HIV Education and Prevention Project of Alameda County (HEPPAC)

What do they do? HEPPAC distributes sterile supplies and naloxone, provides overdose-prevention education, and helps people connect with treatment and other services in the East Bay. Its drug-checking program tests samples and shares results that may inform safer decisions. It also sponsors other harm-reduction programs, whose activities and donation routes need to be distinguished from HEPPAC’s own services. More

Why this approach interests us

• Accessible naloxone can allow a bystander to reverse an otherwise-fatal overdose when timely alternatives are unavailable.
• Sterile supplies and treatment navigation address distinct infection and care-access barriers alongside overdose prevention.
• A functioning local service platform, recent original filings and current checking reports make the work concrete enough to investigate.

Our main reservations

• Distribution, testing and reported reversals do not identify additional unique people or deaths averted by new donations.
• Repeat rescues, alternative supply, later mortality and overlap across services substantially affect the modeled health gain.
• Marginal delivery capacity and household money outcomes are unmeasured; the combined estimate depends on explicit, uncertain priors.

What do you get for your dollar? Our central conditional estimate is about $1.5M per DALY in the Bay Area, combining one modeled equivalent year, compared with $1.5M for the earlier health-only central scenario. Donations could support supplies, education, treatment connections and basic needs—not a verified purchase of a death averted. The model limits naloxone gains to unique victims with finite survival, then subtracts household travel and care costs from nonoverlapping savings and support. An illustrative weighted scenario gives $799K, compared with $813K health-only, but depends strongly on a favorable stress case. Neither is a measured donation return. More

What they do

HEPPAC’s harm-reduction services combine supplies, overdose education, drug checking and connections to care. The potential benefit is not the distribution count itself: naloxone must reach an otherwise-unrescued victim, a treatment connection must change sustained care, and sterile supplies must change infection risk. Basic-needs assistance and avoided missed work can also improve household resources, while travel, care and treatment burdens can offset them.

The direct donation route identifies HEPPAC, EIN 94-3205535, but offers separate buttons for Punks With Lunch and J&MP. Sponsor activity is included in the legal entity’s expense; it cannot automatically be assigned to an ordinary HEPPAC donation. The relevant question is what extra services and outcomes a donation enables beyond public funding, other providers and existing support.

Monitoring and information sharing

FY2025's Punks With Lunch description reports >10,000 people, ≥10,000 doses and >800 community-reported reversals. Identical numbers appear in FY2024. The appropriate interpretation is: “Sponsored-program activity is reported, but current deduplicated HEPPAC victims, reversals, annual reporting windows and marginal attribution remain unavailable.” No summing repeated descriptions, no treating doses as people, and no substituting 800 deaths averted. Sponsor expenses are in the legal-entity cost, but donation routing and shared delivery still prevent a matched sponsor benefit denominator.

HEPPAC's official EBDC index now publishes monthly reports through August 2026. The July infographic reports 50 drug samples tested, 42 results, plus 10 paraphernalia. The August original PDF reports 103 samples, 93 RaDAR results, plus 2 paraphernalia tested and 1 result. Current public checking operations are documented. The missing link is unique people, behavior after results, counterfactual consumption and health follow-up; a sample mismatch is not an overdose prevented. Do not annualize these months.

Current recipient is the direct HEPPAC donation route, EIN 94-3205535. The donation page distinguishes HEPPAC, PWL and Fresno-based J&MP buttons. Their attribution remains separate. Official OPEND URL changed to /opend/; NSB page retains brief training and unlimited refills, making repeat-person correction essential.

The county staff proposal sought $700,000 to maintain SUN staffing through August 31, 2026. That period has elapsed; no execution or successor award was identified. Successor-contract status remains unknown; an elapsed period does not establish private funding need. Keep multi-year contract ceilings separate from annual spending. DHCS free naloxone access supports a substantial alternative-supply counterfactual, with local personal access unknown.

Clinical evidence and native outcomes

No fresh evidence identifies replacements for the earlier model’s unique-person or survival coefficients. We retain the earlier health calculation as an explicit prior illustration rather than invent more favorable activity. Opportunity-equivalent annual totals 20/50/137 divided by 3/2/1.5 opportunities per selected person produce 6.67/25/91.33 synthetic unique victims. They are neither observed reversals nor kit recipients. One finite supported/counterfactual survival trajectory per victim covers one active year, with 5/10/20-year finite follow-up, utilities .55/.65/.75, 3% discount and supported hazards .10/.07/.04. Hazards are broad historical-scale judgments, not contemporary Bay measured mortality or decades of follow-up.

Otherwise-fatal probabilities .05/.10/.15 and naloxone attribution × outcome-additionality (.60×.25, .75×.50, .85×.75) remain explicit. Baseline rescue 0 is only the selected otherwise-unrescued component, not absence of EMS, other distributors or free naloxone. Within-person recurrence never buys another complete lifetime. Selected rescue hazard is added only to the counterfactual; remaining supported-arm all-cause hazard is not mislabeled as exclusive other-cause mortality.

MOUD units remain 10/25/50 synthetic starts, retained-effective shares .5/.6/.7 and bounded .25/.5/1.00 QALY per retained engagement. These composite lifetime-effect sensitivities have no current retained-treatment-year observations. The local navigation study covers five months and 30-day engagement, not annual HEPPAC starts or causal mortality. The 2026 NYC RCT found an imprecise null primary-outcome difference; it does not establish zero effect here, but prevents strong certainty from the observational association.

Syringe units remain .25/1/3 synthetic infections averted, with existing finite 3/5/8-QALY episode sensitivities and explicit causal credit. No new infection surveillance or local baseline transmission was found. Drug-checking fatal-event sensitivity remains visibly unresolved; current sample activity does not justify increasing it. Mortality overlap factors .8/.7/.6 remain coarse reductions, not linked unique-person estimates. The packet cannot establish a union of naloxone, MOUD and checking recipients. An independent test integrates the naloxone curves once; it does not certify other pathways' deduplication.

Current engine exposes 0–10-year common health delay (default legacy 0 retained), independent .01-QALY-per-credited-unique-person harm sensitivity, adverse rescue, short horizons and extreme repeat pressure. It never normalizes away negative outcomes. Early health timing and 20-year favorable prognosis remain material limitations rather than newly validated facts.

Qualitative assessment

HEPPAC’s low-barrier services may reach people whose access to conventional care is poor. Naloxone, sterile supplies and treatment are plausible routes to important benefits, while current reports show continuing service delivery. We are interested in the practical platform, not merely a low modeled ratio.

Our financial and source review does not establish a priced expansion opportunity. The latest returns describe a growing legal entity and several sponsored programs, but reported activity does not show how an additional donation changes unique-person outcomes. We would not treat the current estimate as sufficient to allocate a large donation without resolving that link.

The most useful next evidence would match a dated funding tranche to deduplicated recipients, retained treatment, other rescue options, residence and signed household-resource follow-up. Transparent null and adverse cases matter as much as favorable service counts.

What do you get for your dollar?

We retain full FY2025 legal-entity cost $4,953,186. Program-only expense or the next $100,000 cannot replace the full expense while retaining all portfolio outputs. Salary/staff expansion, fundraising, administrative costs, sponsor activity and unmodeled programs stay in cost. FY2025 return shows receivables and secured debt, so surplus/net assets do not prove unrestricted liquid room. Schedule O's $12,025 donated services require audited expense reconciliation before any gross-resource addition. Public naloxone, volunteer labor and donated supply valuation remain unknown; illustrative gross multipliers are not measured total resources.

For each scenario, annual cohort benefit is multiplied once by gift/full annual expense × deployable share × realization (.20×.70, .45×.85, .70×.95). The gift pays for a share of one annual cohort; health tails follow that cohort and consumption lasts only one year. There is no repeat annual gift-cost mismatch. The engine restricts gifts to at most annual expense; linear scaling is a conditional local illustration, not a claim of linear capacity.

Weights .24/.48/.08 plus .20 zero-credit remain unchanged and uncalibrated. The favorable stress world supplies most weighted gain. The probability of causal private-gift success, coverage and displacement visible; actual EV remains null because these are not identified. The no-benefit case has assumed zero health and money by construction; signed adverse-rescue sensitivity tests a separate negative world without claiming a newly calibrated probability.

These variations dominate the small income overlay. A favorable conditional price is not empirical cost-effectiveness. Required next evidence: dated next-gift budget/deployability, executed successor award, deduplicated victim and rescue linkage, retained MOUD treatment-years, cross-pathway linkage, recipient residence, baseline consumption and signed money follow-up, and matched sponsorship/public-resource accounting.

Health and net household resources

Recipient money outcomes are unknown, not measured zero. Add a distinct explicit one-year conditional cohort: 25/100/300 synthetic basic-needs/navigation recipients, with $6,000/$12,000/$18,000 annual baseline disposable consumption resources. These baselines include consumption financed by support; they are not zero cash wages or local GDP. They are chosen transparent sensitivity anchors for low-resource beneficiaries, with no HEPPAC income panel. Recipient counts are not added to health victim counts to manufacture a unique organization total.

Raw net = saved pay + medical OOP savings + transfer value × (1 − alternative displacement) − fees − travel − care. Modeled net = positive savings/transfers × independent share − full fees/travel/care. Overlap never scales negative costs, including in a positive-net row. The direct fee assumption is $0 because described frontline access is free; treatment/other fees remain unknown and a +$100 annual fee diagnostic is exposed. Central saved pay could mean 8 otherwise-lost work hours × $15 take-home, not lifetime survivor earnings; favorable 20 hours. Travel anchors could mean 24/12/6 visits × $5 and care is a separate opportunity/caregiver burden. Medical savings are modest household cash changes, not total hospital claims or insurer savings. A reduction in billed spending is not automatically a patient gain; payer savings receive no extra credit.

All money coefficients are priors. Zero entries specify that scenario, not empirical absence. Basic-needs value is consumption received, not donor expenditure; displaced alternative support reduces benefit. Funding costs remain in the numerator, so the same transfer is not counted as both payer savings and recipient income. Charity/government-paid services and insurance coverage may make medical household savings zero or treatment burdens negative. No program evidence establishes the chosen dollar amounts.

Apply independent shares .25/.50/.75 only to positive resource rows before the signed net to conservatively limit welfare already captured by health or overlapping services. The shared helper receives independentShare 1, then causal shares .10/.25/.50 reflect incremental service exposure/counterfactual access. These are separate uncertainty judgments, not estimated treatment effects. Negative cash costs remain fully present within that causally credited service cohort. Common recipient geography uses existing nested Bay shares .85/.95/.99 and SF .00/.02/.05, since residence observations remain missing. Direct East Bay delivery supports the direction, but fiscal sponsors outside Bay make an unqualified whole-entity share unsafe.

Use the shared incomeHealthyYearEquivalent helper with 1 year, .5-year delay and 3% discount. The mapping is $50,000 × people × discounted years × ln(1 + signed modeled net/baseline) / $100,000, then causal/geographic shares, following Coefficient Giving's normative comparison. No survivor-wages bonus, GDP, human-capital lifetime, or multi-year perpetuity. One-year annual income equivalents are −.037515436/.046100925/1.959835439; the downside remains negative. With all positive overlap credit removed, travel/care still yield negative money welfare; this is not the same as the separate income-disabled diagnostic.

Funding and previous grants

The official donation page distinguishes HEPPAC from its sponsored programs. County funding, free state naloxone, partner distribution and recipient insurance are relevant alternatives, not extra benefits to add to the donor model. The reviewed county proposal sought continued navigator staffing through August 31, 2026; no executed successor award was established in this review.

A recurring public contract ceiling is not annual expense or a private funding gap. An accounting surplus is not proof of deployable cash. A useful marginal offer would state staffing, full costs, recipient scope, alternative funding and measurable additional services.

Spending breakdown and financial appendix

The latest three available original IRS e-file return reconstructions are fiscal years ending June 2025, 2024 and 2023, not calendar 2025/2024/2023. The receipt JSON contains original URLs, object IDs, filed dates, source sections and values; direct XML downloads were blocked, but original rendered return bodies were successfully read through curl --compressed.

Source: original FY2025, FY2024, FY2023, Part I. Expense growth is real; it is not evidence of additional marginal outcomes.

Average annual legal-entity expense across these three June fiscal years is $3,524,429. FY2025 functional spending is $3,551,686 program services, $1,071,359 management and $330,141 fundraising, reconciling to $4,953,186. Within program services, disclosed amounts are $2,173,527 Alameda/Contra Costa activity, $526,494 EBME/OD2A, $322,029 drug checking, $264,885 NBHRN and $264,751 Punks With Lunch. These sponsor/program lines describe the reported expense mix, not donation additionality or clinical efficiency.

Revision record and model appendix

The complete earlier event-linear and unique-person engines, scenario outputs, option fingerprints and original-source hashes are preserved. Historical weighted Bay prices were $1.1M for the event-linear model and $813K after unique-person and survival revisions. The current combined weighted illustration is $799K; the central scenario is $1.5M. These are different scenario summaries, not alternative labels for the same estimate.

The current evaluator corrects inherited overlap treatment in adverse mortality scenarios: positive mortality channels receive the overlap reduction, while negative channels and independent harms remain fully signed. With adverse rescue, central annual health is -2.13 QALYs and the favorable-input case is −59.5038, rather than the old attenuated losses. Removing all positive overlap credit leaves harms intact. Health-only historical outputs remain inspectable but are not the corrected adverse current result.

Income-disabled diagnostics switch off the entire economic channel and must not be confused with zero positive credit, which still retains fees, travel and care. Combined welfare equivalents are not clinical QALYs. All priors, unquantified portfolio benefits and gross-resource limitations remain visible. Actual organization-specific author, modeling, source-review and report-writing intervals are separate from QA and deployment.