A tender_scoring_execution_map is a versioned reconstruction of the buyer’s evaluation sequence. It records every eligibility gate, criterion, scale, descriptor, weight, threshold, price or cost expression, cap, normalization rule, rounding instruction, subtotal, tie rule and stage dependency exactly as published. Beside each input, it identifies the response location, proof needed, current support state, owner, review depth and reopen trigger. Tested examples show how the published method behaves at its boundaries. The map explains what must be answered and checked. It does not select a bid price, predict a competitor’s price, award an internal score, choose the offer or replace a clarification where the documents are unclear.
Northbar Coast Ferries is preparing a maintenance tender for two passenger terminals. The fictional buyer awards 60 points for quality and 40 for price. Three quality answers are scored from 0 to 5, but every answer must reach 3 or the tender is rejected. The price score is 40 multiplied by the lowest admissible price divided by the tender price. Only the final total is rounded to two decimals. The transition answer is currently supportable at 3. A proposed resilience annex might support 4, but its recovery test has not been approved for external use. The team has highlighted the 20-point weight and started adding pages. It has not yet noticed that moving from 2 to 3 changes admissibility, while moving from 3 to 4 changes only four weighted points.
Treat the published scoring method as a small program that must be reproduced before it can guide work. Parse its inputs, order of operations, gates and unknowns. Run boundary cases against that reconstruction and keep each result traceable to a clause. Then connect the mechanics to the answer: threshold protection comes before optional score improvement; discrete descriptors call for proof of the next stated distinction; relative price formulas remain scenario ranges until other admissible prices exist; rounding happens where the buyer says it happens. When two documents produce different results, record both interpretations and ask through the official channel. A bidder-created formula must never silently settle an ambiguity in the buyer’s method.
Direct answer
Rebuild the buyer’s sequence before assigning answer priority
A scoring table is often read from left to right even when the evaluation runs top to bottom. A pass/fail review may happen before quality scoring. A minimum may reject the tender before price is opened. A presentation may add points after the written submission, and a tie rule may inspect one criterion rather than the total. Put these events in execution order. The answer plan should follow the point at which each event can still change the outcome.
Start with an applicability header: procedure identifier, lot, stage, procurement route, current notice and document versions, permitted variant, currency, tax basis and the response baseline being planned. Do not combine lot formulas because their headings look alike. A 60:40 split for lot 1 and a 50:50 split for lot 2 are different programs, even if both use the same five-point quality scale.
For each event, preserve the source words and create a normalized field beside them. “A score below 3 in any quality question will result in rejection” becomes a criterion gate with operator less than 3 and consequence tender rejected. Keep the quotation and source location. The normalized expression helps testing; it never replaces the clause.
The UK Procurement Act guidance calls for the assessment methodology to describe how criteria are applied, including scoring matrices and disqualifying pass/fail conditions. EU eForms likewise separates criterion numbers into weight, fixed-value and threshold dimensions. Those structures are useful reading aids, not a universal formula. The tender documents still control the case in front of you.
| Event | Published rule | Controlled representation |
|---|---|---|
| Applicability | Lot, stage, variant and document version | Exact scope key |
| Gate | Condition, operator, value and consequence | Boolean test before scoring |
| Quality score | Scale, descriptor, maximum and weight | Allowed raw value and weighted expression |
| Financial score | Inputs, formula, eligibility and units | Named variables with valid domains |
| Aggregation | Subtotals, order, caps and rounding | Ordered calculation with precision |
| Tie or later stage | Trigger and stated resolution | Conditional event after total |
Formula anatomy
Parse the verbs around every number
Two tables can display “20%” and mean different things. One may award 20 available points from a raw score of 0 to 5: raw divided by 5, multiplied by 20. Another may multiply a 100-point technical result by 0.20. A third may rank the criterion as 20 per cent of the non-price subtotal before a separate 60:40 combination. Capture the base, numerator, denominator and unit rather than copying the visible percentage alone.
Read the verbs. “Multiply,” “normalize,” “add,” “average,” “must achieve,” “will be excluded,” “may be moderated” and “rounded” specify operations or consequences. “Relative importance” may describe ordering without publishing a numerical formula. FAR 15.305, for example, allows adjectival ratings, numerical weights and ordinal rankings. Do not convert one form into another merely because a spreadsheet prefers numbers.
Price expressions need typed inputs. Distinguish tender price from evaluated price, total cost from unit rates, pre-tax from tax-inclusive values, and the lowest submitted price from the lowest admissible price. The French DAJ guidance shows why the chosen price method changes the resulting spread. It also explains that a price scoring method is not always disclosed. If the method is absent, no bidder spreadsheet can infer it safely.
List the valid domain for every input. A raw quality score may accept only 0, 1, 2, 3, 4 or 5. A denominator cannot be zero. A price comparison set may include only tenders that survived earlier checks. A score may be capped even if the equation produces more. Invalid inputs should return “cannot calculate” with a reason, not zero and not the spreadsheet’s error-suppression value.
Reproduction test
Require the reconstruction to reproduce the buyer’s examples
A buyer-supplied worked example is the first test fixture. Enter the same raw scores, prices and stage results. Compare each intermediate value, subtotal and final total at the displayed precision. A matching final number can hide offsetting errors, such as rounding one quality row down and using the wrong price basis up. The reconstruction passes only when the path matches, not merely the last cell.
When there is no worked example, use identities that the published method implies. A tender with the lowest admissible price should receive the stated maximum under a lowest-price-divided-by-tender-price formula. A raw maximum should return the full criterion points. An exact threshold should pass if the operator is “at least” and fail if the documents say “more than.” These checks do not prove the buyer’s method is lawful or sensible. They prove that your transcription behaves as written.
Preserve two calculations when wording conflicts. Suppose the scoring schedule says “round every weighted criterion to two decimals,” while an appendix says “round the final score to two decimals.” Label interpretation A and B, calculate a case where they diverge, and send a concise question through the named clarification route. Do not select the interpretation that gives your tender the larger total.
World Bank procurement regulations require evaluation criteria and methodology to be specified in detail in the request document. UNCITRAL’s Model Law says only the disclosed criteria and procedures should be used in the disclosed manner. These are useful controls across jurisdictions, but they do not turn a general rule into a missing clause for a particular tender. Missing method detail stays missing until an authoritative source resolves it.
| Test | Input | Expected observation |
|---|---|---|
| Buyer example | Published sample values | All displayed intermediates and total match |
| Raw maximum | Maximum permitted score | Full criterion contribution, subject to stated cap |
| Exact threshold | Value equal to the minimum | Pass or fail follows the published operator |
| One step below | Next allowed value below minimum | Published rejection or loss applies |
| Price identity | Tender price equals lowest admissible price | Maximum financial points |
| Invalid domain | Zero, blank or wrong unit | Calculation stops with a named defect |
| Rounding divergence | Values with repeating decimals | Precision follows the stated calculation stage |
| Tie | Equal final totals | Published tie event is applied once |
Edge cases
Find the places where a small score change has a different consequence
A weight measures contribution inside a calculation. A gate changes whether that calculation continues. If Northbar’s transition response receives 2 instead of 3, the tender is rejected under the fictional example. The numerical difference inside the 20-point criterion is four points, but the procedural consequence is the loss of the whole bid. The answer plan therefore protects credible evidence for score 3 before considering work intended to support score 4.
Discrete scales create steps. A paragraph that makes an answer “a little better” has no formula effect unless it supports the next available descriptor. Read the difference between 3 and 4 word by word. If 4 requires a tested recovery route and evidence of comparable use, assign the test record and customer evidence decision. Extra prose about commitment does not bridge that gap.
Rounding creates smaller boundaries. Assume three weighted rows produce 12.666, 17.499 and 8.666. Rounding each first yields 38.84; summing first yields 38.831 and then 38.83. A hundredth rarely deserves its own writing project, but it matters when reproducing a total or applying a tie condition. Record it rather than smoothing it away.
Unknown competitor prices create a moving input, not a missing excuse. Use several valid values for the lowest admissible price to see the sensitivity of the published financial formula. Keep the range beside its assumptions and observation date. The result can show how much the financial component moves; it cannot reveal what another supplier will bid or how an evaluator will score quality.
Answer plan
Turn each score event into a bounded answer control
Connect every gate and scored criterion to the exact response surface. Record the portal field, question, attachment, presentation item or price workbook cell. Then state the minimum proof condition using the buyer’s descriptor. For Northbar’s transition answer, score 3 requires a feasible plan with responsibilities, milestones and managed dependencies. The answer control must point to those elements and the records that support them.
Use three review bands. Gate protection checks complete coverage, permitted evidence and consistency with the approved baseline. Descriptor movement checks whether the next published distinction is present and proved. Stability review protects mature answers from late contradictions. These are priority bands, not percentages of the team’s hours. The existing response-budget process remains responsible for deciding the total time and transferring hours between workstreams.
Assign the decision to the owner who controls the input. An answer writer can show where the transition sequence appears. Operations confirms that the sequence is deliverable. The evidence owner approves the test record for this use. Commercial confirms that a priced resource is included. The release owner checks the rendered answer against the formula map. One person may hold several roles, but each decision remains named.
Add a stop condition. If the available evidence supports only 2 and 3 is mandatory, the plan should not disguise the gap as a drafting task. It routes the issue to evidence closure, offer authority or the bid decision. If the published formula itself is ambiguous, the stop condition is clarification. Scoring analysis cannot create authority the tender documents do not provide.
| Formula condition | Answer control | Review disposition |
|---|---|---|
| Pass/fail or minimum | Complete response plus proof of the stated floor | Protect, close gap or stop |
| Next discrete descriptor | Exact additional distinction and permitted evidence | Pursue only if supportable |
| High weighted contribution | Coverage and proof at approved response position | Review influence, not page volume |
| Relative price input | Approved price basis and scenario bounds | Commercial review remains separate |
| Rounding or cap | Calculation check tied to source instruction | Correct model, no prose response |
| Tie or later stage | Named presentation or criterion evidence | Prepare only when trigger applies |
| Conflict or missing rule | Two traced interpretations and divergence case | Clarify, do not assume |
Price boundary
Reproduce price sensitivity without choosing the price
Financial formulas often depend on other admissible tenders. Northbar’s fictional expression is price points = 40 x lowest admissible price / Northbar tender price. Once the commercial baseline fixes Northbar’s evaluated price at 960,000, a lowest admissible price of 840,000 produces 35 points; 900,000 produces 37.5; 960,000 produces 40. These are conditional results, not estimates of what the market will submit.
Validate the comparison set before calculating. “Lowest price” may exclude non-compliant or rejected tenders. The evaluated price may include options, scenario quantities or corrections that differ from the amount on the offer form. Capture the exact source field and unit for Northbar’s own price. Leave the competing input unknown until a stated scenario or an authoritative post-award record supplies it.
Do not reverse-engineer a bid price from a desired total in this workflow. Price setting requires cost, margin, risk, tax, cash, approval and delivery decisions that a scoring reconstruction does not own. A sensitivity table may show a switching condition for the commercial reviewer, but it must not edit the approved price or characterize one scenario as likely without evidence.
Relative scoring also means your score can move while your own tender stays unchanged. Current UK guidance warns buyers that relative price mechanisms can affect value for money and recommends scenario testing at design time. For bidders, the practical consequence is narrower: preserve uncertainty, test plausible domains and avoid converting the formula’s mathematical precision into knowledge of competitors.
| Lowest admissible price | Northbar price | Price points | Permitted reading |
|---|---|---|---|
| 840,000 | 960,000 | 35.00 | Conditional result if 840,000 is the valid minimum |
| 900,000 | 960,000 | 37.50 | Conditional result if 900,000 is the valid minimum |
| 960,000 | 960,000 | 40.00 | Northbar is tied for the valid minimum |
| Unknown | 960,000 | Not calculable | Keep the variable unresolved |
| 0 or blank | 960,000 | Invalid input | Stop rather than silently award zero |
Worked case
Northbar protects the transition threshold before seeking four more points
The Northbar example is fictional. Its published sequence first checks the submission and three quality minimums. Transition has 20 points, maintenance method 25 and reporting 15. Each uses an integer score from 0 to 5 and requires at least 3. Weighted quality is raw score divided by 5, multiplied by the criterion points. Price is considered only for tenders that survive those gates. The final total is rounded to two decimals, with higher quality breaking a tie.
The team transcribes current support as transition 3, maintenance 4 and reporting 3. That gives 12, 20 and 9 quality points, for a subtotal of 41. The unapproved resilience annex is not counted. At a 900,000 lowest-price scenario, the price contribution is 37.5 and the conditional total is 78.5. More important for the answer plan, transition and reporting sit exactly on rejection boundaries.
Transition receives gate-protection review. The plan names the mobilization sequence, buyer dependencies, responsible role, cutover evidence and fallback decision in the response. An operations owner confirms feasibility; an evidence owner approves the existing rehearsal record. Reporting receives the same threshold test with its own evidence. Maintenance gets descriptor-movement review only after both floors are stable. The resilience annex remains a blocked candidate until its scope and disclosure permission are accepted.
The plan never awards Northbar an evaluator score. It states which descriptor the current approved material is designed to support and what could invalidate that position. A changed transition date, loss of the named lead, revised price schedule, clarification on rounding or new threshold wording reopens the affected rows. The frozen map travels with the response baseline so reviewers can repeat every conditional calculation.
Change control
Freeze the formula map with the response version it governs
Give the map a version and bind it to source fingerprints or controlled document identifiers, the selected lot and the response baseline. Record the calculation engine or spreadsheet version, test results, reviewer and approval time. A correct formula attached to an old addendum is still the wrong control.
Reopen by dependency rather than rebuilding everything. A clarification that changes one threshold affects its boundary tests, response mapping and any aggregate result. A revised price sheet affects financial inputs and conditional totals. A new score descriptor affects the evidence test for that criterion. A lot-wide change to rounding may affect every subtotal. The dependency record shows the smallest safe review scope.
Keep the final worksheet inspectable. Hide neither constants nor intermediate values. Protect formulas from accidental editing, but preserve a readable expression and the source beside each one. If a tool evaluates the map, its output should identify the input set, unresolved variables and failed validation rules. A single unexplained “win score” is not an acceptable work product.
After award, compare the assessment summary or debrief with the frozen map. The comparison can test whether the response exposed the planned evidence and whether the published method was understood. It cannot prove that a different answer would have won. Store the finding as evidence for future planning, and keep hindsight separate from the tender facts available before submission.
What good looks like
Useful outcomes from translate tender scoring formula into answer plan
- One controlled formula map reproduces the buyer’s published evaluation sequence without hidden bidder assumptions.
- Eligibility gates, quality thresholds and score contributions remain separate decisions.
- Every calculation cites the criterion, version, clause and stated order of operations it uses.
- Boundary tests expose discontinuities, caps, ties, invalid inputs and rounding effects before drafting is prioritized.
- Each score-changing input is linked to a response location, required proof, owner and review control.
- Unknown competitor prices and evaluator judgements remain ranges or unresolved variables rather than forecasts.
- Price calculation is reproduced for sensitivity testing while the commercial team retains price-setting authority.
- A source change, clarification or bid-baseline change reopens only the affected tests and answer controls.
Operating model
How to run the work
- 01
Fix the evaluation scope
Identify the procedure, lot, stage, current document versions, tender variant and exact response baseline to which the method applies.
- 02
Transcribe the published method
Copy gates, criteria, scales, descriptors, weights, thresholds, formulas, caps, rounding, aggregation and tie rules with clause-level sources.
- 03
Separate facts from interpretations
Mark each element as explicit, derived, assumed, unknown or conflicting and prevent assumptions from entering the controlled calculation.
- 04
Reproduce worked examples
Run any buyer-supplied example through the map and require the same intermediate and final result before using it for planning.
- 05
Test the edges
Calculate exact-threshold, below-threshold, cap, tie, rounding, missing-input and relative-price scenarios.
- 06
Map mechanics to answers
Link every gate and scored input to the response words, evidence, attachment, owner and approval that can affect it.
- 07
Set bounded review priority
Protect admissibility first, then test evidence-backed movement to the next published descriptor and schedule review around fragile inputs.
- 08
Freeze and reopen deliberately
Version the formula map with the tender and response baseline, and rerun affected tests after any controlling change.
Evaluation
Questions that change the decision
- Which lot, stage, variant and tender version does this evaluation method govern?
- What happens before scoring, and which failure stops later evaluation?
- Are scores continuous, integer, ordinal, adjectival or limited to named values?
- Where do minimums apply: criterion, group, technical subtotal, stage or final total?
- Does weighting multiply a raw score, a normalized score or an already weighted point value?
- Which price, cost, baseline or competing value enters each financial expression?
- At what line and precision does the buyer round, cap or truncate?
- How are ties, moderation, consensus and later stages resolved?
- Which published descriptor can the current answer and permitted evidence support?
- Which ambiguity, document change or baseline change requires clarification or recalculation?
Failure modes
Where teams lose control
Treating a percentage weight as if it were the complete scoring formula
Adding weighted points before applying a criterion-level rejection threshold
Assuming every number between scale endpoints is available when the buyer permits only named scores
Rounding each component although the documents require rounding only the final total
Using the lowest observed market price where the formula requires the lowest admissible tender price
Turning an unknown competitor price into a confident award prediction
Planning for a higher descriptor without evidence for its stated distinction
Letting a spreadsheet default, blank cell or zero denominator produce a plausible but invalid result
Ignoring a presentation, moderation round or tie rule because it sits outside the main scoring table
Resolving conflicting buyer instructions through a private bidder convention instead of clarification
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.
- Published formula elements transcribed with clause and version references
- Formula elements classified as explicit, derived, assumed, unknown or conflicting
- Buyer worked examples reproduced exactly at every stated precision
- Gates and thresholds covered by below, exact and above-boundary tests
- Rounding, cap, tie and missing-input cases with recorded expected results
- Scored inputs linked to an answer location and named evidence owner
- Answer priorities justified by a tested score condition rather than weight alone
- Unknown variables reported as scenarios, ranges or unresolved dependencies
- Changes that triggered recalculation and targeted answer reapproval
Questions
Common questions
Should proposal effort follow the scoring weights exactly?
No. Weights describe contribution to the buyer’s method. Gates, evidence gaps, dependencies and answer maturity affect review priority. A separate budget process allocates the team’s total hours.
Can the formula predict the score we will receive?
Only deterministic components can be calculated from known inputs. Evaluator judgement, moderation and competitor-dependent inputs remain uncertain. Plan against published descriptors without presenting an internal estimate as the buyer’s score.
What if the buyer did not disclose its price formula?
Record that the method is unavailable and follow the applicable clarification route if the question is permitted and useful. Do not infer a formula from common practice or another tender.
Should we round every weighted criterion?
Round only where and how the tender documents instruct. If the stage is unclear and different choices change the result, preserve both calculations and request clarification.
How should pass/fail criteria affect the answer plan?
Give them gate-protection review before optional score improvement. Confirm complete coverage, support, ownership and consistency. An unmet mandatory floor is not a writing-polish issue.
Can we model competitor prices?
You can calculate transparent scenarios across a valid range. Label the input and basis. Do not turn a scenario into a prediction or use it as authority to change the bid price.
What if the scoring schedule conflicts with an appendix?
Preserve both clauses, apply the tender’s document-precedence rule if it clearly resolves the conflict, and otherwise ask through the official channel. Do not choose the more favorable result privately.
What can software automate safely?
It can parse candidate fields, run controlled calculations, test boundaries and trace dependencies. People still confirm the controlling text, evidence meaning, commercial price, legal interpretation and release decision.
Sources
Primary references
- Guidance on assessing competitive tenders under the Procurement Act 2023 Government Commercial Function
- Official Procurement Act learning module on assessment and award Government Commercial Function
- Guidance on assessment summaries, criterion scores and reasons Government Commercial Function
- Procurement Act 2023 section 23 on award criteria and assessment methodology The National Archives
- Procurement Act 2023 section 24 on refining award criteria The National Archives
- Procurement Regulations 2024 regulation 31 on assessment summaries The National Archives
- Directive 2014/24/EU, current consolidated English text EUR-Lex
- TED eForms business terms for award criteria, weights and thresholds Publications Office of the European Union
- TED eForms schema guidance for criterion number dimensions Publications Office of the European Union
- TED eForms number-weight codelist Publications Office of the European Union
- World Bank Rated Criteria guidance for borrowers and suppliers World Bank
- World Bank Procurement Regulations, seventh edition World Bank
- ADB guidance note on merit point criteria Asian Development Bank
- ADB merit point criteria builder and weighting controls Asian Development Bank
- FAR 15.304 on evaluation factors and their relative importance Acquisition.gov
- FAR 15.305 on proposal evaluation and rating methods Acquisition.gov
- UNCITRAL Model Law on Public Procurement United Nations Commission on International Trade Law
- UNCITRAL Guide to Enactment of the Model Law on Public Procurement United Nations Commission on International Trade Law
- OECD Recommendation of the Council on Public Procurement Organisation for Economic Co-operation and Development
- OECD guidelines for fighting bid rigging in public procurement Organisation for Economic Co-operation and Development
About the author
Tony Kim
Tony writes about applied AI, dependable product engineering and the systems that turn complex response work into controlled delivery.
Zelius
Managed tender intelligence and bid execution for teams that want the commercial outcome.
Suppliers, founders and commercial teams pursuing public or private opportunities. Start with the workflow, constraints and evidence you already have.