Hoverview

Donor

One programme, four donor frameworks

Four funders, four templates, one set of results underneath. What the templates disagree about is rendering, and most teams still resolve it by retyping.

Write an indicator carefully enough and it will fail a compliance check somewhere. Take a plain one: the number of households receiving a cash transfer, disaggregated by sex of head of household, rising from 0 to 4,200 by March. Under the UK guidance that has governed logframes since 2011, that string is wrong, because the target does not belong inside the indicator. The guidance is unusually direct about it: there is no such thing as a SMART indicator. Under Defra's Biodiversity Challenge Funds, the same string is exactly right, because the column it goes in is headed "SMART Indicators (including disaggregated targets)".

Same programme, same delivery, same number. One template treats the sentence as a definitional error. The other treats it as the model answer.

That is not a disagreement about the work. Nobody at either funder holds a different view about what a cash transfer is or how you count a household. It is a disagreement about how the same underlying fact should be laid out on a page, and the reason it costs anything at all is that most teams keep their results logic inside whichever page the biggest donor asked for.

Four renderings of one output

The scale of the divergence is easier to see with one output pushed through four rulebooks.

Under the UK's 2011 frame, activities leave the matrix. They go to a separate sheet or a Gantt, so the activity rows that carry most of the team's actual planning have no home in the document the donor reads. One outcome, no more. Three indicators per output, maximum. Assumptions may be recorded at outcome and output level only, so an assumption written against an activity is not moved, it is dropped. Every output carries an impact weighting, the weightings across outputs must sum to 100 in steps of five, and the outcome statement is capped at 250 characters. The project title gets 50.

Defra's Biodiversity Challenge Funds keep activities in the matrix as rows, so what the first frame ejected is now mandatory. The indicator rule inverts too: at least two indicators at outcome level and per output, with no ceiling stated. A team that trimmed to one indicator per output to fit the first template is now non-compliant in the opposite direction. No cell may be left empty, means of verification are numbered to mirror the indicators, and the template itself is immutable.

The Green Climate Fund's Simplified Approval Process drops a column the first two treat as fundamental. Its matrix runs baseline, indicator, target, risk, assumptions. There is no means of verification column at all. A team that has spent three weeks agreeing where each number will come from has nowhere in the document to say so. The frame also allows two to three outcomes, hedged in the guidance as "generally (but not always)", which is a sensible thing for a human to write and an awkward thing to enforce.

GFDRR is not a matrix. It is a six-band causal diagram with no indicator columns anywhere, indicators and targets living in a separate results framework, and its published example carries eighteen outcomes against the single outcome the first frame permits.

Four frames. The number of allowed outcomes runs from one to eighteen. Activities are compulsory in one and excluded from another. A column that is structural in two does not exist in a third. And in none of those cases has anything changed about the programme.

Why the templates disagree

The obvious explanation is drift. Templates copied badly, each funder's own procurement people wanting a version with their logo on it, a sector with no standards body and forty years of entropy. That explanation is wrong, or at least it is the smaller half.

The larger half is in the original method. The 1979 Practical Concepts handbook, written by the people who built the logical framework for USAID, states that each of the projects needed to achieve a purpose would be an output, and that this output would become the purpose on the project manager's own logical framework. The level is not a property of the result. It is the result's position relative to a frame. Des Gasper describes the same thing as the general condition of the field: distinctions between levels are treated as contextual rather than inherent.

Which means a funder operating at portfolio level and a funder financing one project in one district are not misusing a shared vocabulary. They are looking at the same causal chain from different heights, and the word that names a row changes with the height. Nine distinct terms appear for the top row across nineteen donor and agency documents, and ten for the second. That count comes from matching the documents against a term list rather than from an impression of reading them, and the same method run with a finance vocabulary over the same nineteen documents returns nothing, which is the control that makes the first number worth quoting.

So the disagreement is structural and it is not going to be argued away. What can change is where a team stores the thing that all four templates are views of.

The rules that are not about columns

Column sets are the visible part and the easy part. The rules that cause more damage are the ones about time and authority, because they are invisible until they bite.

UK Aid Match will not accept a logframe change within three months of an annual review assessment, and requires baseline data on impact, outcome and output indicators within three months of project start. The UK guidance from January 2011 made a logframe mandatory for every newly approved project regardless of value, then set two rule profiles on either side of a one million pound line: a full logframe above it, a shortened one below with a single outcome indicator. Two programmes in the same office under the same funder are governed differently. Defra requires all targets to fall within the project lifetime. Several frames specify who must approve a change at each level, and the approver is not the same person for a statement as for a target.

None of that appears in the matrix. All of it determines whether a correction you make in October is legitimate or a breach.

Then there is the conflict that has no technical resolution at all. WWF's project management standards put a named individual in the responsibility cell. The UK guidance forbids naming individuals, because logframes are published documents. Those two rules are not a formatting difference and no amount of clever field mapping reconciles them, because they encode two different answers to who the document is for: a delivery team that needs to know whose job it is, or a public record that should not expose staff.

What the copies cost

The practical failure is not any one rule. It is that resolving them by hand produces copies, and copies drift.

A team with four funders ends up maintaining four documents that assert the same facts in different shapes on different reporting calendars. Gasper's term for it is double book-keeping, and the mechanism is dull. A figure gets corrected in the version being submitted this month. The other three keep the old number, because nobody remembers they contain it. By month eighteen there are four logframes, no clear answer to which is authoritative, and a quiet inconsistency that will surface in an evaluation two years later as a question about data quality.

The retyping itself is the smaller cost, though it is real and it lands on the people with the least time. The larger cost is that the results logic no longer exists anywhere as a single object. It exists as four partial renderings, and the programme's actual theory of what it is doing is whatever you get by reading all four and reconciling them in your head.

Store the graph, render the frame

The position this leads to is straightforward, and it is a claim about storage rather than about templates.

The results logic is a graph: statements, the causal links between them, the assumptions attached to those links, and the indicators that measure them. That graph is the same for every funder, because it describes the programme. Everything a template varies sits in a separate layer: what each level is called, which columns render and in what order, whether activities appear, how many outcomes are permitted, whether the target lives inside the indicator string or beside it, who may approve a change to what, and when changes are frozen. Held as versioned data rather than as a file, that layer is a rendering instruction, and a donor template becomes a view rather than a second copy.

The consequence worth arguing for is the one about accumulation. Funders are few and their rules are stable for years at a time. A rulebook read once and encoded properly serves every organisation that reports to that funder afterwards. That is the opposite economics from the current arrangement, where each organisation independently decodes the same PDF and stores the result as tacit knowledge in one person's head.

Before accepting a second donor's template as a second document, it is worth answering these:

  1. What are the level names in each frame, and do any two of them use the same word for different rows?
  2. Which frame's rules are strictest on indicator count, and does the strictest one conflict with a floor set elsewhere?
  3. Does any frame require a field that the others do not collect at all?
  4. Where is the authoritative version of each figure, and how would you find out if two renderings disagreed?
  5. Who may change what after approval, under each frame, and is there a blackout period?
  6. Which of these differences would survive if you stored the logic once and generated the documents?

Question six is the one that decides whether this is a tooling problem or a filing problem.

What rendering does not fix

Two limits are worth stating plainly, because the argument above is easy to overstate.

The first is mechanical. Funders do not only specify a shape, they specify a delivery channel, and a portal that requires a particular workbook uploaded in a particular order constrains what can be generated. Generating the content correctly and still having to place it by hand is a partial win, not a clean one.

The second is the WWF and UK conflict from earlier, and it generalises. Where two frames encode incompatible beliefs about what the document is for, rendering does not resolve the incompatibility. It relocates the decision to a single visible place and forces someone to make it once, deliberately, instead of making it four times by accident in four spreadsheets. That is an improvement in honesty more than in effort.

The sector's own answer, moving slowly

None of this is news to the funders. The Grand Bargain named donor reporting as a reform priority precisely because funders use widely different terminology to request what is substantially the same information, and the 8+3 harmonised narrative template was the response. The 2022 independent review reports that by the end of that year more than half of grant-giving signatories were using it in some form with civil society partners. It gives no more precise figure, and it notes that the benefits arrive only once adoption reaches scale.

That is genuine movement on the narrative report. It does not touch the results framework, which is where the structural divergence actually lives, and there is no equivalent harmonisation effort for the matrix itself.

Which leaves the open question, and it is the same one the earlier post ended on. Harmonisation is the right direction and it is a decade's work. A team maintaining four renderings of the same logic has to do something in the meantime, and the honest answer today is that they retype. Whether the alternative holds up under a real donor portal, with a real annual review and a real blackout period, is not something a design argument can settle.

Sources

The frameworks compared

  • Guidance on using the revised Logical Framework, DFID (January 2011). Source for the mandate on all newly approved projects, the full and shortened profiles either side of £1 million, the exclusion of activities from the matrix, the three-indicator cap, assumptions at outcome and output level only, the impact weightings, the 250 and 50 character limits, and the statement that there is no such thing as a SMART indicator
  • Supplementary Logframe Guidance, Biodiversity Challenge Funds, Defra approved. Source for the column headed "SMART Indicators (including disaggregated targets)", the floor of two indicators, activities as matrix rows, and the requirement that no cell be left empty
  • Simplified Approval Process Funding Proposal Template, Green Climate Fund (2022). The project-level logframe now sits inside the funding proposal template as annex 2a. Source for the column set that carries no means of verification, and for the two to three outcomes stated as "generally (but not always)"
  • GFDRR's New Logical Framework, Global Facility for Disaster Reduction and Recovery (2017). Source for the banded causal diagram, the absence of indicator columns and the eighteen outcomes
  • A guide to logframes, UK Aid Match (2020). Source for baseline data within three months of start and the three-month freeze before an annual review assessment
  • WWF Standards of Conservation Project and Programme Management, WWF (2012), hosted by the Conservation Measures Partnership. Source for the named individual in the responsibility cell

Origins and critique of the method

Reporting burden and harmonisation

Our own reading

  • The count of nine distinct terms for the top row and ten for the second comes from Hoverview's reading of nineteen donor and agency logframe documents, matched against a term list rather than read for impression, with a finance vocabulary run over the same nineteen as a null control. The corpus list is available on request