Methodology

Floor Intelligence model (floor-intel-v4l-20260928)

The Floor Health Score and Floor Life Index share one versioned model constant (FLOOR_INTEL_MODEL_VERSION). Worked examples are generated from the production scorer into scripts/fixtures/model-reproducibility-v1/; contract tests fail if scoring constants change without a version bump.

Inputs & defaults

Branches

• Thickness: 33/32" → max 10 sandings, 80-year lifespan cap (both internal planning assumptions); 25/32" → max 6 sandings (MFMA guidance, conditional on proper maintenance and skilled sanding), 62-year lifespan cap (internal planning assumption).
• Usage sand interval (internal planning assumption): heavy (multi-use, no cover) = 8 yr; medium (multi-use + cover) = 9 yr; light = 10 yr. The values sit inside MFMA's 8–12-year sanding example; the mapping to use level is Courtside's.
• Year-precision dates: legacy January 1 records anchor to July 1; recoat-overdue deduction capped at 15 points.
• Measured thickness (25/32" only): remaining sandings from a 0.578" minimum serviceable thickness (internal planning assumption derived from secondary research on NWFA wear-layer geometry; NWFA's guidelines are member publications and were not verified) and 0.03125" (1/32") removal per sand (MFMA: subsequent sandings should remove no more than 1/32").
• Unknown sand/service history (4.A): missing history is unknown evidence — overdue tiers apply only when a date is recorded. A floor with no sand on record is never counted as sand-due, and its next sand is shown as unknown rather than left blank or estimated.
• Floor-cover risk text (2.F, floor-intel-v4l-20260928): the no-cover risk no longer tells users to check "cover/finish compatibility guidance"; MFMA publishes no such guidance. It keeps MFMA's cure-time check and its caution against leaving covers down for long. No score, threshold or weight changed.
• Year last sanded without a recoat year (4.A, floor-intel-v4k-20260928): a year entered for the last sanding is evidence of that sanding only. It drives the sand-cycle tiers, but it is not evidence that nothing was done since, so it does not start the recoat-freshness clock; last service stays a missing input, confidence stays below high, and a planning range is shown. A recoat year or a crew-logged job restores the recoat tiers. A projected next-sand year already in the past is shown as overdue, never as a projection.

Deductions & caps

Rounding

• Health score: clamp to 0–100 after summing deductions.
• Sand due in months: max(0, round((interval − yearsSinceLastSand) × 12)).
• Measured sandings: floor((measured − 0.578) / 0.03125), capped by nominal remaining.
• FLI remaining years: max(0, min(pathA_years, maxLife − floorAge)).
• Planning range (thin evidence): center ±4 / +6 years on remaining life; health band 45–95.

Constants and behavior not previously documented on this page

Audited 2026-09-15 against app.jsx; the model block now lives in floor-life.js. These exist in the shipped model block (covered by the constants hash above) but were missing from this page before this update.

Rationale (floor-life.js, comment above FLOOR_HEALTH_CONDITION_RULES): "Structural/moisture observations survive a refinish; finish, wear, and process observations reset after the newest sand/install." Individual point weights per observation were approved by the owner in business-rules §1.1C (2026-07-09) as a scoring design; they are not calibrated to measured outcomes (see the decision record).

These weights are inert: FLOOR_HEALTH_PHOTO_SCORING_ENABLED = false (floor-life.js) keeps every photo deduction at 0 regardless of the table above. Eligibility gates it would need once enabled — severity ≥ 3, model confidence ≥ 0.85, capture quality ≥ 0.75, analysis confidence ≥ 0.75, at least 2 corroborating photos (floor-life.js) — approved by the owner in business-rules §1.1D (2026-07-15); scoring stays off until it is enabled deliberately (see the decision record).

Sand-cycle urgency ratio thresholds (computeFloorIntel in floor-life.js): the table above names the tiers; the exact ratio is yearsSinceLastSand / sandInterval against >1.0 (overdue, −20), >0.8 (due soon, −12), >0.6 (approaching, −5). Owner-approved baseline (business-rules §1.1C, 2026-07-09); a scoring design, not a published standard.

Recoat-freshness day thresholds (computeFloorIntel in floor-life.js): exactly daysSinceService > 365 / > 270 / > 180, matching the ">1 yr / >9 mo / >6 mo" prose above. No separate rationale beyond the deduction comment.

Confirmed-service precedence (computeFloorIntel in floor-life.js, business-rules §1.1C/§1.1E): when a non-superseded confirmed-service event is on file, the effective "days since service" used for the recoat-freshness deduction is the NEWER of the contractor-reported job date and the confirmed event date — recency can only improve, never worsen, the score. This is production behavior that materially changes deduction inputs and was not previously named on this page.

Methodology classes

Every numeric floor-intelligence rule is mapped to exactly one class in the committed registry (scripts/floor-intel-methodology-registry.js). Contract tests fail if a nonzero numeric literal or numeric-only string in the model's scoring block is not registered, or is registered twice. The literal zero and numbers inside descriptive text are not checked.

• Outside guidance — a figure a publisher states on a public page, cited below with its scope: MFMA's up to six sandings for 25/32" maple and its 1/32" per-sanding removal limit. Nothing else in the model is outside guidance.
• User fact — values supplied by the facility or contractor (year installed, times sanded, thickness, use, cover, HVAC, measured thickness, job history).
• Planning assumption — Courtside-owned lifecycle bounds: the 62 / 80-year lifespan caps, the 33/32" sanding count, the usage sanding intervals, the 0.578" minimum serviceable thickness, and the thin-evidence planning ranges.
• Chosen weight — Courtside-chosen scoring weights and thresholds: every deduction, bucket cap, day and ratio threshold, photo-evidence gate, risk trigger, and health-state band. These are product scoring choices with a recorded owner decision (below), not published standards, and not calibrated to measured floor outcomes.
• Calculated — outputs derived from the rules above (health score, remaining years, projected sand and replacement years, measured sandings). A calculated output has a formula, not a chosen value.

Decision record for planning assumptions and chosen weights

Every planning assumption, chosen weight, and publisher figure the model uses as a value has one recorded decision: its status, accountable owner, where it was recorded, and why. Accepted means the owner approved it as a Courtside scoring or planning choice, not that it is a published standard or calibrated to measured floor outcomes. Accepted-interim means a later decision replaces it once implemented. Pending means an open question for the owner; the value stays in the model, labeled a planning assumption, until it is answered. Generated from scripts/floor-intel-methodology-registry.js.

Worked example: one floor, start to finish

This is the pinned reproducibility scenario sand-cycle-overdue, computed on 2025-08-01 by the same code the app runs. The test suite recomputes it on every build and fails if any number below drifts.

Inputs. Installed 2000, 25/32" maple, sanded 3 times, last sand & refinish 2014-07-01, multi-use, no floor cover, HVAC year-round.

1. Floor age: 2025 − 2000 = 25 years. Planning life cap for 25/32": 62 years (a planning assumption; see the decision record).
2. Service recency: 4049 days since the last recorded service, over the 365-day threshold with an exact date, so −30.
3. Sand cycle: about 11.1 years since the last sand against an 8-year planning interval (multi-use with no cover; 9 with a cover, 10 otherwise), a ratio above 1.0, so −20.
4. Risk: multi-use floor with no cover, so −8.

Score. 100 − 30 − 20 − 8 = 42, clamped to 0–100. Phase: sand_due. Every deduction and interval above is a Courtside chosen weight or planning assumption; its owner, date and reason are in the decision record.

Sources for lifespan and sanding figures

Public pages checked on September 22, 2026 and re-checked on September 28, 2026. No member-only or purchased publication was used.

• MFMA — Life Cycle of Flooring: a properly maintained 25/32" maple floor, sanded by a skilled contractor, can withstand up to six sandings; with sandings every 8 to 12 years the typical lifespan can exceed 75 years. Scope: conditional on maintenance, workmanship, and no environmental damage or abuse.
• MFMA — Open Letter on 33/32" Flooring: it is not uncommon for a 33/32" floor to last more than 100 years. Scope: no 33/32" sanding count and no absolute maximum is stated.
• MFMA — Sanding & Sealing: subsequent sandings by experienced contractors should remove no more than 1/32". Scope: a per-sanding limit; the model treats each sanding as removing the full 1/32".
• MFMA — Annual Maintenance/Refinishing: most gymnasium floors should be recoated annually. Courtside's 6 / 9 / 12-month recoat thresholds and their deductions are its own chosen weights, not an MFMA rule.
• MFMA — Floor Surface Protection: floor covers are not always needed in multipurpose rooms but offer extra protection. The no-cover deduction is a Courtside chosen weight, not an MFMA requirement.
• MFMA — The Use of 33/32 Material (archived 2019 capture): a properly maintained 25/32" floor "will provide four to six additional sandings after installation" with "a life span of at least 30-40 years or more"; a 33/32" floor "will provide eight to ten normal sandings and has a life span of 60 years or more". Scope: no longer on maplefloor.org; lower bounds, not maximums.
• Hardwood Floors magazine — How Many Times Can a Wood Floor Be Sanded? (April 2025): "A traditional sand job will remove 1/32” to 1/16” of actual material", and most solid wood floors "can be sanded six to eight times". Scope: a general hardwood article, not maple-gym specific.

Not supported by any public source checked: a 62-year or 80-year maximum life, a 10-sanding limit for 33/32" maple (the archived MFMA statement above gave "eight to ten normal sandings" as a typical count, not a limit), a mapping of use levels to 8 / 9 / 10-year sanding intervals, or an "MFMA OM-2" publication. The NWFA Technical Guidelines index points to member publications and supplies none of these numbers. These figures stay in the model as Courtside planning assumptions; the decision record gives each one's status and owner, including the two still open.

Floor Health Score

The Floor Health Score is a 0–100 maintenance decision-support score when required installation history is present. It applies transparent deductions for service timing, sand-cycle urgency, remaining sandings, documented risk factors, and structured contractor observations. A floor without an installation year is shown as incomplete rather than assigned a score.

• Contractor evidence: each structured condition is counted once. Structural and moisture observations remain relevant across the floor's history; finish-failure, wear, and process observations reset after the latest sand-and-refinish or installation. This bucket is capped at a 20-point deduction.
• Date precision: legacy dates stored as January 1 are treated as year-only records and anchored to July 1 for elapsed-time math. Their recoat-freshness deduction cannot exceed 15 points.
• Photo documentation: photos, stage tags, and AI findings are reviewable documentation only in the current product. They do not change the Floor Health score or product score.

Photo findings can carry purpose and stage context for review. They remain neutral to scoring in the current app, including when an analysis is pending, failed, stale, or unavailable.

Documentation boundary: a photo can preserve context for human review, but it cannot establish moisture source, active decay, repair scope, or structural safety. Contractor notes remain visible provenance and are not keyword-scored.