggested task

Compare supplied lot dates with manufacturer evidence under an explicitly approved date rule, without certifying safety or changing expiry.

What you could produce

A lot-date discrepancy sheet with calculated candidate dates, evidence authority and hold reasons.

Why an agent / simpler alternative

Mixed narrative and structured evidence can benefit from assisted extraction and explanation. Identity joins, interval calculations and totals require deterministic checks. A supported native report, query or spreadsheet is preferable when it answers the owner’s question reliably.

When this work comes up

Role

Inventory quality administrator

Trigger

A received lot has conflicting date evidence.

Human approval

Owner approves scope, rules and evidence interpretation. Any operational action requires separate appropriate authorization.

Finished output

A lot-date discrepancy sheet with calculated candidate dates, evidence authority and hold reasons.

Proposed evaluation scope — no pilot authorized

Scope

At most 25 non-regulated sample lots, 25 supplier excerpts and one approved rule version.

Limit origin

Editorial evaluation bound, not a verified platform limit.

Overflow action

Stop and ask for approved partitioning; no silent truncation.

Pilot authorized

No

Evidence boundary

Firsthand reports establish stated friction or concerns only. O*NET provides duty context. Official documents establish limited native features or general primitives. This workflow, safeguards, fixtures and outputs are original proposals, not source-documented executions.

Proposed steps

1. 1

   Freeze lot identities and rule versions; isolate duplicate lot IDs across manufacturers.

   Output: lot_scope.json

   Expected output: lot_scope.json

   Checkpoint: Use manufacturer plus item plus lot, not lot text alone.

   Approval gate: Owner approves exact inputs and rule scope before any pilot.

2. 2

   Extract raw date strings and their labels from supplier evidence, retaining page spans and locale.

   Output: date_evidence.csv

   Expected output: date_evidence.csv

   Checkpoint: Do not parse 03/04 until its locale is established.

   Approval gate: Owner resolves ambiguous identity, authority or language before affected rows advance.

3. 3

   Resolve permitted evidence links and label the applicable basis as manufacturing, receipt, printed expiry or unresolved.

   Output: basis_matrix.csv

   Expected output: basis_matrix.csv

   Checkpoint: Conflicting authoritative labels stay unresolved.

   Approval gate: Owner resolves ambiguous identity, authority or language before affected rows advance.

4. 4

   Compute candidate dates with deterministic calendar operations only for rows with complete approved rules.

   Output: candidate_dates.csv

   Expected output: candidate_dates.csv

   Checkpoint: Months are calendar months; never substitute a fixed number of days.

   Approval gate: Owner validates the calculation or decision rule before relying on its result.

5. 5

   Compare candidates to recorded dates and separate arithmetic differences from missing authority.

   Output: date_exceptions.csv

   Expected output: date_exceptions.csv

   Checkpoint: No row is declared safe, compliant or saleable.

   Approval gate: Owner reviews unresolved findings; no operational action is authorized.

6. 6

   Deliver review copies and an exact proposed-field diff for a qualified owner, with no import payload.

   Output: expiry_review.json

   Expected output: expiry_review.json

   Checkpoint: No expiry, stock state or source record is edited.

   Approval gate: Owner accepts the separate artifact; all live actions require separate authorization.

Original starter prompt — proposed

Origin

editorial_proposal

Execution authorized

No

Text

Using only {approved_synthetic_inputs} and {owner_rules}, Compare supplied lot dates with manufacturer evidence under an explicitly approved date rule, without certifying safety or changing expiry. Follow the six-step design. Preserve exact identities, evidence spans and uncertainty. Use inspectable calculations. Write only separate review artifacts described by the output schema. Stop for missing authority, ambiguous evidence or out-of-scope data. Do not change source records, contact anyone, access live accounts, post transactions or execute recovery actions.

Expected output schema — proposed

Primary key

1. manufacturer_id

2. item_id

3. lot_id

Columns

1. raw_date_text

2. date_label

3. date_basis

4. approved_rule_id

5. candidate_date

6. recorded_date

7. difference_days

8. authority_conflict

9. source_spans

10. owner_decision

Record status

draft_review_only

Common provenance columns

1. input_snapshot_id

2. rule_version

3. source_document_id

4. source_row_or_span

5. reviewer_decision

Unknown value policy

Unknown values remain null, not zero. Keep raw identities and evidence. Unresolved rows do not silently enter approved totals.

Synthetic output illustration — not an executed result

Label

Synthetic authored example, not observed output

Synthetic

Yes

Executed

No

Input

Manufactured

2027-01-31

Rule

one calendar month; clamp

Recorded expiry

2027-03-02

Expected

Candidate expiry

2027-02-28

Difference days

2

Release decision

Not specified.

Limits

1. Original proposed and untested design; no source documents this complete agent workflow.

2. A reported problem is not proof of prevalence, willingness to pay, harm magnitude or current demand.

3. Historical product reports require current-installation checks; a current capability document only establishes general primitives.

4. Native reports or a deterministic script may be simpler. Exports, identity mappings, context size, privacy controls and actual tool compatibility remain untested.

5. Numerical fixtures are synthetic and have not been executed. Owner review is necessary; no time saving, cost saving or revenue outcome is asserted.

Candidate platforms — compatibility untested

1. Name

   ChatGPT Work

   Required capabilities

   1. Assess this plan’s input, operation and output requirements before selecting an account.

   Capability source urls

   1. https://learn.chatgpt.com/docs/get-started-with-work

   Selection basis

   Editorial candidate only. Shared source facts are general primitives; this plan remains unexecuted.

   Workflow compatibility

   BotCurrent has not tested this workflow.

Failure modes

1. A label contains an ambiguous numeric date.

2. The month-end rule is absent.

3. A printed expiry conflicts with the calculation.

Recovery and troubleshooting

1. Signal

   A label contains an ambiguous numeric date.

   Recovery

   Keep the raw text and ask the issuer/owner to establish format.

   Resume condition

   The accountable owner supplies the missing authoritative evidence or explicit rule; otherwise retain the affected result as unresolved.

2. Signal

   The month-end rule is absent.

   Recovery

   Leave the candidate date null rather than guessing a rollover.

   Resume condition

   The accountable owner supplies the missing authoritative evidence or explicit rule; otherwise retain the affected result as unresolved.

3. Signal

   A printed expiry conflicts with the calculation.

   Recovery

   Hold both values for the owner; do not pick the later date.

   Resume condition

   The accountable owner supplies the missing authoritative evidence or explicit rule; otherwise retain the affected result as unresolved.

Unexecuted validation plan

Sample setup

Manufactured

2027-01-31

Rule

one calendar month; clamp

Recorded expiry

2027-03-02

Acceptance checks

1. For a supplied one-calendar-month clamp rule, January31,2027 yields February28,2027.

2. If the stored expiry is March2, the difference is two calendar days; no release decision follows.

3. An ambiguous 03/04/2027 remains unparsed until a locale is provided.

4. A missing manufacturing date never falls back to receipt date unless the supplied policy explicitly permits it.

5. Every conclusion traces to a source row or an explicitly labelled owner rule; repeated evaluation of identical frozen inputs must preserve identities and totals.

6. All original copies remain unchanged. Output is a draft review artifact, not an operational decision.

Stop conditions

1. Required identity, source coverage, calculation policy or owner authority is missing.

2. Unexpected real sensitive data is supplied without approved handling.

3. Requested work would require live access, upload, external communication, accounting posting, payment, permission change or irreversible action.

Executed

No

Fixture status

Designed specification only. No acceptance fixture or business workflow executed.

Costs and effort — unknown or hypothetical

Status

unknown_or_estimate

Currency

Not specified.

Items

1. Plan access, model usage, tool runtime and storage costs unknown.

2. Export preparation, evidence cleanup, review and correction effort unmeasured.

3. No native integration, paid service, new account or software installation assumed.

4. Case-specific effort is unknown: source extraction, identity repair, owner-rule preparation, fixture design and review may dominate model cost.

Caveat

No measured time saving, current demand, ROI or price claim. Compare the simpler remedy first.

Canadian adaptation — proposed

Verify local privacy and organizational record-handling requirements before any real-data pilot; preserve original currency, business calendars and time zones. No Canadian deployment or compliance claim.