Nineteen Years With One Organisation

Permanence is audited as a carbon-stock question. Buffer pools, reversal risk, monitoring periods, the arithmetic of how much carbon might come back out of a forest and how much a registry should withhold against that possibility. All of it is necessary. None of it is where the risk actually sits.

What fails first is rarely the trees. It is the organisation that was meant to keep them standing: the community body that signed the agreement, holds the membership register, runs the nursery, and is supposed to distribute the money in year eleven. Forest growth is reasonably predictable over four decades. Institutions are not.

A forty-year commitment is signed by an organisation. The question is not only whether the carbon stays in the ground. It is whether anyone is still answering the phone in year nineteen.

This is a piece about procedure, not about virtue, and not about outcomes. The argument has three parts: the variables that predict whether benefit-sharing actually arrives are institutional; they are observable, but only across a decade; and almost nothing in a standard pre-investment process is positioned to see them. We make no claim here about delivered results at any site.

What the standards actually measure

Start with the primary text, because the market's own instrument is more candid than the commentary about it. Verra's Procedure for Applying the AFOLU Non-Permanence Risk Tool, v5.0 (issued 16 December 2025 under VCS Version 5) requires potential transient and permanent losses in carbon stocks to be assessed "over a period of 100 years." Against that century, the internal risk rating is the sum of four sub-scores: project management, financial viability, opportunity cost and project longevity.

Read what the project-management score is made of. Two points if the management team does not include individuals with at least five years' experience in every area of required skill. Two points if the team "does not maintain a presence in the country," or sits more than "one day of travel from the project site." Minus two if it includes people with more than five years of AFOLU design, carbon accounting and reporting experience. A project fails outright if it has no adaptive management plan with a monitoring plan attached.

These are institutional proxies, and they are sensible ones. They are also almost entirely about the developer, assessed as a snapshot, mostly from documents.

The community side is thinner still. Stakeholder engagement scores zero — no risk at all — once more than 50 per cent of stakeholders living within the project area and reliant on it have been consulted, and more than 20 per cent of those within 20 kilometres of the boundary. Below either threshold the score jumps to 10 and 5 respectively. Consultation is therefore a binary gate crossed at a headcount, not a relationship measured over time. In the opportunity-cost test, where baseline activities are subsistence-driven, the impact of carbon revenues on communities counts only "where the communities have designed and agreed on a benefit-sharing plan." A plan. Not a payment record.

Political risk, meanwhile, is drawn from the mean of the six Worldwide Governance Indicators over the most recent five years of available data — a national average standing in for a question that is answered at the level of a barangay, a parish or a single family's slope. Land tenure is a gate before it is a score: a project fails outright without a binding legal agreement securing the right to operate with every entity holding verified ownership or resource-access rights, and scores 10 where the government has expropriated 10 per cent or more of the project area in the past twenty years.

The arithmetic is unforgiving at the edges: the minimum overall risk rating is 12 regardless of what the tables produce, anything above 60 fails, and the external-risk category fails on its own above 20. Project longevity is its own gate: for projects requesting registration on or after 1 January 2024, longevity under 40 years fails the risk assessment; for earlier requests the fail threshold is 30 years; and the tool notes that anything under 40 years is "ineligible for the Core Carbon Principles label."

So the instrument that converts a century of institutional uncertainty into a buffer percentage reads, at the institutional margin, like a document checklist with a headcount threshold. That is not a criticism of the drafters. It is a description of what is available to them at validation.

The standards know this, and have just moved

The direction of travel is unambiguous, and it is toward the institutional question. Verra's overview of VCS Version 5 updates (released 16 December 2025, last updated 10 September 2026) clarifies that meaningful stakeholder engagement "must occur before the project start date"; introduces a stakeholder engagement plan and a stakeholder input tracker; strengthens grievance redress requirements, explicitly requiring the mechanism to be "available to stakeholders at the earliest stages of project development"; requires a written project implementation agreement carrying the terms of the benefit-sharing mechanism; requires proponents to give benefit-sharing participants financial information on the project's revenues and costs; and requires proponents to ensure ongoing consent, not consent obtained once. Most of it bites for projects with a start date on or after 1 January 2027, and for existing projects at the next baseline reassessment submitted on or after 1 January 2030.

The Integrity Council has been equally direct. Its Continuous Improvement Work Program report on Permanence (May 2025) records that all CCP-eligible crediting programmes have now aligned permanence monitoring thresholds to 40 years, and makes six recommendations. Two of them matter here. Recommendation 4 asks the Council to provide guidance on "the types of risks addressed and acceptable data sources used in project-level risk assessments" — an acknowledgement that the inputs to the risk score are not yet trustworthy or comparable. Recommendation 2 asks that cessation of monitoring and verification trigger a compensation liability equal to everything the project previously contributed to the buffer reserve.

Sit with that second one. Monitoring does not cease because a forest stops growing. It ceases because an organisation stops functioning — it runs out of money, loses its principal, splits, or simply stops filing. Recommendation 2 is an institutional-failure provision in all but name. The market has correctly identified the failure mode and has responded by pricing the liability, which is the right move and still leaves the underwriting problem untouched: nobody has the data to tell a durable organisation from a fragile one at the point of decision.

Nineteen years, compressed

We have one long observation, and it is worth being precise about what it can and cannot support. Our relationship with a single community-based organisation in Mukono district, Uganda, runs from 2007 to now — nineteen years.

It starts in the aftermath of a vanilla price collapse, and the published record of those years is instructive on its own terms. Vanilla had spiked in the mid-2000s and then fell away. Uganda's Daily Monitor reported that neighbouring Buikwe District — carved out of Mukono in 2009 — had about 9,000 vanilla farmers before 2005 and "not more than 2,000" afterwards, growers having abandoned the crop "when its prices tumbled"; a green-bean price that had reached Shs150,000 a kilogramme "nose-dived to almost Shs1,000."

Coffee over those same years moved the other way in the published series, and that is the part worth sitting with. The International Coffee Organization's Annual Review 2007/08 puts the Robustas group indicator at 61.45 US cents per lb for coffee year 2005/06, 82.73 for 2006/07 and 106.36 for 2007/08 — a rise of 28.6 per cent in the final year. So one crop was emptying out at district level while the other climbed on the international indicator — and the indicator is the instrument a desk screen actually reads. What a grower was offered at the farm gate in those years, and whether a trader withdrew or a buyer did not come back, is in neither record. No screening process built on published prices resolves it, then or now.

What we have since is ordinary and slow. A coffee grower and trader survey across the same districts in 2013–14, which asked — among other things, and without appreciating at the time that it was the important one — who a grower actually consults before changing what they plant. Farm-level coffee cost and income data from another region entirely, across the years a disease epidemic ran through it. And a vanilla agroforestry pilot that became a trading business, with that same organisation's director now sitting on our advisory board. The instruments are described where they belong, in the crop-switching piece and on the founder page; what matters here is only when they were collected.

What that proves, statistically, is very little. It is one organisation, not evidence that any method produces durable institutions, and not an endorsement by anyone of anything. What it does show is the shape of the variable. Everything we think we know about whether that organisation would keep faith with a benefit-sharing schedule was learned in years when there was no carbon project, no credit, no registry and no reason to perform for an auditor. Those are the years a due-diligence process is never present for.

The academic literature points the same way. Cox, Arnold and Villamayor-Tomás's review of Ostrom's design principles (Ecology and Society 15(4):38, 2010) analysed 91 studies covering 77 cases of common-pool resource institutions, and found the principles broadly supported — monitoring, graduated sanctions, conflict-resolution mechanisms, nested enterprises — while concluding that "a probabilistic, rather than deterministic" interpretation is warranted. Institutional durability is predictable in the aggregate and uncertain in the individual case. That is exactly how a buffer pool is supposed to work, and exactly why the inputs matter more than the formula.

The variables that predict whether benefit-sharing arrives

Below is the list we work from. It is not a scoring system and we are not proposing it as one. The third column is the point: most of these cannot be read off a site visit, and the ones that can be are the weakest predictors.

Variable	Window needed	Visible on a diligence visit?
Behaviour through a price collapse in the members' main crop	A full commodity cycle, 7–15 years	No
Survival of a leadership succession	A decade, or one founder's tenure	No
Whether money historically reached households or stopped at the centre	Several disbursement cycles	Only from ledgers that predate the project
Records kept in years when no funder required them	Continuous	Partly — ask unannounced for five years of minutes
Membership churn against a register kept for another purpose	Annual, over years	No — a roster is not a register
Advice networks: who growers actually consult before changing a crop	One survey round; better repeated	Yes, if you survey rather than ask the committee
Grievances raised, and what happened to them	Continuous	Only if a mechanism predates the project
Dependence on a single donor or buyer line	One funding cycle	Partly, from accounts
Whether the organisation has ever declined an offer on its members' behalf	Episodic	Rarely, and never from the party asking

Note which way the incentives run on that last row. An organisation that has never said no to anybody is not demonstrating alignment. It is demonstrating that it has not yet had to choose between its members and a counterparty, which is the choice a benefit-sharing schedule eventually forces.

What to do when you do not have nineteen years

Nobody gets to start observing retrospectively. The practical question is what a pre-investment process can do inside the time it actually has. Six things, in order of how much they tell you per unit of effort:

• Ask for the failure, not the record. What stopped in the last downturn, who decided to stop it, and what happened to the people who had been paid to run it. An organisation that can narrate its own last failure precisely is telling you it has institutional memory. One that cannot is telling you something too.
• Read documents produced when nobody was funding them. Minutes from a year with no project, a ledger from a season with no buyer, a register maintained for a school or a savings group. Documents written for an auditor prove only that an auditor was expected.
• Reconcile the register against a list kept for another purpose. Compare the membership roster to a savings group's book, a church list, a cooperative's delivery records. Divergence is not fraud; it is the size of the gap between the institution on paper and the institution in use, and you need that number before you design a benefit-sharing mechanism on top of it.
• Separate consent from consultation, and find who was excluded. The 50 per cent threshold in the risk tool is a floor, not a finding. The useful question is which households are reliant on the area and were not in the room, and whether the reason is geography, gender, tenure status or a live dispute. Under VCS Version 5, consent is something to be maintained, not collected, which makes exclusion a compounding liability rather than a documentation gap.
• Use the grievance channel before you rely on it. Put a real, low-stakes complaint through it and watch what happens. A mechanism that has never been used has not been tested, and one that routes every grievance to the person who caused it is worse than none.
• Price the succession explicitly. Name who signs in year ten. If the answer is one individual with no deputy, no documented process and no second signatory, that is the single largest non-permanence risk in the file, and it will not appear anywhere in the risk score.

Then do the thing that only works if you start early: put the institutional observation on a schedule that outlives the feasibility contract. Short, annual, cheap, written down even in the years when nothing happens — particularly in those years. A nineteen-year record is not an achievement. It is an artefact of somebody having started a file in year one and not thrown it away. The reason so few exist is that the stage before a project exists has no budget holder, which is a separate problem, and a real one.

This is a pricing question, not a values question

None of the above is an argument for being nicer to communities. It is an argument about mispricing.

The buffer contribution a project makes is a direct function of its risk score, and a material part of that score is institutional. If institutional variables are assessed from documents at a point in time, the buffer is wrong in both directions at once. An organisation that kept its books through two commodity collapses contributes at roughly the same rate as one incorporated eighteen months ago with the same paperwork and a well-drafted adaptive management plan. The first subsidises the second, and any buyer paying for durability is paying a blended price across a pool whose risk weights do not distinguish the thing that actually drives reversal.

That is why the Integrity Council's recommendation to pilot stress testing of pooled buffer reserves is more significant than it sounds. Stress testing a reserve only helps if you can characterise what is correlated inside it. Institutional failure is correlated — by country, by commodity, by donor cycle, by the same three funders withdrawing from the same region in the same year. Carbon-stock risk from fire and storm is modelled with real physical data. Institutional risk is modelled, at present, with a checklist.

The same gap runs through the demand side. Durability requirements are tightening on purchasers as well as projects, and the mechanisms named for making shorter-lived nature removals credible are institutional ones. The SBTi Corporate Net-Zero Standard V2.0 (June 2026) says, at footnote 75, that it intends to issue a Call for Evidence on whether shorter-lived removals can deliver climate-equivalent permanence to long-lived removals and neutralise long-lived GHGs "through contractual, financial, or stewardship mechanisms." We read that standard in full elsewhere. A contractual mechanism is only as durable as the counterparty holding the contract, and in a community forestry project that counterparty is a village-level organisation with a chairperson, a bank account and a succession problem.

What we are actually claiming

That the predictable part of permanence failure is institutional. That the predictors are observable but slow, and that a process which begins at feasibility and ends at validation cannot observe them. That the correct response is not better adjectives in a risk narrative but an earlier start date for the observation, and honesty in the file about which variables are measured and which are inferred.

And that the useful unit of account is not the hectare or the tonne. It is the number of years you have been watching the same organisation, and whether you were still watching in the year the price fell.