FlameClear — AI-Native AIM ER&R Fire Suppression Compliance Pack Engine

1 Thesis

US companies that first-fill, service/recharge, or recycle HFC fire suppressants now operate under live ER&R obligations at 40 CFR 84.110: recycled HFCs required for servicing since January 1, 2026; technician training due by June 1, 2026 (or within 30 days of hire); and the first HAWK Fire Suppression Annual Report due February 14, 2027 covering CY2026. Verified

FlameClear sells a Fire Suppression ER&R Compliance Pack: intake of service tickets, cylinder fill logs, purchase orders, inventory snapshots, and training records → AI extraction/normalization against 84.110(g) report fields → recycled-vs-virgin and waste-destination gap matrix → training attestation vault → monthly capture SOPs → HAWK Excel assembly → specialist QA → customer Authorized Official files via CDX/HAWK. Humans own applicability edge cases (mission-critical military / aerospace carve-outs), virgin-vs-recycled classification disputes, and submission attestation. The customer experiences a compliance outcome and evidence vault, not a self-serve co-pilot.

2 Discovery rationale

Terrain scanned this hour: AIM ER&R fire-suppression reporting (open in memory), Form LM-20/LM-21 consultant filings, California SB 1103 commercial-lease QCT packs, AB 2013 GenAI training-data disclosures, TSCA §12(b) export notifications, EPA MSGP annual-report/NOI renewal, DEA ARCOS, BEA FDI (already claimed), FCC Form 499 (telecom claimed), CSDDD/France duty of vigilance (delayed/enterprise-heavy), and NAIC RBC (software-crowded). Decisive evidence for FlameClear: (1) hard statutory deadlines already live or <8 months away; (2) FSSA industry association actively publishing AIM Act training courses and member alerts — visible buyer-community demand; (3) EPA published a dedicated February 2026 fire-suppression fact sheet, June 2026 training overview, HAWK Excel form, and webinar screenshots; (4) adjacent budget exists via Weaver CPA AUP audits for HFC recyclers/allocators and fire-protection consultants at ~$190+/hr; (5) anti-duplication vs LeakClear (facility O&O leak forms) and TransitionDesk (RACHP/foam/aerosol TT product reports) — different buyer, CFR section, form, and workflow.

3 Candidate comparison

4 Hard disqualifier check

5 Rubric scorecard

Average ≈ 4.3 / 5.

6 Opportunity

7 Evidence quality & source-claim matrix

8 Why now

• Verified regulatory clock: Recycled-agent and venting rules are live; training deadline just passed/ongoing for new hires; first annual report is <8 months away with CY2026 data already accruing. Verified
• Verified industry mobilization: FSSA launched an AIM Act technician training course (Mar 2026) and has published repeated AIM/ER&R member communications since 2022. Verified
• Verified tooling readiness: EPA published the HFC Fire Suppression Annual Report Excel and HAWK registration path; leak-report user guide exists; FS user guide still pending — creating a service gap. Verified
• Inferred AI feasibility: Service tickets and cylinder logs are document-heavy and structured enough for OCR + classification with specialist QA. Inferred
• Unverified: Magnitude of mid-market non-compliance / data-gap prevalence outside FSSA core. Unverified

9 Customer & PMF

10 The outcome we sell

First one-feature MVP wedge

• ICP: Regional HFC fire-suppression distributor/installer (1–3 shops) that recharges cylinders
• Trigger event: Mid-year 2026 realization that fill/service logs do not distinguish recycled vs virgin mass for HAWK
• Pain: CY2026 data already accruing; Feb 14 2027 report will fail completeness without retrofit capture
• One-feature MVP: Mid-Year CY2026 Readiness Pack (gap matrix + capture SOP + training vault status)
• Input: 90 days of service tickets, fill logs, agent POs/invoices, inventory snapshot, training roster
• Output: Field-mapped gap matrix + remediation checklist + standing monthly capture workbook + training exception list
• Human chokepoint: Fire-suppression compliance specialist validates recycled/virgin classifications and applicability
• Success metric: ≥80% of ticket volume classifiable; customer adopts monthly capture within 14 days
• What users ask next: Full-year HAWK filing package; multi-branch rollup; Subpart A↔C recycler crosswalk; supplier attestation automation

11 Internal AI engine architecture

1. Intake layer: Secure upload of PDFs/CSVs/photos of tickets, cylinder tags, COAs, POs, inventory exports, training certificates; optional email forward mailbox.
2. Normalization layer: OCR, agent name→CAS/ASHRAE alias map, unit conversion (lb/kg), entity/branch identity resolution, dedupe of ticket IDs.
3. Retrieval & knowledge layer: Indexed 40 CFR 84.110 / 84.31(j), EPA fact sheets, HAWK form schema, SNAP fire-suppression substitutes list, prior gold-standard packs.
4. AI workbench: Extract quantities, classify transaction type (first-fill / service / recycle / waste), draft virgin vs recycled labels from supplier docs, draft training roster status, draft HAWK rows.
5. Deterministic rules: Required-field completeness; mass-balance tolerances; date windows; carve-out flags; Subpart A overlap explanation templates; retention clocks (3 yr / 5 yr).
6. Human chokepoint: Specialist approves classification exceptions, applicability, and submission package.
7. QA layer: Second-pass sampling, mass-balance red-team, screenshot/source-link audit.
8. Delivery layer: Pack PDF + Excel + vault index; AO runbook; optional assisted HAWK upload session.
9. Learning loop: Correction → rule/prompt/taxonomy update; rejected tickets → intake checklist change.
10. Model-portability: Provider-agnostic extraction prompts; schema versioning; eval set of anonymized tickets.

12 AI-vs-human operations pipeline

13 Operations as product

• Structured intake checklist (ticket fields, agent SKU, mass, recycled attestation, tech ID, branch).
• Required evidence lists for first-fill, service, recycle, waste, training.
• Automated completeness gates before specialist review.
• Exception queues: unknown agent, missing recycled COA, mass imbalance > tolerance, carve-out candidates.
• Reviewer assignment by entity complexity / recycler vs installer.
• Confidence scoring on each HAWK row; low-confidence auto-escalates.
• Audit trails + versioned packs; gold-standard anonymized examples; red-team mass-balance checks.
• Postmortem on any EPA rejection, customer rework, or missed training clock.

14 No-holes quality engine

• No HAWK row without source document ID.
• Deterministic rejection of virgin-in-service rows lacking approved exception code.
• Training clock calculator (hire date → due date) with overdue highlighting.
• Subpart A overlap check for recyclers to prevent double-count confusion.
• Dual control: specialist QA memo + customer AO certification.
• Never auto-file without AO; never invent masses; never silently impute recycled status.

15 Pricing, legality & unit economics

Why not hourly: Buyers pay for a filed/ready pack and closed gaps, not hours. Hourly recreates commodity environmental consulting.

Pricing legality: Fixed fees for documentation/reporting support are standard. No contingency on “avoided penalties.” Not legal advice; not PE practice; customer remains the regulated entity and AO.

COGS per Mid-Year Pack (launch estimate)

Revenue/FTE target: $350k–$500k once monthly capture is productized. Automation: ~55% launch → ~75% day 90 → ~85% year 1 on extraction/mapping (not on AO certification).

16 Nonlinear scaling plan

• Throughput target: 2–3 Mid-Year Packs / specialist / week at launch; 5–6 after templates harden.
• Cycle time: Diagnostic 5–7 business days; Mid-Year Pack 10–15; Annual Filing 7–10 after capture year.
• Rework <8%; quality failure <2%; escalation <15% of tickets.
• Margin expands by converting repeated classification fixes into taxonomy rules and by moving customers onto monthly retainers before Q4 crunch.
• Do not scale by hiring field techs — field work stays with clients.

17 Moat & Sam Altman test

Frontier models make OCR and drafting cheaper — strengthening FlameClear’s unit economics. Defensibility comes from (1) living agent/transaction taxonomy, (2) Subpart A↔C crosswalk playbooks for recyclers, (3) gold-standard packs tied to HAWK schema versions, (4) distribution via FSSA/manufacturer channels, and (5) accumulated exception libraries. Commoditization threat: EPA ships a perfect self-serve wizard and manufacturers embed free compliance portals. Counter: mid-market multi-branch messiness + training vault + monthly capture still favor a DFY operator; partner with OEMs rather than compete on software UI.

18 Buyer-specific go-to-market

Primary motion: Association + manufacturer-channel outbound (not consumer waitlist-first). Secondary: diagnostic lead magnet + founder LinkedIn/content.

Distribution proof table

First 30 days of content

• 10 posts: recycled-for-service rule; training clock; HAWK due date; Subpart A vs C; virgin inventory traps; waste destination fields; carve-outs; disposable cylinders; mass-balance tips; AO certification pitfalls.
• 3 teardowns: anonymized ticket lacking recycled COA; training roster with hire-date misses; recycler quarterly vs annual field overlap.
• 2 lead magnets: 84.110(g) field checklist PDF; Mid-Year Capture Workbook.
• 1 webinar: “Dry-run your Feb 2027 Fire Suppression HAWK report in 45 minutes.”
• 1 outbound diagnosis template: “We reviewed N public/service patterns — here are the 5 fields your logs likely cannot populate yet.”

19 Pilot design & early-demand trap

• Pilot cap: 8 entities (max 12 branches total).
• Profile: Distributor-installers first; max 2 recyclers (higher Subpart A complexity).
• Incentive: 30% off Mid-Year Pack for published anonymized lessons-learned rights.
• Feedback cadence: Weekly exception review; product vs custom work logged every ticket.
• Hardening gates: After 5 → freeze intake schema; after 10 → freeze SOPs/QA; at 20 cumulative → pause until COGS/rework/cycle measured.
• Kill criteria: Cannot classify ≥70% tickets without custom engineering; buyers will only pay <$2k; EPA changes eliminate annual report; field labor requested as core SKU.

20 Competitive landscape

Budget validation: Buyers already pay consultants (~$190+/hr), association memberships/training, and (for recyclers) CPA AUPs. FlameClear redirects that spend into a fixed-fee DFY pack focused on 84.110 readiness.

21 Regulation, compliance & licensing boundary

• AI may: Extract, classify, draft HAWK rows, flag gaps, assemble evidence indexes, draft SOPs.
• Operators may: Review classifications, run QA, coach AO on submission mechanics.
• Licensed professionals: Not required to file 84.110 reports; CPA AUPs remain separate for Allocation Program recyclers when applicable. PE not required for reporting packs. Counsel for legal interpretations.
• Must not claim: Legal advice; guaranteed penalty avoidance; SNAP listing decisions; engineering design approval; performing regulated field recovery as FlameClear employees.
• Controls: MSA + scope exclusions; AO certification acknowledgment; audit logs; no contingency-on-penalty pricing.

22 Compact founding team

23 Exhaustive risk register

24 Tech stack & build plan

• Day 0: Notion/Airtable intake + Drive vault + Claude/GPT extraction scripts + Excel HAWK templates + DocuSign AO checklist.
• Day 30: Ticket OCR pipeline, agent taxonomy DB, exception queue UI (simple Next.js internal).
• Day 90: Monthly capture API/email inbox, confidence scoring, partner portal.
• Avoid building customer-facing co-pilot; keep UI internal.

25 Metrics & KPIs

• Throughput packs/week; cycle time; rework %; GM %; COGS/pack; revenue/FTE
• Escalation %; automation %; evidence completeness %; quality failure %
• Customer acceptance %; diagnostic→pack %; pack→retainer %; on-time HAWK filing %

26 What could kill this

• EPA eliminates or indefinitely delays fire-suppression annual reporting.
• Buyers refuse >$3k and DIY Excel proves “good enough.”
• Major OEMs ship free dealer compliance that includes ticket capture + HAWK export.
• Cannot get clean recycled-vs-virgin evidence at scale → liability refusal.
• Scope creeps into field recovery/physical logistics, destroying margin model.

27 90-day validation & launch plan

7-day

• Finalize MSA/exclusions; build field map; recruit 15 outreach targets; publish checklist lead magnet.

30-day

• Close 5 diagnostics; convert ≥2 Mid-Year Packs; run FSSA-adjacent webinar; measure classification exception rate; counsel review of claims.

90-day

• 8 pilot caps filled or waitlist; monthly retainer live; COGS instrumented; kill/continue decision on ASP, exception rate, and census.

Human expert task table — What the human expert actually does