A question-subpart coverage record is a source-controlled tree of the independently answerable asks inside one tender question. Each leaf retains its parent, exact source and version, instruction verb, requested object, scope, condition and quantifier. It then points to the response passage, supporting material, owner and review state. The record shows whether the answer merely mentions a topic or performs the job the buyer requested.
A tender question that occupies six lines can contain twenty separate asks. Some sit in the stem, others in bullets, response guidance, a scoring descriptor, an embedded table or a later clarification. Writers often organize the answer around their own solution and cover the obvious nouns. They miss the second half of a joined instruction, apply one example to a broader scope than it supports, or describe a method where the buyer asked for quantified evidence. The draft can read well and still leave an evaluator with no assessable answer to a subpart.
Treat the issued question as the controlling structure. Reconstruct it from every document that governs the response, then split it into the smallest units that can be answered and reviewed independently. Preserve operators such as “for each”, “at least”, “during mobilisation and operation” and “where applicable”. Map every leaf to a specific response span and test that span against the original instruction verb. Finally, read in both directions: from each ask to its answer, and from every paragraph, table and attachment back to the ask it serves.
Source boundary
The question is larger than the box on the portal
A question number is an address, not the whole ask. The heading may say “Mobilisation”, while the stem asks for an approach, three bullets specify staffing, asset data and continuity, the response guidance requires two examples, and the scoring descriptor rewards quantified risk control. A later clarification may define when mobilisation ends. All of that governs the answer even if the portal shows one text field.
Freeze a question source set before drafting. Record the file name, section, page or cell, version, lot and publication date for each relevant fragment. Include the question stem, nested bullets, adjacent table labels, definitions, submission instructions, response guidance, award criterion, subcriterion, scoring scale and authorized clarifications. Keep conditions of participation separate when the procurement treats them separately.
Reconstruction prevents a common review error: comparing the draft with a shortened copy of the question in a writing plan. The shortened copy has usually lost the words that determine scope. The final reviewer should be able to open every controlling fragment from the record and see why it belongs.
| Source fragment | What to preserve | Typical loss |
|---|---|---|
| Question stem | Instruction verb, object and overall scope | The response answers the topic but not the requested action |
| Bullets or numbered limbs | Each item and its parent relationship | The final bullet disappears inside a broad heading |
| Response guidance | Required examples, periods, format and exclusions | A good narrative lacks the requested proof |
| Award criteria and descriptors | What the evaluator is permitted to assess | Space is spent on material that cannot earn credit |
| Tables and linked fields | Row-level asks, units and attachment references | The narrative and schedule cover different scopes |
| Clarifications or addenda | Replacement wording and effective version | The answer remains correct only for the withdrawn question |
Decomposition
Build a tree, not a flat checklist of nouns
Give the full question a root identifier, then preserve its nesting. A branch might cover mobilisation governance. Its leaves could ask who owns each workstream, how often the board meets and which record proves a decision. Another branch may cover continuity during the same period. Parentage matters because a weekly meeting mentioned under governance does not answer a weekly customer update requested under continuity.
The smallest useful leaf can receive an independent answer and review result. Do not create a new leaf for every grammatical verb. “Identify three mobilisation risks and explain how each will be mitigated” is one coupled task repeated three times: each named risk needs its own treatment. Splitting “identify” from “explain” would allow a risk list to look complete while every mitigation remains absent.
Retain the connectors. “During transition and steady-state operation” sets two periods. “For each depot” distributes an ask across the full population. “At least three” supplies a floor. “Including” may introduce required content, while “such as” may only illustrate it. “Where applicable” calls for a reasoned applicability decision, not silent omission.
Answer contract
Make the instruction verb testable
For each leaf, write a short answer contract. It states what a satisfactory response must do without drafting the response itself. “Explain the asset-data validation method” becomes: name the input, checks, exception route, responsible role and reason the method produces a usable baseline. The contract stops a paragraph from passing merely because it contains “asset data” and “validation”.
Different verbs create different review tests. Describe calls for a concrete account of what happens. Explain adds mechanism or reason. Demonstrate needs a supported proposition, not confidence. Provide calls for the requested item. Quantify needs a value, unit, period and basis. Compare requires shared dimensions across the alternatives. Confirm requires an unambiguous position and any requested support.
Read the noun and qualifiers with the verb. “Quantify service improvement” is incomplete unless the question or guidance defines the service, baseline, period and population. If those elements are absent from the tender, record the uncertainty and use the official clarification route where available. Do not manufacture a number to make the leaf look closed.
| Verb | The answer must do | Weak substitute |
|---|---|---|
| Describe | Give a concrete account with relevant actors, sequence and scope | A list of capability labels |
| Explain | Show how or why the stated result follows | A control name with no mechanism |
| Demonstrate | Make a proposition and support it with applicable proof | An assurance adjective |
| Provide | Supply the named information, document, example or value | A promise to provide it later |
| Quantify | State the value, unit, period, population and basis | A direction such as “substantial reduction” |
| Compare | Assess each option against common, relevant dimensions | Two unrelated descriptions |
| Confirm | State a clear position and attach the support requested | An implied yes |
| Identify and mitigate | Pair each named risk with its treatment, owner and residual position | A risk list followed by generic controls |
Traceability
Map substance, not matching words
Point a leaf to the smallest useful response span: a paragraph range, table row, diagram element or attachment section. A whole-document reference is too vague to review. Record the response version because the same heading can survive while the content beneath it changes.
One span may answer several leaves. A mobilisation table that names the workstream owner, activity, timing, dependency and evidence can support separate role, schedule and governance asks. Count that mapping only after testing the row against each answer contract. Repeating a question word is not coverage, and adding three duplicate paragraphs does not improve it.
One leaf may also need several spans. The method may sit in the narrative, the quantified commitment in a table and the past example in an annex. Link all three and state what each contributes. If the evidence proves only one region, service or period, preserve that boundary rather than extending it to the full proposed solution.
Review states
A tick cannot describe the important gaps
Use states that tell the team what remains. “Identified” means the ask exists in the tree. “Answer mapped” means a response location has been named. “Partial” means the response performs only part of the answer contract. “Unsupported” means the proposition lacks required proof. “Contradictory” means another response element takes an incompatible position. None of these states is ready for release.
“Non-applicable claimed” is also open until the right owner accepts the reason and the tender permits that treatment. “Substantively addressed” means the content performs the requested function. “Supported” means the required evidence or decision is present and within scope. “Unresolved” is preferable to a false green status when the source is ambiguous.
Do not collapse the record into one coverage percentage. Ninety-nine addressed leaves provide little comfort if the hundredth is a pass-or-fail declaration. Show counts by state, but place mandatory status, question weight and consequence beside them so the reviewer sees the gap that can change the submission decision.
Two-way review
Start once from the question and once from the answer
The forward review begins at the root question. Follow every branch to each leaf, open the mapped response span and apply the answer contract. Confirm the right scope, condition and quantity. Then inspect the attached evidence or decision. This pass finds omissions and thin answers.
The backward review begins at the draft. Ask which leaf every paragraph, figure, table row and attachment serves. Material with no defensible link may be irrelevant, left over from another bid or consuming space needed by an uncovered ask. A passage linked to the wrong branch can also reveal a structural problem that the forward pass missed.
Give the final reviewer the issued source set, not just the writer’s headings. A heading such as “Our approach” can conceal three unanswered limbs because it organizes the bidder’s story rather than the buyer’s assessment. The source-first pass restores the evaluator’s point of view.
Worked example
A 500-word mobilisation answer can contain twelve asks
A transport authority asks the bidder to describe mobilisation across 24 depots during the first 90 days. The response must cover asset-data validation, staff onboarding, service continuity and governance; identify at least three risks and explain how each will be mitigated; and provide two comparable examples. The portal allows 500 words plus one risk table.
The root splits into four operating branches, one risk branch with three coupled risk-treatment leaves, and one experience branch with two example leaves. Scope fields preserve all 24 depots and the 90-day period. The word limit belongs to the whole response, while the separate risk table changes where the risk leaves can be answered.
The first draft names an asset audit, induction and weekly board, then gives two projects. It looks complete. The record shows otherwise: the asset passage does not say how exceptions are resolved; onboarding omits temporary staff; continuity has no answer span; governance lacks decision records; the second example covers only a single depot; and one risk has no mitigation. Those are six different repair decisions, not one instruction to “add more detail”.
| Leaf | Answer contract | Mapped response | State |
|---|---|---|---|
| MOB-AD-02 | Explain the exception route for failed asset records across 24 depots | Paragraph 2 names checks but no exception route | Partial |
| MOB-ST-03 | Cover onboarding for permanent and temporary staff in the 90-day period | Paragraph 3 covers permanent staff only | Partial |
| MOB-CO-01 | Describe how service continues during transition | No response span | Identified |
| MOB-GV-03 | Name the record used for governance decisions | Weekly board stated; record absent | Partial |
| MOB-EX-02 | Provide a second example comparable in scale and distribution | Single-depot example in Annex B | Unsupported |
| MOB-RK-03 | Pair the third risk with a specific mitigation | Risk table row 3 has no treatment | Partial |
Change control
Coverage expires when its inputs change
Reopen affected leaves after an amendment, clarification, solution decision, source update, evidence expiry or material edit. A global “reviewed” date is not enough. Record the event, the leaves it can affect, the person who reassessed them and the resulting state. Preserve the earlier mapping so the team can understand what changed.
Cuts made near the deadline deserve the same control. Before deleting a sentence, table row or annex reference, inspect its backward links. After the cut, every dependent leaf returns to review. This makes word-count editing slower for a few minutes and prevents a concise answer from becoming incomplete without anyone noticing.
The record proves content coverage for one question. It does not prove legal compliance, factual truth, overall submission readiness, a particular evaluation score or contract acceptance. Those decisions require their own controls. The useful boundary is narrow: every issued ask is visible, every claimed answer is inspectable and every unresolved gap reaches the right reviewer.
What good looks like
Useful outcomes from answer every part of an RFP question
- Every independently answerable part of the question has a stable identifier and exact source location.
- Conditions, quantities, time periods, populations and other limiting words survive decomposition.
- Each response passage performs the function required by the instruction verb instead of only naming the subject.
- One passage may satisfy several subparts when its content supports each one, without duplicating prose for appearance.
- Partial, unsupported, contradictory and claimed-not-applicable items remain visible to the reviewer.
- Tables and attachments are connected to the narrative claim they support rather than treated as self-explanatory.
- Cuts made to meet a response limit reopen the affected coverage checks.
- The final reviewer can begin with the buyer’s question and account for the whole response.
Operating model
How to run the work
- 01
Freeze the controlling question set
Record the procurement, lot, question number, source files, response field, limits and latest amendments before splitting the ask.
- 02
Reconstruct the full prompt
Read the stem with its bullets, tables, notes, definitions, response guidance, criteria, scoring descriptors and linked fields.
- 03
Build the subpart tree
Create a leaf only when the item can receive its own answer and review decision, while keeping its parent and logical connections.
- 04
Write the answer contract
Translate each instruction verb and its qualifiers into a plain test of what the response must contain.
- 05
Map response and support
Point each leaf to the exact paragraph, table row or attachment and to any evidence or decision needed to sustain it.
- 06
Review in both directions
Trace every ask forward to an adequate answer, then trace every response element backward to a relevant ask.
- 07
Close or expose the gaps
Resolve partial, unsupported and contradictory leaves, or present the remaining exception to the authorized reviewer before release.
Evaluation
Questions that change the decision
- Which files and amendments together form the complete question?
- What is the main instruction verb, and are later verbs separate asks or part of the same response task?
- Which subject, phase, site, service, lot, user group or scenario limits the ask?
- Does “and”, “or”, “including”, “for each” or “where applicable” change how many leaves are needed?
- What would let a reviewer decide that this leaf has been substantively answered?
- Can one response span satisfy several leaves, or does it only repeat their vocabulary?
- Does the answer need a fact, explanation, method, comparison, quantity, example, commitment or attached item?
- Which missing support is a drafting gap, an evidence gap, a solution decision or an approved exception?
- Did a later edit or amendment change any covered leaf?
- Can each paragraph and attachment be justified against the issued question and evaluation basis?
Failure modes
Where teams lose control
The writer answers the heading but misses instructions added in the response guidance.
A joined ask is split so aggressively that the relationship between method and proof disappears.
An “including” list is treated as optional illustration when the wording requires every listed item.
A generic case study is counted as evidence for several scopes it does not cover.
The answer states that a control exists but does not explain how it operates, although the verb is “explain”.
A reference to an attachment is marked complete even though the attachment lacks the requested detail.
A word-count cut removes a qualifier, example or quantified result without reopening the leaf.
The response is polished against the author’s outline rather than checked against the original question.
A late clarification changes a subpart but the coverage record still points to the earlier version.
A coverage percentage hides one unresolved pass-or-fail item.
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.
- subparts with a source file, locator, version and parent identifier
- leaves with instruction verb, requested object, scope, condition and quantifier recorded
- leaves mapped to a precise response location rather than a whole document
- substantively addressed leaves whose response function matches the instruction verb
- supported leaves with adequate evidence or an authorized decision
- partial, unsupported, contradictory and claimed-not-applicable leaves still open
- response passages and attachments with at least one relevant backward link
- coverage checks reopened after source, solution, evidence or response changes
Questions
Common questions
Should every “and” create another subpart?
No. Create separate leaves when the items can receive different answer or review results. Keep inseparable pairs together, such as identifying each risk and explaining its mitigation, so the record cannot mark half the requested task complete.
Can one paragraph answer several subparts?
Yes. Map the paragraph to each leaf and test it separately against each answer contract. Shared prose is efficient when it carries the required substance. A cluster of matching terms is not enough.
Does mentioning a topic count as coverage?
Only when the buyer requested identification or a similarly narrow act. A mention does not explain a mechanism, demonstrate a claim, quantify a result, compare alternatives or provide an item.
How should “where applicable” be handled?
Create the leaf and record the applicability decision. If it applies, answer it. If it does not, preserve the reason, owner and approval required by the bid. Silence gives the reviewer no basis for distinguishing a decision from an omission.
Should scoring descriptors become separate subparts?
Add a leaf when a published descriptor requires distinct assessable content that is not already captured by the question. Otherwise attach it as a review test to the relevant leaf. Do not invent requirements beyond the published basis.
What happens when the tender question is ambiguous?
Record the competing readings and the practical effect of each. Use the authorized clarification route before its deadline where possible. Until resolved, keep the affected leaf open and do not hide the choice inside the prose.
Can an attachment satisfy the question without narrative?
Only if the instructions allow it and the attachment contains the requested content in an assessable location. Map the exact section. A filename or a generic cross-reference does not show what the attachment contributes.
Is full coverage evidence that the answer will score well?
No. Coverage shows that every identified ask has an inspectable response. Evaluators may still judge the method weak, the proof unconvincing or the offer less advantageous under the published methodology.
Sources
Primary references
- Supplier guidance on responding to award criteria and component questions Crown Commercial Service
- Guidance on assessing competitive tenders, updated August 2026 UK Cabinet Office
- Procurement Act 2023, section 23, award criteria The National Archives
- PPN 002 guide to using the Social Value Model UK Cabinet Office
- Directive 2014/24/EU, Article 67, contract award criteria EUR-Lex
- German Procurement Ordinance, section 58, award and evidence German Federal Ministry of Justice and Federal Office of Justice
- German Act against Restraints of Competition, section 127 German Federal Ministry of Justice and Federal Office of Justice
- Practical guide to SME access to public procurement French Directorate for Legal Affairs and OECP
- FAR 15.203, requests for proposals Acquisition.gov
- FAR 15.204-5, proposal instructions and evaluation sections Acquisition.gov
- FAR 15.304, evaluation factors and significant subfactors Acquisition.gov
- FAR 15.305, proposal evaluation against stated factors Acquisition.gov
- Preparing to bid on an opportunity Public Services and Procurement Canada
- Responding to an approach to market Australian Government Department of Finance
- Rated criteria in World Bank procurement World Bank
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.