View 5 · Model

Method & model

Five independent methods run against the same RFP inputs. None of them invents data: a method that lacks its inputs reports as blocked and is excluded from the reconciled ROM rather than contributing a guess.

1 · Comparable historical

data-backed

Scores all 156 archived opportunities against the new RFP using the similarity specification shipped with the corpus. Weights are renormalised over the fields the RFP actually supplies, so an unfilled field never counts as a mismatch. Matches below the similarity floor (0.45) are returned as “no comparable found” rather than padded with weak matches.

Source Fact comparable effort figures · Derived Metric percentiles across the match set. Matches whose totals are byte-identical template carryover are flagged and called out in the result.

2 · Parametric factors

8 of 19 factors supported

Applies corpus-derived effort and duration factors to budget and duration inputs. Any factor the corpus cannot support is blocked with its reason, visible on the Benchmarks view. Factors are pricing behaviour, not delivery productivity.

⚠ Bid archive, not a delivery archive
Nearly every input effort figure was worked back from a price ceiling. The parametric layer reproduces how Cardinality prices work — it is not a validated productivity model.

3 · Bottom-up WBS

requires human input

24 work components, each entered by the estimator in person-months. The archive contains only two WBS tabs, so component values cannot be seeded from history — anything entered here is Assumption. The method returns a 0.9× / 1.0× / 1.2× range around the entered total.

4 · Product fit-gap

blocked — no maturity sheet

Maps RFP requirements to product modules and assigns a disposition (configure, extend, build, integrate, and so on) with an effort rule per disposition. It requires a product module maturity sheet, which is not present in the archive. Until one is loaded, this method reports blocked and contributes nothing to the ROM.

5 · External benchmark

manual entry

Comparable public procurements entered by hand with their source URL. Always labelled External Benchmark and never blended silently into internal figures.

Reconciliation

how the ROM range is formed
  • Each unblocked method contributes p25 / p50 / p75 in person-months.
  • Expected ROM = median of method p50s. Aggressive = lowest p25. Conservative = highest p75.
  • When the spread between the highest and lowest method p50 exceeds 1.5×, the range is widened and a dispersion warning is raised — the methods disagree and the estimate should not be presented as tight.
  • Price is computed separately from effort: role mix × rate card × 160 hours per person-month, then margin, discount and onshore mix. It is never back-solved from a target price.

Fact labelling

Every figure in this workbench carries one of five labels: Source Fact measured in a source workbook, Derived Metric computed from source facts, AI Inference inferred by classification, Assumption supplied by a human, and External Benchmark taken from outside the archive. Nothing is presented without one.