Answer Capsule: Apex Prometheus defines a controlled subcontractor performance scorecard as an evidence-and-review system, not a truth machine. It identifies the company, subcontract, scope, project, evaluation period, criteria version, evaluator, and source record behind each rating. Observations, judgments, totals, objections, corrections, recommendations, and authorized decisions stay separate. When the evidence is disputed or incomplete, the system stops and sends the issue to a qualified person.
A bad scorecard turns jobsite noise into a permanent label. Three punch items become “poor quality.” One schedule miss gets blamed on a subcontractor even though the area was not released. Then somebody treats the total like settled fact.
That is not control. That is a rumor with a decimal point.
General contractors and construction managers across New York City, Staten Island, Brooklyn, and the tri-state area run complicated work under tight schedules. A score may affect corrective action, bidding, or future selection. If nobody can trace that score back to the exact project facts, it has no business carrying that weight.
Draw the Line Between Six Different Stages
A construction subcontractor evaluation belongs inside a lifecycle. It should not replace the stages around it.
- Prequalification asks whether a company appears capable before award.
- Onboarding confirms documents, contacts, access, and readiness before mobilization.
- Performance observation records what happened during delivery.
- Corrective action addresses a defined issue under the applicable contract and policy.
- Final evaluation closes a stated period after an evidence cutoff and review.
- Future selection is a separate governed decision made by authorized people.
Mix those stages together and the score starts doing jobs it was never built to do. A prequalification concern is not a field failure. An interim review is not a final record. A future-work recommendation is not an award decision.
This boundary matters because middlemen love a single number. One number is easy to sell, easy to rank, and easy to hide behind. Builders need the harder thing: a record that can survive a superintendent’s challenge, a subcontractor’s objection, and an executive’s review.
Identify Exactly Who and What Was Evaluated
“ABC Electric — 72” is not an evaluation. It is a loose label.
A controlled contractor performance review should identify:
- Legal company name and relevant aliases.
- Project, location, subcontract, and contract identifier.
- Exact scope and any lower-tier scope involved.
- Evaluation start and end dates.
- Evaluator, reviewer, and approver roles.
- Criteria version and effective date.
- Evidence cutoff date.
- Interim, final, disputed, corrected, or superseded state.
The evaluated object must be narrow enough to understand. A subcontractor may perform strongly on a six-week interior fit-out and struggle on an 18-month phased renovation with occupied spaces. Rolling both into one permanent company grade strips away duration, complexity, crew, supervision, access, and project conditions.
The work has to be judged in its actual box—not in a platform vendor’s generic box.
Version the Rules Before You Add the Points
Every contractor performance evaluation form needs a defined scale. “Good,” “fair,” and “poor” sound plain until five evaluators apply five different meanings.
Store the criteria definitions, weights, critical items, effective dates, and policy or contract basis. If quality carries 30 points, safety 25, schedule 20, cost control 15, and administration 10, preserve that exact version. If the company later changes schedule to 25 and administration to 5, old evaluations must still show the old rules.
Do not quietly recalculate yesterday’s work using today’s formula. That is how records get rewritten without anyone admitting it.
Link Every Rating to Source Evidence
The scorecard should point to the records behind the observation, including their versions and status. Depending on the project, those may include:
- Commitments and subcontract requirements.
- RFIs and responses.
- Submittals and resubmittals.
- Inspection records.
- Punch items and closure evidence.
- Schedule activities and approved updates.
- Change records.
- Daily reports and meeting minutes.
- Closeout documents.
A raw count proves almost nothing by itself. Ten RFIs on a $4 million coordination-heavy mechanical scope may say less than two ignored RFIs on a $60,000 finish scope. Twenty punch items across 300 apartments are different from 20 punch items in five rooms.
For every negative or positive observation, capture scope, date, source, version, status, exposure, responsibility, and correction state. If responsibility is disputed, mark it disputed. Do not let software guess blame because a name appears on a document.
Separate the Fact, the Judgment, and the Total
A defensible evidence-linked contractor scorecard uses separate records for separate acts.
- Observation: what a source record shows.
- Context: scope, exposure, baseline, access, responsibility, and missing information.
- Judgment: the evaluator’s reasoned rating under a named criterion.
- Summary: a calculation based on the approved criteria version.
- Recommendation: one person’s proposed next step.
- Decision: an authorized action under the applicable authority matrix.
Suppose a daily report shows a crew arrived at 9:30 a.m. The fact is the timestamp. The context may be that building access started at 9:00, the delivery elevator was unavailable until 9:20, or the subcontract required a 7:00 start. The judgment comes only after that context is reviewed. The total comes later still.
Software companies make money by compressing all six layers into a dashboard tile. Apex Prometheus takes the opposite position: if a person cannot open the tile and replay how it was built, the tile is decoration, not operational control.
Use Interim Reviews to Fix Work, Not Ambush People
An interim review should create a usable checkpoint while correction is still possible. Identify the period, evidence cutoff, open questions, and required follow-up. Give the subcontractor the recorded issues and the chance to submit context or evidence.
The final evaluation should close a stated period only after review. It must not silently inherit every unverified note from the interim record. Open disputes should remain visible. Missing evidence should trigger abstention on the affected criterion rather than a made-up neutral score.
A simple state model can read:
draft → under review → issued for response → response received → approved final
A disputed record can branch to:
objection received → assigned review → decision recorded → corrected or upheld
That state control matters more than fancy graphics. It tells the crew, office, and executive exactly what is settled and what is not.
Correct the Record Without Erasing It
Corrections happen. A document was attached to the wrong subcontract. A schedule event was later reclassified. An evaluator missed an approved change. The answer is not to overwrite the old score and pretend it never existed.
Create a replacement evaluation with a stable ID. Link it to the original. Record the reason, reviewer, approver, effective date, and people or systems notified. Mark the prior version superseded while retaining it under access and retention controls.
That gives both sides a clean audit trail. It also blocks the oldest trick in the software middleman’s book: change the number, keep no receipt, and tell everyone to trust the screen.
Put Dollars on Bad Scoring Before It Costs Real Money
Consider a hypothetical $2.5 million interior package with an 8% expected gross margin. That is $200,000 of expected gross profit. A careless rating wrongly removes a capable bidder from a future bid list. The next qualified price is only 3% higher. That difference is $75,000—more than one-third of the expected gross profit before the next job even starts.
Now look at administration. Assume a project executive, PM, superintendent, and contract administrator each spend two hours reconstructing a disputed evaluation. At loaded rates of $175, $125, $110, and $85 per hour, one dispute consumes $990 before legal or executive escalation. Ten poorly documented disputes cost $9,900 in labor just to rebuild records that should have been linked on day one.
Those figures are scenarios, not promised savings. The point is simple: sloppy evaluation has a price. Evidence lineage is cheaper than archaeology.
Churchill Painting Corp serves as Apex Prometheus’s field proof model for how systems must respect real crews, scopes, and operating pressure. This article does not claim Churchill uses a released Apex scorecard or that any scorecard produced a measured result. The proof principle is narrower: build around the way work is actually recorded and reviewed, then test with controlled data before touching a real vendor decision.
Give AI a Wrench, Not the Foreman’s Authority
AI can help retrieve records, map evidence to criteria, cite versions, summarize long files, flag missing context, and draft an evaluation for review. It can also identify contradictions: an RFI marked late in one system but answered on time in another.
AI should not approve a rating. It should not infer responsibility from document counts. It should not decide corrective action, qualification, eligibility, suspension, bid access, award, personnel action, contract meaning, or legal position.
When evidence conflicts, identity is uncertain, or context is missing, the correct machine action is abstain and escalate. Authorized people keep the keys.
Start any proof work with synthetic packets: varied scopes, project durations, positive and negative evidence, missing records, objections, corrections, and replacement evaluations. Test whether reviewers agree, where they disagree, and whether the audit export can replay every result. Do not feed real subcontractor data into an unapproved workflow just because a software salesman says the model is ready.
Frequently Asked Questions
What belongs in a subcontractor performance scorecard?
Include company identity, project, subcontract, scope, evaluation period, criteria version, evaluator roles, source evidence, comments, rating scale, review state, subcontractor response, approval, and correction history. A total without those records is not an auditable evaluation.
How often should subcontractors be evaluated?
Follow the governing contract and company policy, then account for duration and risk. If interim and final reviews are used, state the schedule, evaluation period, and evidence cutoff in advance. A six-week package and an 18-month package do not need identical review timing.
Which records can support a contractor rating?
Commitments, RFIs, submittals, inspections, punch records, schedule events, changes, daily reports, administration records, and closeout documents may support a rating. Each record still needs version, status, scope, party, responsibility, exposure, and correction context.
Can a subcontractor dispute a score?
A controlled process should record receipt of the objection, the subcontractor’s comments and evidence, the assigned reviewer, the decision, and any replacement evaluation. Actual rights and deadlines depend on the governing contract, policy, and law.
Can AI decide whether a subcontractor gets future work?
No. AI can organize and cite evidence or draft material for review. Authorized people must retain rating approval, corrective-action, qualification, restriction, suspension, bidding, and award decisions.
A scorecard should make hard decisions more traceable, not make them automatic. Map the records, states, response rights, correction path, and authority before anyone starts handing out grades.
Come see what time it is — apexprometheus.ai