A versioned Q&A baseline is a time-bounded manifest of the official clarification publications available to a defined bidder population, plus an item-level account of how their questions, answers, attachments and states relate. It records whether each publication is cumulative, supplemental, corrective, split by scope or unresolved; preserves every observed representation; and selects the exact item set permitted for current bid work. Its strongest conclusion is complete through a stated observation time and source boundary. A recent filename, a high version number, an email notification or a larger row count cannot establish that conclusion alone.
A regional water utility is procuring telemetry upgrades at 63 treatment and pumping sites. Its portal first publishes “Clarification Log v1” on 12 May with questions 1 through 27. On 19 May, “Responses batch 2” contains questions 28 through 41. On 23 May, the document area adds “Clarification Log final” with questions 1 through 39, while the interactive Q&A view shows 42 items. The email alert merely says that the log was updated. Question 14 has different answer text in v1 and the final file. Questions 40 and 41 appear only in batch 2, and question 42 appears only in the portal view. One workstream replaces every older file with the final one. Another appends all three downloads and now has duplicates. Neither can say which answer set controls the design, price or submission plan.
Treat each portal view, download, attachment and message as an observed publication object, not as a synonym for the complete Q&A. First prove its procurement, stage, lot, audience, source, issue or observation time and exact bytes where available. Then determine the publisher’s update method from explicit wording and item evidence. Build an item-by-version matrix that distinguishes additions, unchanged carry-forwards, revised answers, omissions, withdrawals, references and unresolved identity. Only then release one current working baseline. Preserve every prior object and the reason it ceased to drive work. If the publication relationship cannot be proved, keep the affected answers and bid decisions blocked rather than selecting the tidiest file.
Direct answer
The latest Q&A object is not necessarily the complete Q&A set
Answer two questions separately. Which official object was published or observed most recently? Which exact set of questions, answers and attachments should current bid work use? A buyer can publish a cumulative log that replaces its predecessor, a supplement that adds only new answers, a corrected file that revises selected rows, or separate logs for different lots and stages. The latest timestamp identifies an event. It does not describe the publisher’s merge rule.
In the water utility case, the 23 May file is the newest named download. It contains questions 1 through 39, which is inconsistent with both the batch containing questions 40 and 41 and the portal count of 42. That evidence defeats an unqualified claim that the file is complete. It does not prove that the file is defective. Questions 40 through 42 might be supplemental, private, withdrawn, displayed under another stage or omitted by mistake. Each possibility requires a different record and action.
Release a baseline only when the official publication objects and item relationships support it through a stated time. The result may be a single cumulative file, a defined union of a base log and supplements, or an interactive view plus named attachments. It may also remain qualified, such as complete except for unresolved questions 40 through 42. Precision protects the bid better than forcing a clean answer that the sources do not support.
| Object | Observed evidence | What remains unproved |
|---|---|---|
| Clarification Log v1 | 12 May, questions 1 to 27 | Whether later files replace every row |
| Responses batch 2 | 19 May, questions 28 to 41 | Whether it is a supplement or later cumulative input |
| Clarification Log final | 23 May, questions 1 to 39 | Why questions 40 and 41 are absent and whether question 14 was corrected |
| Interactive Q&A view | 42 visible items at the checked time | Whether all items share the same scope and publication status |
| Email alert | States that the log was updated | Which object changed and what the update contained |
Boundary
Fix the procedure, audience and official locations before counting versions
Record the procurement identifier, current stage, relevant lots, invited or public bidder population, authorized account and observation time. A public notice attachment and an authenticated portal thread may both be official but serve different audiences. The European Commission Funding and Tenders portal documentation, for example, distinguishes public answers from private answers visible only to the economic operator that asked. A public export cannot prove completeness for a private participant record, and private content must not be moved into a public baseline.
List every location named by the tender instructions: interactive Q&A, downloadable schedules, document register, change notices and participant messages. Include a location even when it is inaccessible, empty or returns an error. “Not checked” and “checked, no item visible” are different facts. Notifications belong in the chronology but do not replace the objects they announce. Record redirects and alternate labels because a buyer may move a log without changing its business name.
The baseline boundary is not a claim about the whole internet. It states which buyer-designated sources were checked with which access and when. UK Procurement Act duties, EU rules, German VgV provisions and French profile-buyer rules address official documents, communication and timing in their respective scope. They do not guarantee that every portal presents versions in the same way. Use the procedure’s actual instructions and applicable regime, not a platform habit learned elsewhere.
Evidence
Preserve the publication before trying to merge its answers
Capture the source URL or portal locator, visible title, publisher, issue label, publication time, observation time, file name, media type, byte size and a cryptographic fingerprint where the file can lawfully be retained. A fingerprint proves whether two captured byte sequences are identical. It does not prove that their meaning, authority or completeness is the same. A portal may regenerate an unchanged spreadsheet with different internal metadata, while a same-named PDF can contain different text.
Keep the issued original separate from OCR, extracted rows, normalized text and analyst notes. Derived representations make comparison practical, but the buyer’s object remains the evidence. Preserve the extraction method and link each parsed item back to a page, cell, row or portal item. If a scan is unreadable, mark that region unreadable rather than silently omitting it from the controlled set.
Do not overwrite a prior download when a new object has the same filename. Store each observation as a separate source event and add a relationship only after review. The older file may explain why a draft, price or approval changed. The current file may later disappear. Retaining both is not permission to use both for live drafting. Storage status and working-baseline status are separate controls.
| Field | Purpose | Unsafe shortcut |
|---|---|---|
| Procedure, stage, lot, audience | Prevents cross-scope merging | Treating all buyer Q&A as one list |
| Official locator and access context | Shows where and how the object was observed | Citing a local filename alone |
| Issue, publication and observation times | Separates buyer chronology from bidder discovery | Sorting only by download time |
| Label, media type, size and fingerprint | Supports representation identity | Assuming same name means same bytes |
| Original and derived representation links | Makes extraction inspectable | Replacing the source with normalized rows |
| Declared update relationship | Records publisher intent where stated | Inventing replacement from chronology |
Version semantics
Determine whether the buyer replaced, added, corrected or split the log
Start with explicit publisher evidence. Phrases such as “supersedes version 2,” “questions 31 to 35 added,” “supplement,” “corrected answer 14” and “lot 3 only” describe different relationships. A revision table can be stronger than the filename. One NHS England clarification log published through Contracts Finder records version 1 for questions 1 to 27, version 2 for questions 28 to 30 and version 3 for question 31. That table supports an additive revision history inside the combined log. It does not establish the update method used by another buyer.
When the publisher is silent, compare item evidence without turning the comparison into authority. If every earlier identifier and answer appears unchanged and new identifiers are appended, cumulative replacement is a supported working hypothesis. If the new file contains only later identifiers, supplement is more plausible. If some earlier answers change or disappear, the relationship needs explicit review. Use behavior_unresolved when the evidence fits more than one model.
Labels such as final, latest, updated and consolidated deserve to be recorded, not worshipped. A final file can still be limited to one lot, omit attachments or precede a later correction. A high package version can contain a Q&A file whose own revision is older. German federal e-Vergabe document lists show Q&A files inside successive package versions, which is useful evidence of chronology but still requires inspection of the files and their scope.
Item control
Reconcile questions and answers by identity, not by row position
Give each observed item a source-specific identity that includes the official question number where present, stage, lot, publication object and locator. Match across publications using the buyer identifier first, then corroborate with question text, answer text, dates and cross-references. If numbering restarts by lot or stage, “question 14” is not globally unique. If the buyer renumbers a carried item, record the old and new identifiers as a proposed relationship until the content and context support it.
Compare question and answer separately. A spelling change in the question may leave the answer intact. A revised answer can change one quantity while the question remains byte-for-byte identical. Preserve answer versions as A14.1 and A14.2 or another clear local scheme, with the buyer’s identifiers unchanged. Record the smallest material change and its source locator. Detailed semantic effect belongs in the appropriate conflict or response-change control, but AN-119 must decide which observed answer version enters the baseline.
Omission is a state, not a deletion instruction. Use omitted_in_later_object until the buyer marks the item withdrawn, a cumulative replacement rule is proved, the item is relocated to a named source or an authorized clarification resolves it. If a supposedly cumulative file drops question 40, keeping question 40 active without qualification may be wrong; deleting it may also be wrong. Quarantine the dependent work and expose the gap.
| Item | Observed history | Baseline disposition |
|---|---|---|
| Q14 | Answer differs between v1 and final | Use neither version as silently settled; prove correction status and route impact |
| Q28 to Q39 | Present in batch 2 and final | Match identifiers and text; retain one current answer version per item |
| Q40 and Q41 | Present in batch 2, absent from final | Quarantine as unexplained omissions until update behavior is proved |
| Q42 | Visible only in interactive portal view | Capture scope and publication state before adding to the baseline |
| Referenced workbook | Link appears beside Q31 but file version is unclear | Resolve the exact attachment before closing Q31 |
Completeness
Prove coverage with more than the highest question number
Reconcile several independent counts: visible portal items, answered and unanswered states, unique controlled identifiers, rows in each download, attachments and cross-referenced objects. The highest identifier can coexist with gaps, withdrawn rows or separate numbering sequences. A file with 42 rows may contain headers, duplicates and non-question notices. Count evidence helps locate discrepancies; it does not replace item inspection.
Follow every “see answer,” “see revised schedule” and “document reissued” reference. The answer is not operationally complete if its target cannot be identified in the current document set. Record target_missing, target_ambiguous, target_superseded or scope_mismatch. AN-024 owns proof of the broader amendment chain and AN-078 owns detailed comparison between controlled document versions. This dossier keeps the Q&A baseline open until the referenced object can be linked or explicitly quarantined.
Check privacy and audience before merging. A private answer can explain why one supplier’s view has 42 items while a public export has 41. That does not authorize disclosure to the wider team or incorporation into a public register. Preserve an existence and scope record appropriate to permissions, and let the procurement’s communication rules determine who may see the content.
Release
Name the exact answer set that may feed current bid work
The release record lists each accepted question identity, answer version, attachment, lot and source publication. It also lists quarantined items and the bid objects they block. Avoid a label such as “use final log” when the operational set is actually v1 questions 1 to 27 except revised answer 14, plus batch 2 questions 28 to 41, plus portal question 42. The released manifest must say what the set contains.
Assign baseline_reviewed, active, quarantined, superseded_for_work and unresolved_relationship separately. A file can be retained and superseded for work. An item can be current but blocked from use because its attachment is missing. Approval covers the stated source boundary and checked time, not every future portal event. Make the release owner and evidence visible to technical, commercial, compliance and response teams.
Prevent leakage from superseded material. Search current drafts, calculations, assumptions, requirements and evidence notes for prior answer identifiers or quotations. A clean document repository is not enough if a spreadsheet still prices v1 answer 14. Route the actual content effect to AN-111 or another specialist dossier. AN-119 closes when the input baseline is controlled, not when every response change is complete.
| Control | Required evidence | Failure state |
|---|---|---|
| Boundary | Procedure, stage, lots, audience, sources and checked time | boundary_incomplete |
| Publication lineage | Relationship for every observed Q&A object | behavior_unresolved |
| Item set | One disposition for every observed question and answer version | item_gap |
| Referenced objects | Current attachment or explicit quarantine | target_unresolved |
| Work isolation | Superseded content excluded from live bid inputs | superseded_leak |
| Expiry | Next check and event-based reopen triggers | stale_baseline |
Escalation
Ask the buyer about the missing relationship, not which file looks nicer
First repeat the source search with cleared filters, every page, the current document register and the authorized participant view. Capture the observation. If questions 40 through 42 remain unexplained and the clarification route is open, ask whether the 23 May final file is cumulative, whether batch 2 remains applicable, what happened to questions 40 and 41, and where question 42 belongs. Cite the exact object labels and issue times. Do not attach the team’s preferred merge or disclose unnecessary design and price consequences.
If the question deadline has passed, follow the procurement-specific late-question rule and the separate post-cutoff recovery control. A visible message field does not prove that a late question is permitted. Internal reviewers may decide which work can continue across every viable baseline, whether the unresolved gap is bid-stopping and who owns any legal or commercial exposure. They cannot declare one buyer publication withdrawn.
Never infer a tender deadline extension from a new Q&A version. Applicable rules may require or prompt the buyer to reconsider time when material information or document changes arrive. Only the authoritative notice or procedure record changes the deadline. Keep the Q&A publication event, any formal document modification and any deadline decision as three linked but separate records.
Closure
Close through a time, then watch the official source again
The water utility baseline can close when the buyer or publication evidence establishes the update method, all 42 visible items receive a disposition, revised answer 14 is identified, questions 40 and 41 are either retained or formally displaced, question 42 is scoped, and every referenced attachment is linked or quarantined. The release says complete through the observation at a precise date and time. It does not say complete forever.
Set both time-based and event-based reopen triggers: the next promised answer date, any document-area change, a new participant message, a changed portal count, a replaced file fingerprint, the final clarification publication and the tender deadline review. Check even when no alert arrives. Record no_change_observed as an evidence-bearing result rather than leaving a blank row.
Preserve superseded publications, item mappings, decisions and baseline manifests under the applicable retention and access policy. A later review should be able to answer what the team saw, what it used, why it changed and which bid objects were exposed. That history supports accountability without turning obsolete answers into current instructions.
What good looks like
Useful outcomes from latest complete tender Q&A log
- The baseline names one procurement, stage, lot set, bidder population, official source boundary and checked time.
- Every observed Q&A view, file, message and attachment has its own source record and retained representation.
- The record distinguishes cumulative replacement, supplement, correction, scope split and unknown update behavior.
- Question identity does not depend on row position, file order or similar wording alone.
- Each item is classified as new, unchanged, revised, omitted, withdrawn, reinstated, referenced elsewhere or unresolved.
- Changed answer text and missing items remain visible until the buyer’s publication relationship supports a disposition.
- The released baseline lists the exact answer versions and attachments that current bid work may use.
- Superseded files remain available for chronology, rollback and explanation without leaking back into drafting.
- Completion expires at a stated trigger and never becomes a timeless claim that no later answer exists.
Operating model
How to run the work
- 01
Fix the Q&A control boundary
Record the procurement, stage, lots, bidder audience, authorized access, official Q&A locations and exact observation time before comparing anything.
- 02
Capture every publication object
Retain each portal view, export, file, attachment and relevant notification with its locator, visible label, timestamps, size and byte fingerprint where available.
- 03
Determine the update method
Use publisher statements, revision tables and item evidence to classify each object as cumulative, supplemental, corrective, scope-specific, withdrawn or unresolved.
- 04
Build stable item identities
Match questions by official identifier, stage, lot, question text, answer context and references, preserving collisions instead of deduplicating by appearance.
- 05
Reconcile items across versions
Mark additions, exact carry-forwards, answer revisions, omissions, withdrawals, reappearances and attachment changes for every item.
- 06
Resolve gaps and contradictions
Search the complete official boundary, follow cross-references and use the permitted buyer route when a material item or version relationship remains unexplained.
- 07
Release one working baseline
Name the exact current item and attachment set, quarantine unresolved rows, record approvals and prevent superseded answers from feeding live response work.
- 08
Expire and recheck the decision
Set the next official check and reopen triggers through the final-answer milestone, tender deadline and any later buyer publication event.
Evaluation
Questions that change the decision
- Which official locations and access role define the Q&A publication boundary for this bidder and stage?
- Does the buyer expressly call the new object cumulative, additional, corrected, replacement, final or limited to a named scope?
- Does a revision table describe the item ranges added in each version?
- Can an item be matched by a buyer-issued identifier, or is the proposed match based only on wording and position?
- Did the answer text, question text, scope, status, attachment or cross-reference change?
- Does omission from a later file mean withdrawal, movement to another publication or an unexplained gap?
- Do the portal count, downloaded item set, unanswered items and referenced attachments reconcile?
- Which current bid objects depend on a revised or unresolved answer?
- Who may approve the released baseline and who must decide a material unresolved publication conflict?
- What event or time makes the completeness decision stale?
Failure modes
Where teams lose control
The newest dated file is assumed to contain every earlier answer even though it is only a supplement.
A file marked final is trusted despite containing fewer items than the official interactive view.
All downloads are appended, causing revised answers to appear as two simultaneously valid instructions.
An omitted answer is treated as withdrawn without an explicit buyer act or complete source search.
Rows are matched by number even though the buyer restarted numbering for a new stage or lot.
Similar question wording is deduplicated even though the answers concern different sites or conditions.
A revised answer keeps its old attachment link, or a carried answer points to a superseded document.
An email alert proves that activity occurred but is mistaken for the updated source itself.
Old files are deleted, removing the chronology needed to explain changed prices or commitments.
A completeness decision survives beyond the next publication event and silently becomes stale.
Measurement
Measure the finished job
Measure the completed workflow, including review effort and exceptions. Output volume on its own is not evidence of a better process.
- official Q&A publication objects captured within the declared boundary
- objects with proved cumulative, supplemental, corrective or scope-specific behavior
- items matched across versions by official identifier and corroborating evidence
- answer revisions and attachment changes awaiting owner disposition
- later-version omissions without an evidenced withdrawal or relocation
- portal counts, file counts and controlled baseline counts reconciled
- live bid objects still using a superseded answer
- elapsed time from official publication to baseline release
- baseline checks completed by the next required trigger
Questions
Common questions
Is the file with the newest date always the current clarification log?
No. The newest object may be cumulative, supplemental, corrective or limited to one lot or stage. Prove its relationship to earlier publications before releasing a working set.
Should we append every Q&A download into one spreadsheet?
Only after item-level reconciliation. Blind appending can duplicate unchanged answers and leave revised answers looking simultaneously valid.
Does a file marked final replace every earlier answer?
The label is evidence but not enough by itself. Check the publisher’s statement, revision table, scope, item coverage, changed answers and referenced attachments.
What if an answer disappears from a later log?
Record an unexplained omission until an official source supports withdrawal, replacement or relocation. Block material dependent work instead of silently deleting or preserving the answer.
Can a file hash tell us which Q&A is authoritative?
No. A hash can support byte identity for a captured file. Authority, scope, update behavior and completeness require publication and procedure evidence.
Why can the portal count differ from the downloaded log?
Possible reasons include filters, unanswered or withdrawn items, private answers, separate stages, delayed exports or an error. Preserve the discrepancy and test those explanations at the official source.
Does a new Q&A version extend the tender deadline?
No bidder should infer an extension. Record any material change and applicable time rule, but use only an authoritative deadline notice as the current submission time.
When is the Q&A baseline complete?
When the declared official sources, publication relationships, item set, attachments and gaps reconcile through a stated observation time, with an approved working set and explicit reopen triggers.
Sources
Primary references
- Procurement Act 2023, section 12 on covered procurement objectives UK Legislation
- Procurement Act 2023, section 21 on tender notices and associated documents UK Legislation
- Procurement Act 2023, section 31 on modifying a procurement UK Legislation
- Procurement Act 2023, section 54 on time limits UK Legislation
- Cabinet Office guidance on modifying a competitive procurement, updated 17 August 2026 Cabinet Office
- Cabinet Office guidance on electronic communications, updated 17 August 2026 Cabinet Office
- Official Procurement Act e-learning module on clarifications and notified changes Cabinet Office
- Published tender instructions requiring suppliers to monitor a regularly updated Q&A log Contracts Finder
- NHS England clarification log with a three-version revision table NHS England via Contracts Finder
- Find a Tender notice publishing a dated clarification questions V1 attachment Find a Tender service
- Directive 2014/24/EU, especially Articles 18, 47 and 53 EUR-Lex
- European Commission Funding and Tenders portal guide to public and private Q&A European Commission
- German federal e-Vergabe list with successive Q&A file versions Beschaffungsamt des BMI
- UNCITRAL Model Law on Public Procurement, Article 15 United Nations Commission on International Trade Law
- World Bank Procurement Regulations, seventh edition, September 2025 World Bank
- WTO Agreement on Government Procurement, Articles X and XI World Trade Organization
About the author
Tony Kim
Tony writes about applied AI, dependable product engineering and the systems that turn complex response work into controlled delivery.
Ziva
Proposal software for source-grounded RFP, RFI, DDQ and questionnaire response work.
Bid, proposal, presales, security and compliance teams. Start with the workflow, constraints and evidence you already have.