A question-to-template dependency map is a versioned record connecting one tender question or requirement to the buyer-controlled surface needed to answer, declare, calculate, evidence or deliver it. The source end keeps the exact instruction and scope. The target end identifies the current form, response schedule, workbook, field or portal control and the smallest recoverable location inside it. A typed edge states what must happen between them. The map can prove that an answer belongs in a named field, that a narrative value must agree with a workbook cell, or that a form is missing. It does not complete the template, choose answer content, alter buyer files or certify the final submission.
Proposal plans usually assign questions as if every answer were a block of prose. Buyer packs rarely behave that way. One instruction asks for a method statement in a Word schedule. Another asks for the same service level in a pricing workbook. A personnel answer may be incomplete until each named expert has a separate form. A portal can collect a declaration that does not exist as a file. Meanwhile, an annex mentioned in the RFP may be absent, superseded or supplied only after selection. If the team drafts against a homemade document, it may lose mandatory fields, repeat a value inconsistently or discover too late that the buyer surface cannot accept the planned answer.
Do not label an answer template-dependent merely because a form is nearby. Prove the instruction, the current target and the relationship between them. Bind both ends to the same procurement, stage, lot, bidder configuration and package snapshot. Preserve a visible locator for reviewers and a representation-specific locator for software. Give missing files, ambiguous targets and unverified versions explicit states. Draft research can continue in a controlled sidecar, but no dependent answer is final until the current buyer surface and every material reconciliation are known. An agent may discover and test links; it may not fill, unlock, repair, upload or submit the buyer artifact.
Dependency test
A question is template-dependent only when the buyer surface changes what complete means
Begin with an issued proposition, not with a folder of forms. “Describe the transition plan” is a question. “Complete Schedule T, row 4.2, for the transition plan” adds a dependency. The answer is not ready merely because the prose exists; it must reach the named target. By contrast, a schedule listed as background reading does not create an answer dependency. Record the exact verb, object and condition that connect the two.
The relationship can be direct or distributed. A narrative response may live in one field while a service level appears in a separate workbook and a supporting declaration sits in a portal. The question depends on all three only if the documents make each one part of that response or require the values to agree. Co-occurrence, similar labels and reviewer preference are not enough.
Some procurement systems state these links with unusual clarity. FAR 15.204-5 allows proposal instructions to specify a format or severable parts. The World Bank February 2025 consulting RFP directs consultants, within that document family, to prepare technical and financial proposals using the standard forms in Sections 3 and 4. Those are scoped examples. They do not prove that an unrelated buyer requires the same sections or permits the same form.
Portal data deserves the same test. The European Commission eSubmission guide separates structured tender data from uploaded financial and technical documents, and the interface blocks progression until its mandatory data and documents are present. A portal control can therefore be the required response surface even when no template file exists. A screenshot or field name from another procedure cannot establish that dependency for the current tender.
| Test | Evidence needed | If evidence is absent |
|---|---|---|
| Instruction | Exact wording requiring or conditioning use of the surface | Keep the relationship unproved |
| Target identity | Current buyer object, authoritative source and version | Return template_missing or version_unverified |
| Target location | Recoverable field, row, cell, section or portal control | Return target_ambiguous |
| Scope | Matching lot, stage, entity, participant and condition | Return scope_unresolved |
| Operation | What the answer must do in relation to the target | Do not use a generic related_to edge |
Object identity
Bind the dependency to the buyer’s current object, not a familiar master form
Record a template as a controlled object. Useful fields include the buyer title, package identifier, source URL or portal location, issued filename, media type, file size, publication or retrieval time, version label, checksum, amendment relationship and access state. A filename such as “Technical_Response_Final_v2.docx” is evidence to preserve, not proof of authority. The same name can be reused after replacement, and a browser download can add a suffix that the buyer never issued.
A public standard library is not the live tender pack. The World Bank framework page publishes dated standard procurement documents, while an actual borrower RFP contains a procurement-specific Data Sheet and selected forms. EBRD likewise maintains a library whose forms are included in particular procurement documents. These sources help a reviewer understand possible form families. The current buyer package still decides which instantiated form, revision and optional clauses apply.
Version evidence must reach the target field. If an addendum replaces Schedule B, every dependency to cells or controls in the old Schedule B expires, even when most labels look unchanged. Compare the issued change statement and the native objects. Do not transfer old coordinates by nearest text match. Re-resolve the target and require review where rows moved, identifiers changed or a field was split.
Missing is a result, not an invitation to improvise. Record whether the object was absent from the controlled package, inaccessible under the customer’s permitted account, named differently, corrupted, password protected or promised for later release. Each state has a different next action. A generic form, a prior tender or an internal replica can support analysis, but none may be presented as the buyer’s current target.
| State | Meaning | Permitted next action |
|---|---|---|
| current_exact | The issued object and version are proved for this scope | Resolve its target fields |
| version_unverified | The object exists but currentness is not proved | Check package and amendment evidence |
| superseded | A later buyer issue expressly replaced it | Keep history and remap affected links |
| template_missing | The instruction names an object outside the accessible package | Raise a precise package or clarification request |
| access_required | The object may exist behind a permitted account or stage | Ask the authorized owner to retrieve it |
| unreadable | The received object cannot be inspected reliably | Preserve it and request an authoritative usable copy |
Field identity
Give the instruction and the target two locators each
A dependency has two ends. At the source, preserve the question identifier, exact wording, surrounding heading, printed or rendered page, clause, table row and a machine-recoverable selector. At the target, preserve the buyer-visible field label plus the native address that locates it in the recorded representation. A reviewer needs to see why the link exists. An agent needs to recover the same two occurrences without trusting search rank.
Native structure matters. ECMA-376 defines the vocabularies, representation and packaging used by Office Open XML. A DOCX may contain WordprocessingML paragraphs, tables or content controls. Microsoft’s content-control documentation shows that controls can have titles, tags, XML mappings and repeating behavior. Use those properties when present, but bind them to the file version. Copy and paste, nesting or regeneration can break a mapping while the visible label remains.
For a workbook, store the workbook identity, sheet name, cell or range, nearby row and column labels, formula or validation role and any named range that the file exposes. “Cell F12” alone is weak because an inserted row changes its meaning. “Service Levels, row Incident severity 1, column Proposed minutes, F12 in checksum X” gives a person and an agent independent ways to test the target. Do not edit protected structure to make it easier to parse.
PDF and portal targets need the same discipline. A fillable PDF may expose a field name, while a flat PDF may offer only a page, label and text or geometry selector. A portal field may have a visible prompt, stable control identifier, section path and observed behavior in one authorized session. Record the observation time and account scope. Never infer hidden fields, bypass access or replay an action merely to improve the locator.
| Surface | Buyer-visible locator | Representation-bound locator |
|---|---|---|
| Word form | Section, prompt and table position | Content-control tag or ID, XML path, or text plus context |
| Spreadsheet | Sheet, row label and column heading | Cell or range, named range, formula role and file checksum |
| Fillable PDF | Page, field caption and nearby heading | Form field name or object reference tied to the PDF |
| Flat PDF | Page, table row and printed label | Text quote with context or bounded page coordinates |
| Portal | Stage, section and visible prompt | Observed field identifier and permitted-session timestamp |
Relationship semantics
Name the operation that joins the question to the template
A generic link called related_to cannot control response work. Use a directed verb. answer_in means the substantive response belongs in the target. copy_value_to means an approved fact must be transcribed there. reconcile_with means two separately authored surfaces must agree. attach_from means the required deliverable is the completed or signed form. calculated_by means a buyer formula produces a value used by the answer. declares_in means the answer is made through a specified declaration surface. conditional_on means another field or bidder choice activates the target.
One question can have several edges. A staffing question might require a short narrative in a response schedule, one personnel form per proposed expert and a portal declaration about subcontracting. Keep each edge separate because its owner, state and consequence can differ. The narrative can be ready while one expert form remains missing. The question as a whole stays blocked if the missing edge is material.
Several questions can also converge on one form. A technical schedule may collect architecture, implementation, support and security responses. Preserve separate question identifiers and approval states even when they share a table. A reviewer must be able to tell which question authorized each entry. Do not merge answers merely because the buyer gives them adjacent rows.
Carry conditions and subjects on the edge. “Complete for each consortium member” is different from “complete once for the bidding group.” “For every lot tendered” differs from “one response covering all selected lots.” The dependency map records the condition and links to the response configuration. AN-072 owns the full object graph and instance count; AN-074 proves which questions depend on which template instances.
| Type | Operational meaning | Release test |
|---|---|---|
| answer_in | The substantive answer belongs in the target | Approved content is present in the current target |
| copy_value_to | A controlled value must be transcribed | Target value matches the approved source and unit |
| reconcile_with | Separate surfaces express the same proposition | Values and scopes agree or the difference is explained and allowed |
| attach_from | The completed form is a required deliverable | Correct instance exists with required review |
| calculated_by | The buyer artifact derives an answer value | Formula lineage and displayed result are verified |
| declares_in | The response is made through a declaration control | Authorized declarant and exact proposition are reviewed |
| conditional_on | A choice or fact activates another target | Trigger is decided and dependent state recomputed |
Readiness
Separate useful drafting from permission to call the answer finished
A missing form does not always stop research. Writers can collect evidence, draft a bounded sidecar response and list assumptions without modifying or imitating the absent buyer artifact. Mark that work content_ready_template_blocked. The label says that content may be mature while the submission response is not. It prevents a polished draft from bypassing the missing target.
Use controlled states for the dependency itself: linked_current_exact, linked_current_with_review, template_missing, version_unverified, target_ambiguous, scope_unresolved, source_conflict, superseded_dependency, not_template_dependent and human_review_required. Confidence is not a substitute for state. A high similarity score between a question and a worksheet tab cannot turn an ambiguous target into an exact link.
The gate should be proportional. A missing optional worksheet can remain a named review item if the buyer confirms that it does not apply. A missing mandatory declaration, pricing schedule or evaluated response field blocks finalization. Record the material effect, owner, due date and permitted resolution route. Do not mark the whole bid blocked when only one question is affected, and do not release the affected question because unrelated sections are complete.
Reconciliation is part of finalization. If a service level appears in the narrative, SLA table and price assumptions, all linked propositions need compatible scope, units and approval. The dependency map does not decide the service level or price. It proves where the approved decision must appear and whether each occurrence still agrees after insertion.
| Question state | What is true | What remains prohibited |
|---|---|---|
| not_template_dependent | No buyer surface is required by the reviewed sources | Inventing a form requirement |
| content_in_progress | Research or drafting can continue in a controlled workspace | Calling the submission answer complete |
| content_ready_template_blocked | Substance is reviewed but a material target is unresolved | Final approval, transfer or submission |
| linked_for_population | Current target and operation are proved | Population without the required content and authority approvals |
| reconciled_for_review | All material target occurrences agree and have evidence | Final submission actions outside the authorized workflow |
Worked example
Map a fictional water-utility security RFP without filling its forms
Assume a fictional water utility issues an operational technology security RFP. The controlled package contains Instructions v3, Technical Response Schedule B v3 in DOCX, SLA Workbook D v2 in XLSX, Personnel Form P2 and Addendum 04. The package index still mentions Hosting Declaration H8, but no H8 object is present. The bidder is responding to Lot 1 as a sole supplier. These facts are invented for the example and describe no real procurement.
Question T14 asks the bidder to describe incident triage and escalation. Instructions v3 says the answer must be entered in Schedule B. The target is Schedule B, section 4, row T14, visible label “Incident triage and escalation,” with content-control tag TECH-T14 in the recorded file. That creates answer_in. A nearby workbook tab called Incident Costs is not linked by the instruction and receives no edge.
Question T15 asks for maximum response times by severity. Its narrative belongs in Schedule B row T15, while the exact minutes must also be entered in SLA Workbook D, sheet Service Levels, rows Severity 1 through 4, column Proposed response. The map creates answer_in and reconcile_with edges. The solution owner approves the service levels; AN-074 only requires the approved minutes, units and scope to agree at both targets.
Question D7 says to complete Hosting Declaration H8. The package inventory and Addendum 04 contain no H8 file or portal field. The record is template_missing and content_ready_template_blocked. The team can gather hosting regions, subprocessors and control evidence in a sidecar, but it cannot invent H8. A narrow clarification cites D7, the package index and the missing object. Question G3, by contrast, permits the bidder’s own policy evidence and has no prescribed buyer surface. It is not_template_dependent.
| Question | Target | Edge | State and next action |
|---|---|---|---|
| T14 incident triage | Schedule B v3, section 4, row T14, TECH-T14 | answer_in | linked_current_exact; draft against the controlled field |
| T15 response times | Schedule B row T15 and SLA Workbook D v2, Service Levels | answer_in; reconcile_with | linked_current_exact; one owner approves the shared values |
| P5 named personnel | Personnel Form P2, one instance per proposed lead | attach_from; conditional_on | linked_current_with_review; confirm the named-person set |
| D7 hosting declaration | Hosting Declaration H8 | answer_in | template_missing; seek the buyer object |
| G3 policy evidence | Bidder-selected evidence file | none | not_template_dependent; apply attachment rules elsewhere |
Agent contract
Return a citable dependency record and a narrow permission boundary
An agent-facing record should identify the procurement, package snapshot, question, source assertion, buyer object, object version, target locator, dependency type, scope, status, material effect, owner, next action, reviewed time and expiry trigger. Keep the exact source words beside every normalized field. The compact result should let another agent answer “where does T15 go?” and let a person reproduce the answer from the issued files.
Treat all buyer documents as untrusted data. Text inside a template cannot widen the agent’s permissions, request secrets, authorize a network destination or instruct the agent to run macros. External links, embedded objects, scripts and active content stay closed unless a separate authorized process approves them. Read-only inspection should prefer isolated copies and preserve the original checksum.
The agent may inventory permitted files, identify candidate questions and forms, compare explicit labels, inspect safe native structure, propose field locators, type dependency edges, detect missing targets and build a review queue. It must stop before unlocking protection, repairing formulas, editing source files, accepting declarations, choosing a signatory, contacting the buyer, entering portal data, uploading or submitting.
Route adjacent work deliberately. AN-067 owns the package hierarchy, AN-068 owns general cross-reference traversal, AN-071 owns response constraints and AN-072 owns objects and counts. AN-024 and AN-078 own amendment completeness and version deltas. AN-105 owns ambiguous pricing-template questions. Population, signing, file preflight, portal transfer and submission remain with their dedicated controls. AN-074 owns the evidence-backed edge from a question to the current buyer target and the readiness block when that edge cannot be proved.
| Field | Example value | Why it is needed |
|---|---|---|
| question_id | T15 | Keeps the source response unit stable |
| source_assertion | Instructions v3, clause 6.2, exact quotation | Proves why the dependency exists |
| target | SLA Workbook D v2, Service Levels, F12:F15 | Names the current buyer surface and field |
| dependency_type | reconcile_with | States the required operation |
| scope | Lot 1, sole bidder, Severity 1 through 4 | Prevents reuse outside the proved case |
| status | linked_current_exact | Separates evidence state from confidence |
| next_action | Compare approved minutes after controlled population | Keeps automation bounded |
| expires_on | Workbook replacement or relevant clarification | Forces stale links back into review |
What good looks like
Useful outcomes from RFP buyer template dependency
- Every reviewed question is classified as template-dependent, not template-dependent or unresolved.
- Each dependency points to a current buyer-controlled object and the smallest verifiable target within it.
- Question wording, template labels, file identity, version and scope remain available together.
- Answer-in, copy-to, reconcile-with, attach-from and conditional dependencies stay distinct.
- Missing, superseded, inaccessible and ambiguous templates block only the affected work.
- Repeated forms are tied to the correct lot, entity, member, expert, site or option.
- Draft content can progress outside the buyer file without being mistaken for a finished response.
- Changes to a question or template reopen every affected dependency and downstream review.
- Agents can return a compact map with evidence, uncertainty and a permitted next action.
Operating model
How to run the work
- 01
Freeze the question and package scope
Name the procurement, package snapshot, stage, selected lots, bidder structure and exact question set before looking for template targets.
- 02
Capture the dependency claim
Preserve the instruction that says where or how the response must be supplied, including its source version, context, condition and scope.
- 03
Resolve the current buyer object
Match the named form or surface to a package record, authoritative download or permitted portal view and record version evidence rather than relying on its filename.
- 04
Locate the exact target
Pair the buyer-visible label, heading, sheet or cell with a native identifier or recoverable selector that works only against the recorded version.
- 05
Type and scope the edge
State whether the answer lives in, copies to, agrees with, attaches from or is conditionally activated by the target, then preserve its lot, entity and participant scope.
- 06
Set the drafting gate
Allow bounded research or sidecar drafting where useful, but block finalization when a material target, version, scope or reconciliation remains unresolved.
- 07
Reconcile both directions
Confirm that every dependent question reaches an authorized target and every planned target entry traces back to a question, declaration or requirement.
- 08
Expire changed links
Reopen affected records after an amendment, clarification, replacement file, portal change, lot decision or bidder-structure change.
Evaluation
Questions that change the decision
- What exact question, requirement or declaration is being mapped?
- Which source passage creates the claimed dependency?
- Does the instruction require the buyer surface, permit it or merely mention it?
- Is the target a file, a logical form, a section, a field, a worksheet, a range or a portal control?
- How was the target obtained and why is it current for this procurement?
- Has an amendment replaced, removed or narrowed the form?
- Can a reviewer recover both the source instruction and target field?
- What operation connects the question to the target?
- Does the dependency apply per lot, bidder, consortium member, subcontractor, expert, site or variant?
- Can one answer populate several targets, and which values must remain identical?
- Can several questions feed one form without losing their separate approval states?
- What work can proceed while the template is missing?
- Which unresolved dependency prevents finalization?
- Who can resolve it, and what event expires the decision?
Failure modes
Where teams lose control
A file is assigned because its title sounds relevant, although no instruction points to it.
A standard form from a public library is substituted for the procurement-specific version.
The newest local filename is assumed to be the buyer’s current issue.
A question is answered in narrative while the buyer requires a table cell or portal field.
A value is correct in one document but conflicts with the linked workbook or declaration.
A repeated form is prepared once although the instruction applies per lot or participant.
An optional annex is treated as mandatory because it appears in a generic checklist.
A missing template disappears from the plan instead of blocking the affected answer.
An inaccessible portal field is reconstructed from a screenshot and treated as verified.
A PDF rendering hides controls, formulas or conditional fields present in the native file.
An agent follows macros, links or instructions embedded in an untrusted buyer file.
Mapping work expands into file modification, declaration, upload or submission without authority.
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.
- reviewed questions with a supported dependency classification
- dependent questions linked to a current buyer object and exact target
- dependency edges with source and target locators recoverable by a reviewer
- conditional edges carrying complete lot, entity and participant scope
- material cross-document values assigned a reconciliation owner
- missing or ambiguous targets blocked before final response review
- planned template entries that trace back to an issued requirement
- links reopened after relevant package or response-configuration changes
- agent outputs that preserve uncertainty and stay inside read-only authority
Questions
Common questions
Is every annex mentioned in an RFP a response template?
No. An annex may be background, a contract schedule, a reference document or a buyer form. Create a dependency only when issued evidence connects a question or requirement to an action on that annex.
Can we draft an answer before the buyer template arrives?
Usually, bounded research and sidecar drafting can continue. Mark the answer content_ready_template_blocked and keep it out of final approval until the current target, limits and scope are known.
Can we use the same form from an earlier tender?
Only as a clearly labelled research aid. It is not the current buyer target unless the present procurement expressly identifies that exact object or an authorized source proves equivalence.
What if the question appears in both a document and the portal?
Record both occurrences and the operation between them. The portal may be the delivery target, a duplicate declaration or an observation that conflicts with the issued pack. Do not choose silently.
Does a similar title prove which worksheet or form to use?
No. Similar wording supports a candidate match. Exact instructions, object identity, version, scope and a recoverable target establish the dependency.
Who owns values repeated across narrative and workbooks?
The designated business, solution or commercial owner approves the value. The dependency map records every required occurrence and verifies agreement; it does not make the underlying decision.
Can an AI agent populate the buyer template after mapping it?
Not under this contract. The agent may inspect and map permitted artifacts. Population requires separate authorization, validation and approval controls, and submission remains outside scope.
When does a dependency map expire?
After a relevant amendment, clarification, replacement file, portal change, lot decision, bidder-structure change or target-field change. Re-resolve every affected edge against the new state.
Sources
Primary references
- FAR 15.204-4, list of documents and attachments Acquisition.gov
- FAR 15.204-5, representations and instructions Acquisition.gov
- Quick guide for open procedures in eSubmission European Commission
- Downloadable eSubmission Quick Guide for Economic Operators European Commission
- Procurement specific questionnaire under the Procurement Act 2023 UK Government Commercial Function
- Project Procurement Framework and current standard documents World Bank
- Request for Proposals for Consulting Services, February 2025 World Bank
- Library of Forms and Alternative Options European Bank for Reconstruction and Development
- EBRD Form TEC-1, Technical Proposal for goods European Bank for Reconstruction and Development
- Commission Implementing Regulation (EU) 2016/7, ESPD standard form EUR-Lex
- ECMA-376, Office Open XML file formats Ecma International
- Content controls in Word Microsoft Learn
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.